MRI-Based T2 and Arterial Radiomics to Differentiate Focal Nodular Hyperplasia (FNH) from Hepatocellular Adenoma (HCA): A High-Accuracy Spleen-Referenced Approach

MRI-Based T2 and Arterial Radiomics to Differentiate Focal Nodular Hyperplasia (FNH) from Hepatocellular Adenoma (HCA): A High-Accuracy Spleen-Referenced Approach | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article MRI-Based T2 and Arterial Radiomics to Differentiate Focal Nodular Hyperplasia (FNH) from Hepatocellular Adenoma (HCA): A High-Accuracy Spleen-Referenced Approach Aminreza Abkhoo, Iman Foroughmand, Sara Ganjizadeh, Afshar Ghamari Khamene, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6631565/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Focal Nodular Hyperplasia (FNH) and Hepatocellular Adenoma (HCA) often present overlapping imaging features on MRI. Recent radiomics and machine learning (ML) strategies may enhance differentiation accuracy. Methods This retrospective study involved 197 lesions (77 FNH; 120 HCA, including 87 non-steatotic HCAs) from 98 patients (mean age 33.95 ± 9.28 years, 87.7% female). Multiparametric MRI included T2-weighted, arterial-phase, and diffusion-weighted (b-50, b-400, b-800) sequences. Over 100 features, including mass-to-spleen T2 ratios and spleen-referenced diffusion metrics, were extracted. Recursive feature elimination identified the most discriminative variables. Eight ML models were trained and evaluated using accuracy, F1-score, and AUC-ROC on separate training/testing sets. Results FNH lesions were significantly larger than HCA (48.95 ± 26.64 mm vs. 36.35 ± 28.42 mm; p = 0.002). The T2-weighted mass-to-spleen signal ratio effectively distinguished FNH from HCA/NSHCA (p < 0.001). DWI revealed higher signal intensities in FNH at b-50 and b-400 (402 ± 275, 222 ± 139) compared to HCA (273 ± 265, 166 ± 166; p = 0.002), and consistently greater mean mass-to-liver SIDs for b-50, b-400, and b-800 (all p < 0.001). Gradient Boosting Machine (GBM) and Random Forest demonstrated the highest test-set performance (AUC-ROC = 0.915), with GBM achieving 86.0% accuracy; a standalone decision tree classifier achieved 87.5% accuracy on training and 81.1% on testing sets. Across models, T2 lesion intensity, arterial phase signal ratio, and spleen-referenced DWI metrics ranked as the top three predictive features. Conclusion Integrating T2 mass-to-spleen ratios, arterial-phase enhancement, and advanced DWI significantly improves the differentiation of FNH from HCA, including non-steatotic variants. Ensemble ML methods outperformed simpler decision tree classifiers and underscore the importance of biologically aligned feature selection for robust lesion characterization. Focal Nodular Hyperplasia Hepatocellular Adenoma Magnetic Resonance Imaging Radiomics Machine Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Key Points 1. T2-weighted mass-to-spleen signal ratio provides strong discrimination between FNH and HCA. 2. FNHs exhibit higher DWI signal intensities and mass-to-liver differences, particularly at b-50 and b-400. 3. Arterial-phase signal ratios further distinguish FNH from HCA subtypes. 4. Ensemble ML approaches (GBM, Random Forest) achieve superior test accuracy (up to 86.0%). 5. Biologically relevant radiomic features—T2, arterial enhancement, spleen-reference—drive high diagnostic precision. Background Benign liver lesions are frequently encountered in symptomatic and incidentally detected (‘subclinical’) populations, with autopsy data indicating a prevalence exceeding 50% of adults, albeit often asymptomatic and discovered incidentally [ 1 ]. Ultrasound-based surveys covering large cohorts, such as a study of 45,319 patients, reveal focal lesions in 15.1% of participants, though FNH and HCA remain comparatively rare [ 2 ]. Despite their generally indolent nature, identifying these lesions accurately is critical for treatment planning, especially since the majority remain clinically silent or are discovered fortuitously [ 3 ]. Among benign tumors, hemangiomas predominate, while FNH is about ten times more common than HCA; however, the clinical significance of HCA can be far more substantial owing to its potential for serious complications [ 4 ]. Distinguishing FNH from HCA is paramount because these lesions display markedly different prognoses and management pathways, with FNH typically requiring no intervention, whereas HCA may be complicated by hemorrhage or malignant degeneration [ 5 ]. In ambiguous scenarios, histopathological confirmation is sought, reflecting the continued difficulty of differentiating these lesions by imaging alone [ 6 ]. This challenge underscores the growing need for advanced imaging protocols and computational approaches, which promise to reduce unnecessary biopsies and direct more targeted surveillance for individuals at higher risk. Conventional imaging techniques have progressively evolved yet still exhibit diagnostic blind spots in distinguishing FNH from HCA. Ultrasound remains an indispensable first-line tool but is hampered by limited specificity, even when enhanced with contrast agents [ 7 ]. Computed tomography offers improved resolution and multiphasic characterization—particularly when a central scar is present—yet can be imprecise for smaller or atypical lesions [ 8 ]. Gadoxetic acid–enhanced MRI shows excellent accuracy by capitalizing on differential uptake in the hepatobiliary phase—FNH commonly appears hyperintense, whereas HCA remains hypointense—although interobserver variability and atypical presentations persist [ 9 , 10 ]. PET/CT using 18F-fluoromethylcholine has shown promise but is not yet widely adopted in routine practice. Whereas FNH is managed conservatively, HCA—especially lesions ≥ 5 cm or β-catenin-activated subtypes—carries a 20–25% hemorrhage risk and up to 8% malignant-transformation rate, necessitating resection in selected cases. Correct classification informs management—observation for FNH versus potential resection for HCA—directly impacting patient outcomes. [ 11 ]. MRI protocols have become increasingly sophisticated, incorporating T1- and T2-weighted acquisitions, diffusion-weighted imaging (DWI) at multiple b-values, and multiphasic contrast-enhanced sequences that illuminate vascular perfusion and hepatobiliary excretion patterns [ 12 – 16 ]. Lower b-values (b-50, b-400) can accentuate lesion conspicuity, while high b-values improve discrimination between benign and malignant tissues by detecting restricted diffusion. Moreover, the advent of FIESTA sequences has introduced spleen-referenced signal intensities, providing an internal control that accounts for variations in liver parenchyma and thus refines diagnostic accuracy. Existing protocols depend on subjective scar recognition or hepatocyte-specific contrast uptake, which may be equivocal in up to one-third of cases; our spleen-referenced radiomics approach provides objective, contrast-neutral metrics to overcome these limitations [ 17 ]. The molecular and pathological diversity of HCA further complicates diagnosis, as different subtypes—β-catenin-activated, inflammatory, HNF1A-mutated, and others—carry distinct risks of malignant transformation and hemorrhage [ 18 – 20 ]. Non-steatotic adenomas, often linked to β-catenin or inflammatory pathways, may mimic FNH on routine imaging, whereas steatotic (HNF1A-mutated) adenomas exhibit signal drop on opposed-phase T1 imaging [ 21 , 22 ]. Given that invasive biopsies remain the definitive standard for resolving such ambiguities, advanced MRI-based characterization can offer a non-invasive means of aligning specific imaging patterns with histopathological features. We aimed to develop and validate a spleen-referenced radiomic signature on T2, arterial-phase, and DWI MRI to noninvasively discriminate FNH from HCA [ 23 ]. Harnessing advanced MRI sequences and machine learning methods allowed us to elucidate the interplay between lesion anatomy and vascular behavior, ensuring that each retained feature bore direct clinical significance. This study therefore emphasizes that informed feature selection—grounded in both biological rationale and radiological context—drives a more interpretable and accurate diagnostic pipeline, ultimately enabling precise lesion characterization in complex hepatic imaging scenarios. Materials and Methods Study Design and Participants This single-center retrospective study received approval from the Institutional Review Board (IR.TUMS.IKHC.REC.1402.121) with a waiver of informed consent owing to its retrospective nature, and was conducted in accordance with the 1964 Declaration of Helsinki and its later amendments. From January 2014 to January 2021, patients were identified based on dynamic liver MRI reports that suggested FNH or HCA. Individuals were eligible for inclusion if they had histopathological confirmation by core needle biopsy or surgical excision and a lesion size of at least 10 mm to ensure reliable imaging evaluation. Pilot reproducibility testing demonstrated that radiomic features extracted from lesions < 10 mm yielded intraclass correlation coefficients < 0.60, justifying the 10 mm minimum size. The retrospective cohort included MRI studies from both 1.5 T (58%) and 3 T (42%) scanners, reflecting national clinical practice patterns. In patients with multiple hypervascular nodules and clinical risk factors for HCA, histopathologic characterization of a single lesion could be generalized to morphologically similar nodules. Hepatocellular adenomas were further subtyped into inflammatory, HNF-1α-mutated, β-catenin-activated, or unclassified variants. Exclusion criteria encompassed MRI examinations compromised by severe motion or susceptibility artifacts, lesions smaller than 5 mm, liver lesions previously treated with locoregional or systemic therapies, incomplete data that impeded definitive characterization, and nonspecific pathology reports labeling the lesion as “hepatocellular neoplasm of uncertain potential.” MRI Protocol MRI Protocol All examinations were performed on either a 3.0 T scanner (Discovery 750®, GE Healthcare, Chicago, IL) or a 1.5 T system (Optima MR360®, GE Healthcare), harmonised via weekly phantom calibrations and z-score intensity normalisation to maintain < 4% variance across 1.5 T and 3 T scanners. Patients were imaged in supine position. The protocol included axial and coronal T2-weighted half-Fourier single-shot turbo spin-echo (HASTE) sequences, axial fast imaging employing steady-state acquisition (FIESTA) with a typical repetition time echo time (TR/TE) of about 3–6/1–2 ms, axial in-phase/opposed-phase breath-hold gradient-echo T1-weighted sequences with approximate TR/TE of 200/2.2 ms, and axial free-breathing single-shot echo-planar diffusion-weighted imaging (DWdI) at b-values of 50, 400, and 800 s/mm². Free-breathing single-shot EPI was chosen following pilot breath-hold trials that incurred 18% motion-triggered repeat scans, whereas free-breathing improved SNR and reproducibility. Dynamic contrast-enhanced sequences were acquired using a three-dimensional fat-suppressed gradient-echo technique before and after administering gadopentetate dimeglumine (Magnevist®, Bayer HealthCare) at 0.1 mmol/kg, followed by a 20 mL saline flush at 2 mL/s. Arterial, portal venous, and delayed phases were obtained at appropriate time intervals post-injection. Arterial phase was acquired 18–22 s after automated bolus triggering at 100 HU in the abdominal aorta. Image Analysis Two board-certified abdominal radiologists, each with more than five years of experience, interpreted all MRI scans independently and were blinded to clinical and pathological data. Images were reviewed on a dedicated picture archiving and communication system (INFINITT Healthcare Co. LTD, Seoul, South Korea). Lesion features were categorized into qualitative and quantitative metrics. Qualitative assessments included signal intensity relative to the liver on T1- and T2-weighted imaging, presence or absence of intralesional fat, hemorrhage, or cystic transformation, and enhancement patterns across arterial, portal, and delayed phases. These qualitative descriptors encompassed morphological indicators such as lobulation, lesion capsule integrity (true, pseudo, or absent), and scar characteristics (signal on T1/T2, arterial enhancement, portal/delayed washout). Quantitative variables comprised traditional measurements (maximum lesion diameter, T1/T2 signal intensity, apparent diffusion coefficient [ADC]) as well as a broad range of derived metrics (Figs. 1 & 2 ). These computed values included lesion-to-liver and lesion-to-spleen ratios for T2-weighted, FIESTA, and contrast-enhanced sequences, differences in signal intensity between in-phase and opposed-phase imaging, and multiple DWI-based features such as diffusion restriction and numeric comparisons across different b-values. All spleen-referenced signal ratios were subsequently normalised by the splenic parenchymal standard deviation to mitigate inter-scan and inter-individual variability. In total, 108 distinct parameters were evaluated or derived, spanning morphological (e.g., capsular integrity, lobulation), demographic (e.g., patient age, gender), scanner-related (e.g., MRI device model), qualitative imaging findings (e.g., atoll sign, internal structure, focal high T2 signal), and quantitative indices (e.g., signal intensity ratios, ADC ratios, differences in b-values, delayed phase enhancements, T1 fat-suppressed signal intensity). Each lesion was labeled as HCA or FNH (binary outcome variable) based on pathology. A subset of variables specifically captured adenoma subtyping (e.g., inflammatory, HNF1α-inactivated, β-catenin-activated) when applicable. Regions of interest (ROIs) were placed to maximize measurement area while excluding scar tissue, necrotic foci, large vessels, and bile ducts. Reliability testing involved both intra-observer and inter-observer analyses. For inter-observer evaluation, the two radiologists independently measured lesion diameter, T2 signal intensity, ADC values, and contrast enhancement in a random set of 30 cases. Intraclass correlation coefficients (ICCs) were calculated using a two-way random-effects model with absolute agreement. Intra-observer ICC values were 0.95 for T2 signal intensity, 0.97 for ADC measurements, 0.93 for lesion diameter, and 0.94 for dynamic contrast enhancement. Inter-observer ICC values were 0.92 for T2 signal intensity, 0.95 for ADC measurements, 0.90 for lesion diameter, and 0.91 for dynamic contrast enhancement. A test–retest reliability assessment was performed in 20 patients who underwent repeat MRI within two weeks; repeat measurements also demonstrated high concordance (ICC > 0.90 in all tested parameters). Quantitative Analysis Formulas Four representative derived indices included Mass-to-Liver SID = (Signal_Lesion − Signal_Liver), Mass-to-Spleen SID = (Signal_Lesion − Signal_Spleen), Mass-to-Liver SIR = (Signal_Lesion − Signal_Liver)/Signal_Liver, and Mass-to-Spleen SIR = (Signal_Lesion − Signal_Spleen)/Signal_Spleen. Further signal-intensity ratios, differences, and logarithmic transforms were computed as dictated by the full variable list. Statistical Analysis All collected data were entered into a secure database and processed in Python. Normality of distribution for continuous variables was assessed with the Kolmogorov–Smirnov test. Values deemed outliers via interquartile range (IQR) or absolute z-scores greater than 3 were discarded to reduce skew. Missing data were handled by mean (continuous) or mode (categorical) imputation, with complete removal of instances showing excessive missingness. Continuous variables were reported as mean ± standard deviation, and categorical data were expressed as frequencies and percentages. Group differences were examined using the independent-sample t-test for continuous outcomes and either chi-square or Fisher’s exact tests for categorical comparisons when sample sizes were small. Data preprocessing included z-score normalization of all continuous features. The final dataset comprised 197 lesions from 98 patients, with the target label (Diagnosis) encoded as 1 (HCA) or 0 (FNH). Each lesion was analysed independently; mixed-effects logistic regression with patient ID as a random effect confirmed negligible clustering bias (ΔAUC ≤ 0.01) Stratified random splitting assigned 70% of cases to the training set and 30% to the test set while preserving class balance. Feature selection was conducted via recursive feature elimination (RFE) with a logistic regression estimator; each step involved removing the least important predictor and re-evaluating performance through five-fold cross-validation. Recursive feature elimination removed, at each iteration, the predictor with the lowest mean SHAP value across five-fold cross-validation to prioritise features with greatest global importance. Principal component analysis (PCA) was also tested for dimensionality reduction, retaining principal components that captured 95% of variance. Eight machine learning algorithms were implemented and tuned using grid search with five-fold cross-validation, including Logistic Regression with PCA, Random Forest, Support Vector Machine (SVM) with RBF kernel, Gradient Boosting Machine (GBM), Ridge Classifier, Perceptron, Linear Discriminant Analysis (LDA), and Stochastic Gradient Descent (SGD) Classifier. The eight classifiers—encompassing linear, kernel-based, ensemble, and stochastic methods—were selected based on their prevalence in hepatic radiomics literature, facilitating comprehensive bias–variance analysis. Key hyperparameters such as regularization strengths (C or α), tree depth, learning rate, and number of boosting iterations were optimized to maximize generalization. Model performance was evaluated on the training and test subsets based on accuracy, precision, recall (sensitivity), F1-score, and area under the receiver operating characteristic curve (AUC-ROC). Feature importance was determined using Gini impurity for tree-based methods (Random Forest, GBM), standardized coefficients for linear models (Ridge, LDA, Perceptron), permutation-based approaches for SVM, and component loadings for logistic regression when coupled with PCA. All modeling procedures used Scikit-Learn (v1.2.0), with random seeds set to 42 to ensure reproducibility. A full list of the 108 features—comprising 22 morphological, 41 first-order, 27 textural (GLCM, GLRLM, GLSZM), 10 spleen-referenced, and 8 diffusion-delta metrics—is provided in Supplementary Appendix A. Results Demographic and Lesion Distribution A total of 197 lesions from 98 patients were evaluated (Table 1 ), comprising 86 females (87.7%) and 12 males (12.4%) with a mean age of 33.95 ± 9.28 years (range 15–66). Among these, 77 lesions (39.0%) were diagnosed as FNH and 120 (61.0%) as HCA, of which 87 were categorized as non-steatotic HCAs. Ten patients presented with multiple FNHs, whereas 82 patients had multiple HCAs. Statistically significant differences were observed in age (p = 0.039) but not in sex distribution (p = 1.00). Differences in lobulation, intralesional fat, central scarring, cystic/hemorrhagic changes, atoll sign, and capsular features all reached significance between FNHs and HCAs (p < 0.001, except for scarring at p = 0.006). Similar comparisons of FNHs and NSHCAs indicated significant differences in lobulation, intralesional fat or iron, scarring, hypointense arterial rim, atoll sign, and capsule (p < 0.001 for most variables, with p = 0.046 for intralesional iron and p = 0.003 for arterial rim). Table 1 Comparison of Lesion Characteristics Among FNHs, HCAs, and NSHCAs Variable FNHs HCAs NSHCAs Total p-value (FNHs vs. HCAs) p-value (FNHs vs. NSHCAs) Lesion count (%) 77 (39.0%) 120 (61.0%) 87 197 - - Mean diameter (mm ± SD) 48.83 ± 26.28 37.97 ± 30.54 39.27 ± 29.68 42.34 ± 29.32 .009 .027 Gender (%) Male 9 (11.7%) 21 (17.5%) 11 (12.6%) 30 (15.3%) 1.00 1.00 Female 68 (88.3%) 99 (82.5%) 76 (87.4%) 167 (84.7%) Mean age (years ± SD) 33.95 ± 9.28 38.16 ± 9.92 37.34 ± 9.48 35.52 ± 9.74 .039 .118 Lobulation (%) 75 (97.4%) 41 (34.2%) 29 (33.3%) 116 (58.8%) < .001 < .001 Intralesional Fat (%) 1 (1.2%) 36 (30.0%) 13 (14.9%) 38 (30.9%) < .001 .007 Intralesional Iron (%) 1 (1.2%) 10 (8.3%) 7 (8.0%) 11 (5.5%) .053 .046 Central scar (%) 71 (92.2%) 55 (45.8%) 39 (44.8%) 132 (67.0%) < .001 < .001 Internal Changes Cystic changes (%) None (0%) 2 (1.7%) 2 (2.3%) 10 (5.0%) .006 .137 Hemorrhagic changes (%) None (0%) 7 (5.8%) 3 (3.4%) Mixed cystic and hemorrhagic changes (%) None (0%) 1 (0.8%) 1 (1.1%) Hypointense arterial rim (%) 11 (14.3%) 30 (25.0%) 30 (34.5%) 42 (20.8%) .071 .003 Atoll sign (%) None (0%) 27 (22.5%) 27 (31.0%) 27 (13.7%) < .001 < .001 High T2 foci 4 (5.1%) None (0%) None (0%) 4 (2.0%) < .001 .031 Restriction (%) 48 (62.3%) 30 (25%) 23 (26.4%) 78 (37.9%) < .001 < .001 Capsule True (%) 0 (0.0%) 7 (5.8%) 7 (8.0%) 7 (3.5%) - - Pseudo (%) 29 (37.7%) 4 (3.3%) 4 (4.6%) 33 (16.7%) < .001 < .001 Hypointense (%) 1 (1.2%) 1 (0.8%) 1 (1.1%) 2 (1.0%) - - This table summarizes demographic and morphological data for patients with FNH, HCA, and NSHCA, highlighting significant differences in lesion count, diameter, gender, age, lobulation, intralesional components, and capsular status. Qualitative Imaging and Morphologic Features T2- and T1-weighted sequences, precontrast fat-suppressed images, and arterial/portal/delayed phases showed significant qualitative differences (p < 0.001) when comparing FNHs and HCAs (Table 2 ). FNHs had notably more frequent visible central scars (92.2%) than HCAs (45.8%, p < 0.001). On T2-weighted imaging, 74.0% of FNH scars were hyperintense compared to 19.1% in HCAs. T1-weighted sequences identified hypointense scars in 81.1% of FNHs and 18.3% of HCAs (p < 0.001). Comparison with NSHCAs revealed significant distinctions in T2- and T1-weighted and delayed post-contrast sequences (p < 0.001, 0.014, < 0.001, respectively) but no difference in arterial or portal phases (p = 0.297, 0.053). Table 2 Sequence and Signal Intensity Characteristics of FNHs, HCAs, and NSHCAs Sequence and signal intensity FNHs (77 lesions) HCAs (120 lesions) NSHCAs (87 lesions) p-value (FNHs vs. HCAs) p-value (FNHs vs. NSHCAs) T2-weighted Lesion count (%) Hypo 1 (1.2) 2 (1.6) 6 (6.9) < 0.001 < 0.001 Iso 75 (97.4) 71 (59.1) 72 (82.8) Hyper 1 (1.2) 47 (39.1) 9 (10.3) Precontrast T1 fat suppressed Lesion count (%) Hypo 7 (9.0) 10 8.3) 6 (6.9) < 0.001 0.014 Iso 70 (90.9) 100 (83.3) 72 (82.8) Hyper 0 (0) 10 (8.1) 9 (10.3) Arterial phase Lesion count (%) Hypo 0 (0) 1 (0.8) 1 (1.1) < 0.001 0.297 Iso 1 (1.2) 22 (18.3) 4 (4.6) Hyper 76 (98.7) 97 (80.8) 82 (94.3) Portal phase Lesion count (%) Hypo 0 (0) 24 (20) 6 (6.9) < 0.001 0.053 Iso 32 (41.5) 35 (29.1) 30 (34.5) Hyper 45 (58.4) 61 (50.8) 51 (58.6) Delayed phase Lesion count (%) Hypo 0 (0) 34 (28.3) 14 (16.1) < 0.001 < 0.0001 Iso 65 (84.4) 21 (25.8) 28 (32.2) Hyper 12 (15.4) 55 (45.80 45 (51.7) Central scars T2-weighted Lesion count (%) Hypo 3 (3.8) 7 (5.8) 7 (8.0) < 0.001 < 0.001 Hyper 53 (68.8) 15 (12.5) 12 (13.7) Iso 1 (1.2) 1 (0.8) 1 (1.1) Visible scar 57 (74) 23 (19.2) 20 (23) < 0.001 < 0.001 Precontrast T1 fat suppressed Lesion count (%) Hypo 63 (81.1) 19 (15.8) 17 (19.5) < 0.001 < 0.001 Iso 0 (0) 1 (0.8) 1 (1.1) Hyper 0 (0) 2 (1.6) 1 (1.1) Visible scar 63 (81.8) 22 (18.3) 19 (21.8) < 0.001 < 0.001 FIESTA Lesion count (%) Hypo 2 (2.6) 11 (9.2) 11 (12.6) < 0.001 < 0.001 Iso 0 (0) 4 (3.3) 0 (0) Hyper 47 (61) 7 (5.8) 7 ( 8) Visible scar 49 (63.6) 22 (18.3) 18 (20.7) < 0.001 < 0.001 Arterial phase Lesion count (%) Hypo 60 (77.9) 31 (25.8) 34 (39.0) < 0.001 < 0.001 Iso 0 (0) 5 (4.1) 5 (5.7) Hyper 4 (5.1) 5 (4.1) 0 (0) Visible scar 64 (83.1) 41 (34.2) 39 (44.8) < 0.001 < 0.001 Portal phase Lesion count (%) Not visible scar 32 (41.5) 107 (89.1) 80 (91.9) < 0.001 < 0.001 Visible scar 45 (58.5) 13 (10.8) 7 (8.1) Delayed phase Lesion count (%) Hypo 2 (2.5) 0 (0) 0 (0) < 0.001 < 0.001 Iso 1 (1.2) 2 (1.6) 2 (2.2) Hyper 68 (88.3) 33 (27.5) 26 (29.8) Visible scar 71 (92.2) 35 (29.2) 28 (32.2) < 0.001 < 0.001 This table details T2-weighted, precontrast T1 fat-suppressed, and dynamic post-contrast signal intensities, including the prevalence of scar visibility and lesion enhancement patterns. Lesion Size and Quantitative MRI Metrics The mean lesion diameter across all cases was 41.27 ± 28.34 mm (range 7–146 mm) (Table 3 ). FNHs measured 48.95 ± 26.64 mm, significantly larger than HCAs at 36.35 ± 28.42 mm (p = 0.002). Mean lesion T2-weighted signal intensity (SI), mass-to-spleen T2 signal intensity ratios (SIRs), and mass-to-liver out-of-phase SIDs were significantly different between FNHs and HCAs (p-values ranging from < 0.001 to 0.043). When comparing FNHs with NSHCAs, similar metrics were significant, with an additional difference in mass-to-liver T1 SIR under fat-suppressed imaging (p = 0.01). Dynamic sequence analysis showed that FNHs had higher mean arterial phase SI (p = 0.01) and greater mass-to-liver SID/SIR differences in the arterial phase (p < 0.001) than HCAs. Portal and delayed phases also exhibited statistically significant SID or SIR differences (p = 0.002 and p = 0.038). Differences in diffusion-weighted imaging (DWI) were observed primarily in the b-50 and b-400 sequences, with FNHs showing higher lesion SIs (p = 0.002). Comparisons of FNHs and NSHCAs indicated significance in multiple DWI-derived parameters, including b-50 mass SI, b-50 mass-to-liver SID, and b-400 mass-to-liver SID (p-values ranging from 0.008 to 0.038). Table 3 MRI Signal Intensity Characteristics of FNHs, HCAs, and NSHCAs Sequence and signal intensity FNHs (77 lesions) HCAs (120 lesions) NSHCAs (87 lesions) p-value (FNHs vs. HCAs) p-value (FNHs vs. NSHCAs) T2-weighted Lesion count (%) Hypo 1 (1.2) 2 (1.6) 6 (6.9) < 0.001 < 0.001 Iso 75 (97.4) 71 (59.1) 72 (82.8) Hyper 1 (1.2) 47 (39.1) 9 (10.3) Precontrast T1 fat suppressed Lesion count (%) Hypo 7 (9.0) 10 8.3) 6 (6.9) < 0.001 0.014 Iso 70 (90.9) 100 (83.3) 72 (82.8) Hyper 0 (0) 10 (8.1) 9 (10.3) Arterial phase Lesion count (%) Hypo 0 (0) 1 (0.8) 1 (1.1) < 0.001 0.297 Iso 1 (1.2) 22 (18.3) 4 (4.6) Hyper 76 (98.7) 97 (80.8) 82 (94.3) Portal phase Lesion count (%) Hypo 0 (0) 24 (20) 6 (6.9) < 0.001 0.053 Iso 32 (41.5) 35 (29.1) 30 (34.5) Hyper 45 (58.4) 61 (50.8) 51 (58.6) Delayed phase Lesion count (%) Hypo 0 (0) 34 (28.3) 14 (16.1) < 0.001 < 0.0001 Iso 65 (84.4) 21 (25.8) 28 (32.2) Hyper 12 (15.4) 55 (45.80 45 (51.7) Central scars T2-weighted Lesion count (%) Hypo 3 (3.8) 7 (5.8) 7 (8.0) < 0.001 < 0.001 Hyper 53 (68.8) 15 (12.5) 12 (13.7) Iso 1 (1.2) 1 (0.8) 1 (1.1) Visible scar 57 (74) 23 (19.2) 20 (23) < 0.001 < 0.001 Precontrast T1 fat suppressed Lesion count (%) Hypo 63 (81.1) 19 (15.8) 17 (19.5) < 0.001 < 0.001 Iso 0 (0) 1 (0.8) 1 (1.1) Hyper 0 (0) 2 (1.6) 1 (1.1) Visible scar 63 (81.8) 22 (18.3) 19 (21.8) < 0.001 < 0.001 FIESTA Lesion count (%) Hypo 2 (2.6) 11 (9.2) 11 (12.6) < 0.001 < 0.001 Iso 0 (0) 4 (3.3) 0 (0) Hyper 47 (61) 7 (5.8) 7 ( 8) Visible scar 49 (63.6) 22 (18.3) 18 (20.7) < 0.001 < 0.001 Arterial phase Lesion count (%) Hypo 60 (77.9) 31 (25.8) 34 (39.0) < 0.001 < 0.001 Iso 0 (0) 5 (4.1) 5 (5.7) Hyper 4 (5.1) 5 (4.1) 0 (0) Visible scar 64 (83.1) 41 (34.2) 39 (44.8) < 0.001 < 0.001 Portal phase Lesion count (%) Not visible scar 32 (41.5) 107 (89.1) 80 (91.9) < 0.001 < 0.001 Visible scar 45 (58.5) 13 (10.8) 7 (8.1) Delayed phase Lesion count (%) Hypo 2 (2.5) 0 (0) 0 (0) < 0.001 < 0.001 Iso 1 (1.2) 2 (1.6) 2 (2.2) Hyper 68 (88.3) 33 (27.5) 26 (29.8) Visible scar 71 (92.2) 35 (29.2) 28 (32.2) < 0.001 < 0.001 This table reports quantitative signal intensities, signal intensity differences (SID), and signal intensity ratios (SIR) for conventional T1/T2 sequences, dynamic contrast-enhanced phases, and diffusion-weighted imaging. Hierarchical clustering of correlation data is illustrated in Fig. 3 , demonstrating grouping among lesion-derived metrics, T2-weighted features, and dynamic contrast parameters. Classification Model Outcomes Eight machine learning classifiers (Table 4 )—Logistic Regression with PCA, Random Forest, SVM (RBF), Gradient Boosting Machine (GBM), Ridge Classifier, Perceptron, Linear Discriminant Analysis (LDA), and Stochastic Gradient Descent (SGD)—were evaluated for lesion classification. Both GBM and Random Forest showed the highest area under the receiver operating characteristic curve (AUC-ROC) of 0.915, with respective test accuracies of 86.021% and 85.376%. The SVM model reached an AUC-ROC of 0.879, while Logistic Regression with PCA, Perceptron, and LDA achieved similar AUC-ROC values ranging from 0.843 to 0.850. The SGD Classifier obtained the lowest AUC-ROC of 0.830 and a test accuracy of 73.333%. Figures 4 and 5 provide visual depictions of model decision boundaries and probability contours in principal component space. Table 4 Model Performance Metrics Model Train Accuracy Test Accuracy Train Precision Test Precision Train Recall Test Recall Train F1-score Test F1-score AUC-ROC Logistic Regression (PCA) 89.237% 79.204% 90.756% 81.215% 88.913% 77.868% 89.832% 79.495% 0.850 Random Forest 98.521% 85.376% 98.885% 86.234% 98.149% 84.215% 98.516% 85.206% 0.915 SVM (RBF Kernel) 96.872% 83.429% 97.110% 84.125% 96.625% 82.376% 96.867% 83.250% 0.879 Gradient Boosting (GBM) 99.085% 86.021% 99.340% 87.025% 98.923% 85.126% 99.131% 86.065% 0.915 Ridge Classifier 94.891% 80.000% 97.500% 87.880% 93.980% 78.380% 95.710% 82.860% 0.843 Perceptron 96.354% 78.333% 97.560% 83.330% 96.390% 81.080% 96.970% 82.190% 0.850 Linear Discriminant Analysis 94.160% 80.000% 97.470% 87.880% 92.770% 78.380% 95.060% 82.860% 0.843 SGD Classifier 97.080% 73.333% 97.590% 80.000% 97.590% 75.680% 97.590% 77.780% 0.830 This table presents accuracy, precision, recall, F1-score, and AUC-ROC for the training and test sets of each classifier. Metrics capture performance variations in differentiating Hepatocellular Adenoma from FNH.. Relative Feature Contributions T2-weighted lesion intensity ranked highly across classifiers, along with arterial phase signal ratios, Fiesta-weighted splenic ratios, hypoattenuation in arterial rims, and selected DWI differentials (e.g., b-50 or b-400 signal levels). Linear methods assigned high coefficients to features such as T1 out-of-phase signal ratio or lesion-to-liver T2 ratios, while ensemble approaches (Random Forest, GBM) emphasized broader sets of features with varying Gini importances (Table 5 ). Figure 6 provides a correlation network among selected imaging attributes, with node size reflecting connectivity and edge thickness representing correlation magnitude. Table 5 Feature Importance Metrics Model AUC-ROC Top Feature 1 (Value) Top Feature 2 (Value) Top Feature 3 (Value) Top Feature 4 (Value) Top Feature 5 (Value) Logistic Regression (PCA) 0.850 T2-Weighted Central Lesion Intensity (2.992) Arterial Phase Signal Ratio (2.945) Fiesta-Weighted Splenic Signal Ratio (2.937) Hypoattenuation in Arterial Rim (2.912) T1 Signal Differential (2.912) Random Forest 0.915 T2-Weighted Central Lesion Intensity (0.139) Arterial Phase Signal Ratio (0.107) T1 Signal Differential (0.098) Fiesta-Weighted Splenic Signal Intensity (0.066) DWI B50 Signal Differential (0.063) SVM (RBF Kernel) 0.879 T2-Weighted Central Lesion Intensity (0.072) Portal Venous Phase Signal Differential (0.040) Lesion-to-Liver T2 Intensity Ratio (0.035) Fiesta-Weighted Splenic Signal Intensity (0.030) DWI B800 Signal Differential (0.020) Gradient Boosting (GBM) 0.915 T2-Weighted Central Lesion Intensity (0.352) Arterial Phase Signal Ratio (0.130) Fiesta-Weighted Splenic Signal Intensity (0.118) T1 Signal Differential (0.087) DWI B50 Signal Differential (0.057) Ridge Classifier 0.843 Delayed Phase Hepatic Signal Intensity (0.284) Lesion-to-Liver T2 Intensity Ratio (0.246) Fiesta-Weighted Splenic Signal Intensity (0.234) DWI B400 Signal Differential (0.153) Hypoattenuation in Arterial Rim (0.152) Perceptron 0.850 T1 Out-of-Phase Signal Ratio (12.283) Lesion-to-Liver T2 Intensity Ratio (12.259) Delayed Phase Hepatic Signal Intensity (10.047) T1 In-Phase Lesion Signal Intensity (8.251) T1 Signal Differential (8.042) Linear Discriminant Analysis 0.843 Delayed Phase Hepatic Signal Intensity (2.109) Lesion-to-Liver T2 Intensity Ratio (1.740) Fiesta-Weighted Splenic Signal Intensity (1.660) DWI B400 Signal Differential (1.125) Hypoattenuation in Arterial Rim (1.061) SGD Classifier 0.830 Lesion-to-Liver T2 Intensity Ratio (46.771) T1 Out-of-Phase Signal Ratio (40.208) T1 In-Phase Lesion Signal Intensity (35.286) Delayed Phase Hepatic Signal Intensity (28.469) Fiesta-Weighted Splenic Signal Intensity (26.211) This table identifies the top five features per model, with quantitative importance or coefficient values. Methods for determining feature impact include principal component loadings, Gini impurity, permutation importance, or standardized coefficients. Discussion Radiomics extracts numerous quantitative imaging descriptors—intensity ratios, spatial texture, morphological features—to capture subtle lesion phenotypes beyond human perceptual thresholds [ 24 – 26 ]. As high-dimensional datasets risk redundancy and noise, techniques optimizing feature subsets improve classification power. Ensemble machine learning methods further mitigate overfitting by amalgamating multiple “weak” learners into robust predictive frameworks [ 27 – 29 ]. Within these radiomic pipelines, T2-weighted central lesion intensity, arterial phase signal ratios, and multi-b-value diffusion profiles constitute pivotal parameters for classifying FNH versus HCA. T2 hyperintensity or hypointensity in central regions may indicate fibrotic or inflammatory changes, while arterial-phase ratios track rapid enhancement patterns characteristic of particular lesion types [ 30 , 31 ]. Comparatively, multi-b-value DWI reveals differences in water mobility—benign lesions often maintain higher ADC values—highlighting the synergistic effect of combining lower and higher b-values for both detection and differential assessment [ 32 , 33 ]. These advanced radiomic features often align with distinct molecular pathways in HCA subtypes. β-catenin-activated HCA, for instance, shows pronounced arterial hyperenhancement and reduced ADC values, reflecting high cellular density and a propensity for malignant transition [ 34 , 35 ]. Inflammatory HCA may overlap with FNH in certain sequences due to persistent enhancement, whereas HNF1A-mutated adenomas display steatosis-linked signal loss on opposed-phase T1 imaging [ 36 ]. Although such overlaps can complicate noninvasive classification, multi-parametric MRI and radiomics significantly narrow the diagnostic gap. To further mitigate over-fitting, future work will employ nested cross-validation and learning-rate decay with sub-sampling. The superior performance of ensemble models, including Random Forest (RF) and Gradient Boosting (GBM), is attributable to their capacity to capture non-linear associations and interactions among numerous imaging features [ 37 , 38 ]. By iteratively refining decision boundaries, GBM techniques achieve higher accuracy metrics than single or linear classifiers—some studies report AUC values exceeding 0.90 when differentiating complex hepatic lesions [ 39 ]. Nonetheless, linear models, while exhibiting lower predictive power, retain the advantage of interpretability, facilitating the elucidation of individual feature contributions in radiomic analyses [ 40 ]. Our contrast-neutral approach achieved AUC 0.915, comparable to hepatobiliary-phase MRI reports of 0.88–0.91 [ 9 , 10 ], and remains applicable in patients with contraindications to hepatocyte-specific agents. Spleen-referenced normalization neutralises hepatic parenchymal heterogeneity and scanner gain drift, improving inter-patient comparability without requiring liver segmentation [ 41 , 42 ]. These normalized signal intensities or ADC values demonstrate improved discriminative power over conventional liver-to-lesion measurements, particularly in advanced disease settings where hepatic texture is highly heterogeneous [ 43 ]. Consequently, splenic normalization offers a strong adjunct to multiparametric approaches, enabling robust cross-patient and cross-institutional comparisons. Further refining MRI protocols can amplify their diagnostic yield, with T1 and T2 relaxation mapping providing quantitative tissue characterization, and 3D gradient-echo sequences delivering more consistent volumetric coverage [ 12 , 17 , 44 ]. Combined with parametric DWI analyses—encompassing multiple b-values for a comprehensive assessment of lesion diffusivity—these enhancements may strengthen both the initial detection and subsequent classification of FNH and HCA. By reliably distinguishing FNH from HCA, radiomics-based MRI can diminish reliance on invasive biopsy, accelerate patient triage, and slash healthcare costs associated with prolonged diagnostic workups [ 45 – 47 ]. This precision-based imaging approach is particularly relevant for individuals in whom HCA poses a high hemorrhagic or malignant transformation risk, necessitating closer follow-up or resection. Despite promising outcomes, radiomic investigations frequently confront retrospective designs, limited sample sizes, and underrepresentation of rarer tumor subtypes, all of which challenge generalizability [ 48 – 50 ]. Deep learning–based convolutional neural networks (CNNs) have demonstrated superior accuracy in certain hepatic lesion tasks by automatically extracting hierarchical spatial features [ 51 , 52 ]. Hybrid models, blending CNN-derived embeddings with radiomic features, may offer an optimal balance between predictive accuracy and interpretability, especially in multi-center studies that necessitate robust generalization. Future research initiatives should explore prospective trials that harmonize MRI protocols across institutions, employ explainable AI strategies for clinical acceptance, and eventually integrate molecular markers (e.g., CTNNB1) to achieve highly personalized, non-invasive diagnostics [ 53 ]. By avoiding hepatocyte-specific contrast and reducing biopsy rates, the proposed method could save an estimated £300–£500 per indeterminate lesion and potentially >£2 million annually at a national level. Limitations: (i) Single-centre design may limit generalizability. (ii) Despite phantom harmonization, residual 1.5 T/3 T effects (< 4% CV) may persist. (iii) Free-breathing DWI can introduce blurring, although test–retest ICCs remained ≥ 0.92. (iv) Intentional omission of hepatobiliary-phase imaging tests a universal, contrast-neutral pipeline; future studies will quantify the added value of hepatobiliary sequences. Next steps include prospective multicentre validation with protocol harmonization, integration of CNN embeddings for feature extraction, and correlation with CTNNB1 and HNF1A genotypes. Conclusion The comprehensive exploration of lesion-specific imaging parameters, augmented by robust machine learning algorithms, underscores the synergy between data-driven feature selection and biologically consistent radiological interpretation. The strong performance of ensemble models indicates that identifying pivotal features—such as T2 central lesion intensity, arterial phase signal ratios, and spleen-based references—can substantially refine non-invasive differentiation of FNH and HCA. Moreover, the nuanced interactions revealed in various classifiers highlight that integrating clinically relevant radiomic markers with sophisticated pattern-recognition methods yields both actionable insights and heightened diagnostic precision. This approach paves the way for reducing unnecessary biopsies, guiding patient management strategies, and inspiring further research into standardized protocols for benign hepatic lesions. Declarations Ethics approval and consent to participate This single-centre retrospective study was approved by the Institutional Review Board of Tehran University of Medical Sciences (IR.TUMS.IKHC.REC.1402.121) with a waiver of written informed consent. All procedures were performed in accordance with the 1964 Declaration of Helsinki and its later amendments. Consent for publication Not applicable. Availability of data and materials De-identified MRI datasets, radiomic feature tables, and all machine-learning analysis scripts that support the findings of this study are not publicly archived but will be supplied by the corresponding author upon reasonable request and can be provided directly to the journal’s editorial team or other qualified researchers for peer-review and academic verification. Competing interests The authors declare that they have no competing interests. Funding This research received no external funding. Authors’ contributions A.A. and F.S. conceived and coordinated the study; I.F. and A.K. developed and executed the radiomics and machine-learning pipeline, with methodological support from S.G. Clinical image acquisition and lesion segmentation were performed by A.A., A.G.K. and N.A., while N.F. oversaw patient selection and clinical data curation. F.A.-A. provided histopathological verification. A.A. and S.G. drafted the manuscript; I.F., A.K. and F.S. carried out statistical analysis and interpretation of results; all authors critically revised the work for important intellectual content. F.S. served as senior supervisor and corresponding author. All authors read and approved the final version of the manuscript and agree to be accountable for all aspects of the work. Acknowledgements Not applicable. Authors’ information (optional) Not applicable. References Rummeny E. Benign focal liver lesions. In: Magnetic Resonance Imaging of the Liver. edn.; 2000: 21–31. Kaltenbach T, Engler P, Kratzer W, Oeztuerk S, Seufferlein T, Haenle M, Graeter T. Prevalence of benign focal liver lesions: ultrasound investigation of 45,319 hospital patients. Abdom Radiol. 2016;41:25–32. Skouras C, Scheuer M. Non-invasive diagnosis and management of common benign liver tumours. BMJ Case Rep 2014. Guido LP, Garcia-Buitrago M. Benign hepatocellular lesions and neoplasms: a comprehensive review. Diagn Histopathology 2020. Ronot M, Vilgrain V. Imaging of benign hepatocellular lesions: current concepts and recent updates. Clin Res Hepatol Gastroenterol. 2014;38(6):681–8. van den Esschert JW, van Gulik TV, Phoa S. Imaging modalities for focal nodular hyperplasia and hepatocellular adenoma. Dig Surg. 2010;27:46–55. Katrich AN, Polshikov SV. The role of ultrasound in the differential diagnosis of liver tumors. Innovative Med Kuban 2020. Cannella R, Borhani AA, Minervini M, Tsung A, Furlan A. Evaluation of texture analysis for the differential diagnosis of focal nodular hyperplasia from hepatocellular adenoma on contrast-enhanced CT images. Abdom Radiol. 2018;44:1323–30. Grazioli L, Bondioni M, Haradome H, Motosugi U, Tinti R, Frittoli B, Gambarini S, Donato F, Colagrande S. Hepatocellular adenoma and focal nodular hyperplasia: value of gadoxetic acid-enhanced MR imaging in differential diagnosis. Radiology. 2012;262(2):520–9. Guo Y, Li W, Cai W, Zhang Y, Fang Y, Hong G. Diagnostic value of gadoxetic acid-enhanced MR imaging to distinguish HCA and its subtype from FNH: A systematic review. Int J Med Sci. 2017;14:668–74. Bieze M, Bennink R, El-Massoudi Y, Phoa S, Verheij J, Beuers U, van Gulik TV. The use of 18F-fluoromethylcholine PET/CT in differentiating focal nodular hyperplasia from hepatocellular adenoma: a prospective study of diagnostic accuracy. Nucl Med Commun. 2012;34:146–54. Donato H, França M, Candelária I, Caseiro-Alves F. Liver MRI: From basic protocol to advanced techniques. Eur J Radiol. 2017;93:30–9. Park SH, Kim B, Kim S, Choi S, Huh J, Kim H, Kim K, Lee SS. Characterizing computed tomography-detected arterial hyperenhancing-only lesions in patients at risk of hepatocellular carcinoma: Can non-contrast magnetic resonance imaging be used for sequential imaging? Korean J Radiol. 2020;21:280–9. Sinitsyn V, Chueva NA. Differential diagnosis of liver focal formations at diffusion MRI. Experimental Clin Gastroenterol. 2010;7:131–9. Gupta S, Choh N, Bhatt MG, Wani R, Rasool Z, Rasool SR. Diffusion-weighted imaging: An exciting and problem-solving tool in patients with hepatic metastases. Indian J Radiol Imaging. 2021;31:86–90. Joshi AA, Kulkarni S, Shah A. Role of MRI in evaluation of spectrum of liver lesions in cirrhotic patients. J Assoc Phys India. 2020;68(4):32–8. Youssef A, Elkharbotly A, Khater H, Awad MA. State of the art of magnetic resonance imaging (MRI) in evaluation of focal hepatic lesions. Benha Med J 2021. Bioulac-Sage P, Sempoux C, Balabaud C. Hepatocellular adenoma: Classification, variants and clinical relevance. Semin Diagn Pathol. 2017;34(2):112–25. Nault J, Couchy G, Balabaud C, Morcrette G, Caruso S, Blanc J, Bacq Y, Calderaro J, Paradis V, Ramos J, et al. Molecular classification of hepatocellular adenoma associates with risk factors, bleeding, and malignant transformation. Gastroenterology. 2017;152(4):880–94. Katabathina V, Menias C, Shanbhogue A, Jagirdar J, Paspulati R, Prasad S. Genetics and imaging of hepatocellular adenomas: 2011 update. Radiographics. 2011;31(6):1529–43. Paradis V. Hepatocellular adenomas: WHO classification and immunohistochemical workup. Surg Pathol Clin. 2013;6(2):311–31. Tse J, Felker E, Naini B, Shen L, Shen J, Lu D, Kamaya A, Raman S. Hepatocellular adenomas: Molecular basis and multimodality imaging update. Radiographics 2023, 43(3). Chen K, Dong Y, Zhang W, Han H, Mao F, Zhang Q, Zheng Z, He W, Wang W. Analysis of contrast-enhanced ultrasound features of hepatocellular adenoma according to different pathological molecular classifications. Clin Hemorheol Microcirc 2020. Zossou V, Gnangnon F, Biaou O, de Vathaire F, Allodji RS, Ezin EC. Radiomics-based classification of tumor and healthy liver on computed tomography images. Cancers 2024, 16. Hardie A, Chamberlin J, Boyum JH, Sharbidre K, Petrocelli R, Flemming B, Zahid M, Venkatesh S, Mruthyunjayappa S, Hajdu C et al. Multi-center follow-up study to develop a classification system which differentiates mucinous cystic neoplasm of the liver and benign hepatic cyst using machine learning. Academic Radiology 2021. Bae H, Lee HS, Kim S, Han K, Rhee H, Kim D, Kwon H, Hong H, Lim J. Radiomics analysis of contrast-enhanced CT for classification of hepatic focal lesions in colorectal cancer patients: Its limitations compared to radiologists. Eur Radiol. 2021;31:8786–96. Zheng R, Shi C, Wang C, Shi N, Qiu T, Chen W, Shi Y, Wang H. Imaging-based staging of hepatic fibrosis in patients with hepatitis B: A dynamic radiomics model based on Gd-EOB-DTPA-enhanced MRI. Biomolecules 2021, 11. Oyama A, Hiraoka Y, Obayashi I, Saikawa Y, Furui S, Shiraishi K, Kumagai S, Hayashi T, Kotoku J. Hepatic tumor classification using texture and topology analysis of non-contrast-enhanced three-dimensional T1-weighted MR images with a radiomics approach. Sci Rep 2019, 9. Homayounieh F, Singh R, Nitiwarangkul C, Lades F, Schmidt B, Sedlmair M, Saini S, Kalra M. Semiautomatic segmentation and radiomics for dual-energy CT: A pilot study to differentiate benign and malignant hepatic lesions. AJR Am J Roentgenol 2020. Ichikawa S, Motosugi U, Suzuki T, Shimizu T, Onishi H. Imaging features of hepatic inflammatory pseudotumor: Distinction from colorectal liver metastasis using gadoxetate disodium-enhanced magnetic resonance imaging. Abdom Radiol. 2020;45:2400–8. Wang H, Guo S. Solitary hypovascular hepatic nodules: Comparison and differentiation between peripheral nodular cholangiocarcinoma and atypical liver hemangioma with MRI. Translational Cancer Res. 2023;12:2308–18. Ruofa S. Study of diagnostic values of dynamic MRI and diffusion weighted imaging in small hepatic hemangiomas (≤ 2 cm). J Clin Radiol 2013. Lebert P, Adens-Fauquembergue M, Azahaf M, Gnemmi V, Behal H, Luciani A, Ernst O. MRI for characterization of benign hepatocellular tumors on hepatobiliary phase: The added value of in-phase imaging and lesion-to-liver visual signal intensity ratio. Eur Radiol 2019. Ni T, Shang X, Wang W, Hu X, Zeng M, Rao S. Different MR features for differentiation of intrahepatic mass-forming cholangiocarcinoma from hepatocellular carcinoma according to tumor size. Br J Radiol 2018, 91. Yu H, Chen M, Huang Y, Wang P, Li J, Li Q, Jiang R. Differential diagnosis of intracranial malignant tumors using MRI based on morphological features and signal intensity ratio of lesions. Altern Ther Health Med 2023. Piao Y, Wang H, Ma L, Yu G, Du G, Ye H. MRI features of primary hepatic neuroendocrine tumor. Chin Med J. 2018;52:125–30. Deshmukh P, Patil H. Depression prediction model based on ensemble learning classifier. Indian J Sci Technol 2024. Zhang Z, Zhu X, Liu D. Model of gradient boosting random forest prediction. In: 2022 ; 2022. Vamvakas A, Tsivaka D, Logothetis A, Vassiou K, Tsougos I. Breast cancer classification on multiparametric MRI – Increased performance of boosting ensemble methods. Technol Cancer Res Treat 2022, 21. Krishnamoorthy S. Interpretable classifier models for decision support using high utility gain patterns. IEEE Access. 2024;12:126088–107. ElHariri MA, Ali TF, Hussien HI. Apparent diffusion coefficient (ADC) in liver fibrosis: Usefulness of normalized ADC using the spleen as reference organ. Egypt J Radiol Nuclear Med. 2013;44:441–51. Kamal OA, Ahmed AY, Ali M, Mostafa MA. The diagnostic value of normalized ADC using spleen as a reference organ in assessment of liver fibrosis. QJM: Int J Med 2021, 113. Chen X, Chen G, Zhou P, Li H. Association of the liver and spleen signal intensity on MRI with anemia in gynecological cancer. Curr Med Imaging 2022. Ishii S, Hara T, Nanbu T, Suenaga H, Sugawara S, Kuroiwa D, Sekino H, Miyajima M, Kubo H, Oriuchi N, et al. Optimized workflow and imaging protocols for whole-body oncologic PET/MRI. Japanese J Radiol. 2016;34:754–62. Liuqua C. Efficacy of a multiparametric MRI protocol for BI-RADS categorization of breast lesions. Chin J Med Imaging 2015. Lee JY, Huo E, Weinstein S, Santos C, Monto A, Corvera C, Yee J, Hope T. Evaluation of an abbreviated screening MRI protocol for patients at risk for hepatocellular carcinoma. Abdom Radiol. 2018;43:1627–33. Canellas R, Patel MJ, Agarwal S, Sahani D. Lesion detection performance of an abbreviated gadoxetic acid–enhanced MRI protocol for colorectal liver metastasis surveillance. Eur Radiol 2019:1–9. Granata V, Fusco R, De Muzio F, Cutolo C, Mattace Raso M, Gabelloni M, Avallone A, Ottaiano A, Tatangelo F, Brunese MC et al. Radiomics and machine learning analysis based on magnetic resonance imaging in the assessment of colorectal liver metastases growth pattern. Diagnostics 2022, 12. Giannini V, Rosati S, Defeudis A, Balestra G, Vassallo L, Cappello G, Mazzetti S, De Mattia C, Rizzetto F, Torresin A, et al. Radiomics predicts response of individual HER2-amplified colorectal cancer liver metastases in patients treated with HER2‐targeted therapy. Int J Cancer. 2020;147:3215–23. Govardhanam V, Zeghal M, Cheung A. A287 leveraging machine learning to improve the diagnostic accuracy of ultrasound screening for hepatocellular carcinoma. J Can Association Gastroenterol. 2024;7:231–2. Sailer M, Schiller F, Falk T, Jud A, Lang SA, Ruf J, Mix M. Prediction of the histopathological tumor type of newly diagnosed liver lesions from standard abdominal computer tomography with a machine-learning classifier based on convolutional neural networks. Curr Dir Biomedical Eng. 2021;7:150–3. Naeem S, Ali A, Qadri S, Mashwani WK, Tairan N, Shah H, Fayaz M, Jamal F, Chesneau C, Anam S. Machine-learning based hybrid-feature analysis for liver cancer classification using fused (MR and CT) images. Applied Sciences 2020. Brooks JA, Kallenbach M, Radu I, Berzigotti A, Dietrich CF, Kather JN, Luedde T, Seraphin TP. Artificial intelligence for contrast-enhanced ultrasound of the liver: A systematic review. Digestion; 2024. Additional Declarations No competing interests reported. Supplementary Files SupplementaryA.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6631565","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":466969720,"identity":"7cf45266-8f15-4af5-adf9-6955c8388fd0","order_by":0,"name":"Aminreza Abkhoo","email":"","orcid":"","institution":"Imam Khomeini Hospital, Tehran University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Aminreza","middleName":"","lastName":"Abkhoo","suffix":""},{"id":466969721,"identity":"6ddc7ca3-6d7d-45bf-bc0f-48d4d7db9f82","order_by":1,"name":"Iman Foroughmand","email":"","orcid":"","institution":"Johns Hopkins University","correspondingAuthor":false,"prefix":"","firstName":"Iman","middleName":"","lastName":"Foroughmand","suffix":""},{"id":466969722,"identity":"357ea365-5d6b-4cc3-bdf3-f402d9ddf37a","order_by":2,"name":"Sara Ganjizadeh","email":"","orcid":"","institution":"Iran University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Sara","middleName":"","lastName":"Ganjizadeh","suffix":""},{"id":466969723,"identity":"ca7252c9-ce15-47ac-966a-a11dc8672971","order_by":3,"name":"Afshar Ghamari Khamene","email":"","orcid":"","institution":"Imam Khomeini Hospital, Tehran University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Afshar","middleName":"Ghamari","lastName":"Khamene","suffix":""},{"id":466969724,"identity":"eecd8ff5-a78d-4275-839b-ae7b5fd976d4","order_by":4,"name":"Faezeh Salahshour","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFklEQVRIiWNgGAWjYBAC9gYog4+Bh4EZxOAHEQkFuLXwHIAy2GBaJEGGJBiQosUALIJPC/vhwx8+MNTJs7GfPfy5oMIm3/j86sQPDwwY5PnFDmDXwpOWYDiD4bBhG09emvSMM2mW22683SwBdJjhzNkJWLXYM+QYJPMwHGBsY8gxY+ZtO2xgduPsBpCWBIPb2LXw8L//cPgPQ519G/8b488gLcYzzm7+gVeLRA5jMwMDc2KbRI6BNEiLAX/vNvy2SDwzZuwxOJzcJvHGTJrnTJqBxA3ebRYJBhI4/cLDn/z4w4+KOtt+/hzjzzwVNgb8/Wc33/xRYSPPL41dCwSgxIIEWKUEHuUYgP8AKapHwSgYBaNgBAAAGY1YD9sbjOAAAAAASUVORK5CYII=","orcid":"","institution":"Imam Khomeini Hospital, Tehran University of Medical Sciences","correspondingAuthor":true,"prefix":"","firstName":"Faezeh","middleName":"","lastName":"Salahshour","suffix":""},{"id":466969725,"identity":"8da0e1b1-2a15-47dc-a8fa-63b4ef6af209","order_by":5,"name":"Amirhossein kamalian","email":"","orcid":"","institution":"Shahid Sadoughi University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Amirhossein","middleName":"","lastName":"kamalian","suffix":""},{"id":466969726,"identity":"bcb67938-1d87-4384-8b96-0136ee93a998","order_by":6,"name":"Niloufar Ayoubi","email":"","orcid":"","institution":"Imam Khomeini Hospital, Tehran University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Niloufar","middleName":"","lastName":"Ayoubi","suffix":""},{"id":466969727,"identity":"551389d3-24b2-43b9-8f36-25b24f058605","order_by":7,"name":"Farid Azmoudeh-Ardalan","email":"","orcid":"","institution":"Imam Khomeini Hospital, Tehran University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Farid","middleName":"","lastName":"Azmoudeh-Ardalan","suffix":""},{"id":466969728,"identity":"ee2c56fa-65e7-4196-b2a6-bbe29a364ed7","order_by":8,"name":"Nasir Fakhar","email":"","orcid":"","institution":"Tehran University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Nasir","middleName":"","lastName":"Fakhar","suffix":""}],"badges":[],"createdAt":"2025-05-09 22:38:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6631565/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6631565/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":84216242,"identity":"c34d98b3-ba77-4562-a200-35ebff0176ab","added_by":"auto","created_at":"2025-06-09 10:41:58","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":221781,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparative Analysis of Mass-to-Spleen Signal Ratio Using T2-Weighted MRI Images for FNH and HCA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe calculation methodology for the mass-to-spleen signal ratio in FNH and HCA using T2-weighted MRI images. FNH images are positioned on the left, while HCA images are on the right. Annotations highlight the selected regions of interest (ROIs) within each lesion and the adjacent spleen tissue, used to measure signal intensities. The analysis reveals a significant distinction in the mass-to-spleen signal ratios, with HCA exhibiting a ratio of 0.854, contrasted with FNH's ratio of 0.499, underscoring the ratio's efficacy as a differentiator between these lesion types. Overlaid annotations report lesion area (mm²), minimum, maximum and average signal intensity, standard deviation of signal intensity, total sum of signal intensities, and lesion length (mm), illustrating the distinct diffusion profiles of FNH versus HCA.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/c590516f9892e45cf1508a2b.jpg"},{"id":84216244,"identity":"bdbe1ebd-00a0-4240-aa9f-6f25580de353","added_by":"auto","created_at":"2025-06-09 10:41:58","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":107435,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparative Analysis of DWI and ADC Maps for FNH and HCA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eA comparative analysis of Diffusion-Weighted Imaging (DWI) and Apparent Diffusion Coefficient (ADC) maps for Focal Nodular Hyperplasia (FNH) and Hepatocellular Adenoma (HCA). The left panel showcases a DWI image of FNH at a b-value of 400, where FNH exhibits higher signal intensity, highlighting its distinct radiological signature. In contrast, the right panel displays a DWI image of HCA under the same b-value, demonstrating marked differences. Two circular ROIs—one over the lesion and one over adjacent splenic tissue—provide signal intensity measurements. Additional annotations include area measurements (mm²), minimum, maximum, average signal intensity (SI), standard deviation (SD), total sum of signal intensities, and lesion length (mm), offering a comprehensive comparison of the lesion characteristics.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/2542089e45b0daac707589cd.jpg"},{"id":84214844,"identity":"4283726b-ff9d-4c26-a8b7-a03fb15419e3","added_by":"auto","created_at":"2025-06-09 10:33:58","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":420630,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClustered Heatmap of the Pairwise Correlation Matrix of Imaging Features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThis figure displays a hierarchical clustering of imaging variables. The color scale represents correlation coefficients, and dendrograms group similarly correlated features across T2-weighted, DWI, and contrast-enhanced metrics.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/6341e35ea03c1223c70b7020.jpg"},{"id":84214840,"identity":"07d05214-0732-4b27-9b42-5d11b7c73869","added_by":"auto","created_at":"2025-06-09 10:33:58","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":155047,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeatmap Representation of Class Separation Across Linear Classifiers\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThis figure illustrates the decision scores of Logistic Regression, Ridge Classifier, and LDA in principal component space, displaying classification confidence for each data point.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/c17e383264bfbb5b7812a542.jpg"},{"id":84216246,"identity":"1674bda3-2fed-4c17-8f94-7ba8e7dc4c50","added_by":"auto","created_at":"2025-06-09 10:41:58","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":133473,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eProbability Contour Maps for Nonlinear Classifiers in Principal Component Space\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThis figure shows the probability surfaces of SVM (RBF), Random Forest, and GBM in principal component space, highlighting varying likelihood contours for each class.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/5f9a86f76c60e80451b98651.jpg"},{"id":84216245,"identity":"13ad04f2-0549-4f9c-b1ad-1245824ded30","added_by":"auto","created_at":"2025-06-09 10:41:58","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":128789,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCorrelation Network Graph of Selected Imaging Features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThis figure displays interrelationships among key imaging variables. Node size indicates connectivity (node degree), while edge thickness conveys correlation strength, highlighting prominent associations among T2, arterial enhancement, DWI, and other diagnostic parameters.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Picture6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/5ca74e432a5646085be8e4d2.jpg"},{"id":90842574,"identity":"f571f5e8-02e2-46cf-b256-edb3e0265097","added_by":"auto","created_at":"2025-09-08 21:01:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2946881,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/75749399-d287-42a0-8b1d-e45d9554fa5d.pdf"},{"id":84214842,"identity":"7043fcef-90e2-400d-bba0-6b7666455a4f","added_by":"auto","created_at":"2025-06-09 10:33:58","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":28637,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryA.docx","url":"https://assets-eu.researchsquare.com/files/rs-6631565/v1/066beae6618dfba52d22d173.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eMRI-Based T2 and Arterial Radiomics to Differentiate Focal Nodular Hyperplasia (FNH) from Hepatocellular Adenoma (HCA): A High-Accuracy Spleen-Referenced Approach\u003c/p\u003e","fulltext":[{"header":"Key Points","content":"\u003cp\u003e1. T2-weighted mass-to-spleen signal ratio provides strong discrimination between FNH and HCA.\u003c/p\u003e\u003cp\u003e2. FNHs exhibit higher DWI signal intensities and mass-to-liver differences, particularly at b-50 and b-400.\u003c/p\u003e\u003cp\u003e3. Arterial-phase signal ratios further distinguish FNH from HCA subtypes.\u003c/p\u003e\u003cp\u003e4. Ensemble ML approaches (GBM, Random Forest) achieve superior test accuracy (up to 86.0%).\u003c/p\u003e\u003cp\u003e5. Biologically relevant radiomic features\u0026mdash;T2, arterial enhancement, spleen-reference\u0026mdash;drive high diagnostic precision.\u003c/p\u003e"},{"header":"Background","content":"\u003cp\u003eBenign liver lesions are frequently encountered in symptomatic and incidentally detected (\u0026lsquo;subclinical\u0026rsquo;) populations, with autopsy data indicating a prevalence exceeding 50% of adults, albeit often asymptomatic and discovered incidentally [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Ultrasound-based surveys covering large cohorts, such as a study of 45,319 patients, reveal focal lesions in 15.1% of participants, though FNH and HCA remain comparatively rare [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Despite their generally indolent nature, identifying these lesions accurately is critical for treatment planning, especially since the majority remain clinically silent or are discovered fortuitously [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Among benign tumors, hemangiomas predominate, while FNH is about ten times more common than HCA; however, the clinical significance of HCA can be far more substantial owing to its potential for serious complications [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDistinguishing FNH from HCA is paramount because these lesions display markedly different prognoses and management pathways, with FNH typically requiring no intervention, whereas HCA may be complicated by hemorrhage or malignant degeneration [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. In ambiguous scenarios, histopathological confirmation is sought, reflecting the continued difficulty of differentiating these lesions by imaging alone [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. This challenge underscores the growing need for advanced imaging protocols and computational approaches, which promise to reduce unnecessary biopsies and direct more targeted surveillance for individuals at higher risk.\u003c/p\u003e \u003cp\u003eConventional imaging techniques have progressively evolved yet still exhibit diagnostic blind spots in distinguishing FNH from HCA. Ultrasound remains an indispensable first-line tool but is hampered by limited specificity, even when enhanced with contrast agents [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Computed tomography offers improved resolution and multiphasic characterization\u0026mdash;particularly when a central scar is present\u0026mdash;yet can be imprecise for smaller or atypical lesions [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Gadoxetic acid\u0026ndash;enhanced MRI shows excellent accuracy by capitalizing on differential uptake in the hepatobiliary phase\u0026mdash;FNH commonly appears hyperintense, whereas HCA remains hypointense\u0026mdash;although interobserver variability and atypical presentations persist [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. PET/CT using 18F-fluoromethylcholine has shown promise but is not yet widely adopted in routine practice. Whereas FNH is managed conservatively, HCA\u0026mdash;especially lesions\u0026thinsp;\u0026ge;\u0026thinsp;5 cm or β-catenin-activated subtypes\u0026mdash;carries a 20\u0026ndash;25% hemorrhage risk and up to 8% malignant-transformation rate, necessitating resection in selected cases. Correct classification informs management\u0026mdash;observation for FNH versus potential resection for HCA\u0026mdash;directly impacting patient outcomes. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMRI protocols have become increasingly sophisticated, incorporating T1- and T2-weighted acquisitions, diffusion-weighted imaging (DWI) at multiple b-values, and multiphasic contrast-enhanced sequences that illuminate vascular perfusion and hepatobiliary excretion patterns [\u003cspan additionalcitationids=\"CR13 CR14 CR15\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Lower b-values (b-50, b-400) can accentuate lesion conspicuity, while high b-values improve discrimination between benign and malignant tissues by detecting restricted diffusion. Moreover, the advent of FIESTA sequences has introduced spleen-referenced signal intensities, providing an internal control that accounts for variations in liver parenchyma and thus refines diagnostic accuracy. Existing protocols depend on subjective scar recognition or hepatocyte-specific contrast uptake, which may be equivocal in up to one-third of cases; our spleen-referenced radiomics approach provides objective, contrast-neutral metrics to overcome these limitations [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe molecular and pathological diversity of HCA further complicates diagnosis, as different subtypes\u0026mdash;β-catenin-activated, inflammatory, HNF1A-mutated, and others\u0026mdash;carry distinct risks of malignant transformation and hemorrhage [\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Non-steatotic adenomas, often linked to β-catenin or inflammatory pathways, may mimic FNH on routine imaging, whereas steatotic (HNF1A-mutated) adenomas exhibit signal drop on opposed-phase T1 imaging [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Given that invasive biopsies remain the definitive standard for resolving such ambiguities, advanced MRI-based characterization can offer a non-invasive means of aligning specific imaging patterns with histopathological features. We aimed to develop and validate a spleen-referenced radiomic signature on T2, arterial-phase, and DWI MRI to noninvasively discriminate FNH from HCA [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHarnessing advanced MRI sequences and machine learning methods allowed us to elucidate the interplay between lesion anatomy and vascular behavior, ensuring that each retained feature bore direct clinical significance. This study therefore emphasizes that informed feature selection\u0026mdash;grounded in both biological rationale and radiological context\u0026mdash;drives a more interpretable and accurate diagnostic pipeline, ultimately enabling precise lesion characterization in complex hepatic imaging scenarios.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eStudy Design and Participants\u003c/h2\u003e\n \u003cp\u003eThis single-center retrospective study received approval from the Institutional Review Board (IR.TUMS.IKHC.REC.1402.121) with a waiver of informed consent owing to its retrospective nature, and was conducted in accordance with the 1964 Declaration of Helsinki and its later amendments. From January 2014 to January 2021, patients were identified based on dynamic liver MRI reports that suggested FNH or HCA. Individuals were eligible for inclusion if they had histopathological confirmation by core needle biopsy or surgical excision and a lesion size of at least 10 mm to ensure reliable imaging evaluation. Pilot reproducibility testing demonstrated that radiomic features extracted from lesions\u0026thinsp;\u0026lt;\u0026thinsp;10 mm yielded intraclass correlation coefficients\u0026thinsp;\u0026lt;\u0026thinsp;0.60, justifying the 10 mm minimum size. The retrospective cohort included MRI studies from both 1.5 T (58%) and 3 T (42%) scanners, reflecting national clinical practice patterns. In patients with multiple hypervascular nodules and clinical risk factors for HCA, histopathologic characterization of a single lesion could be generalized to morphologically similar nodules. Hepatocellular adenomas were further subtyped into inflammatory, HNF-1\u0026alpha;-mutated, \u0026beta;-catenin-activated, or unclassified variants. Exclusion criteria encompassed MRI examinations compromised by severe motion or susceptibility artifacts, lesions smaller than 5 mm, liver lesions previously treated with locoregional or systemic therapies, incomplete data that impeded definitive characterization, and nonspecific pathology reports labeling the lesion as \u0026ldquo;hepatocellular neoplasm of uncertain potential.\u0026rdquo;\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eMRI Protocol\u003c/h3\u003e\n\u003cdiv class=\"Heading\"\u003eMRI Protocol\u003c/div\u003e\n\u003cp\u003eAll examinations were performed on either a 3.0 T scanner (Discovery 750\u0026reg;, GE Healthcare, Chicago, IL) or a 1.5 T system (Optima MR360\u0026reg;, GE Healthcare), harmonised via weekly phantom calibrations and z-score intensity normalisation to maintain\u0026thinsp;\u0026lt;\u0026thinsp;4% variance across 1.5 T and 3 T scanners. Patients were imaged in supine position. The protocol included axial and coronal T2-weighted half-Fourier single-shot turbo spin-echo (HASTE) sequences, axial fast imaging employing steady-state acquisition (FIESTA) with a typical repetition time echo time (TR/TE) of about 3\u0026ndash;6/1\u0026ndash;2 ms, axial in-phase/opposed-phase breath-hold gradient-echo T1-weighted sequences with approximate TR/TE of 200/2.2 ms, and axial free-breathing single-shot echo-planar diffusion-weighted imaging (DWdI) at b-values of 50, 400, and 800 s/mm\u0026sup2;. Free-breathing single-shot EPI was chosen following pilot breath-hold trials that incurred 18% motion-triggered repeat scans, whereas free-breathing improved SNR and reproducibility. Dynamic contrast-enhanced sequences were acquired using a three-dimensional fat-suppressed gradient-echo technique before and after administering gadopentetate dimeglumine (Magnevist\u0026reg;, Bayer HealthCare) at 0.1 mmol/kg, followed by a 20 mL saline flush at 2 mL/s. Arterial, portal venous, and delayed phases were obtained at appropriate time intervals post-injection. Arterial phase was acquired 18\u0026ndash;22 s after automated bolus triggering at 100 HU in the abdominal aorta.\u003c/p\u003e\n\u003ch3\u003eImage Analysis\u003c/h3\u003e\n\u003cp\u003eTwo board-certified abdominal radiologists, each with more than five years of experience, interpreted all MRI scans independently and were blinded to clinical and pathological data. Images were reviewed on a dedicated picture archiving and communication system (INFINITT Healthcare Co. LTD, Seoul, South Korea). Lesion features were categorized into qualitative and quantitative metrics. Qualitative assessments included signal intensity relative to the liver on T1- and T2-weighted imaging, presence or absence of intralesional fat, hemorrhage, or cystic transformation, and enhancement patterns across arterial, portal, and delayed phases. These qualitative descriptors encompassed morphological indicators such as lobulation, lesion capsule integrity (true, pseudo, or absent), and scar characteristics (signal on T1/T2, arterial enhancement, portal/delayed washout).\u003c/p\u003e\n\u003cp\u003eQuantitative variables comprised traditional measurements (maximum lesion diameter, T1/T2 signal intensity, apparent diffusion coefficient [ADC]) as well as a broad range of derived metrics (Figs. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e \u0026amp; \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). These computed values included lesion-to-liver and lesion-to-spleen ratios for T2-weighted, FIESTA, and contrast-enhanced sequences, differences in signal intensity between in-phase and opposed-phase imaging, and multiple DWI-based features such as diffusion restriction and numeric comparisons across different b-values. All spleen-referenced signal ratios were subsequently normalised by the splenic parenchymal standard deviation to mitigate inter-scan and inter-individual variability. In total, 108 distinct parameters were evaluated or derived, spanning morphological (e.g., capsular integrity, lobulation), demographic (e.g., patient age, gender), scanner-related (e.g., MRI device model), qualitative imaging findings (e.g., atoll sign, internal structure, focal high T2 signal), and quantitative indices (e.g., signal intensity ratios, ADC ratios, differences in b-values, delayed phase enhancements, T1 fat-suppressed signal intensity). Each lesion was labeled as HCA or FNH (binary outcome variable) based on pathology. A subset of variables specifically captured adenoma subtyping (e.g., inflammatory, HNF1\u0026alpha;-inactivated, \u0026beta;-catenin-activated) when applicable. Regions of interest (ROIs) were placed to maximize measurement area while excluding scar tissue, necrotic foci, large vessels, and bile ducts.\u003c/p\u003e\n\u003cp\u003eReliability testing involved both intra-observer and inter-observer analyses. For inter-observer evaluation, the two radiologists independently measured lesion diameter, T2 signal intensity, ADC values, and contrast enhancement in a random set of 30 cases. Intraclass correlation coefficients (ICCs) were calculated using a two-way random-effects model with absolute agreement. Intra-observer ICC values were 0.95 for T2 signal intensity, 0.97 for ADC measurements, 0.93 for lesion diameter, and 0.94 for dynamic contrast enhancement. Inter-observer ICC values were 0.92 for T2 signal intensity, 0.95 for ADC measurements, 0.90 for lesion diameter, and 0.91 for dynamic contrast enhancement. A test\u0026ndash;retest reliability assessment was performed in 20 patients who underwent repeat MRI within two weeks; repeat measurements also demonstrated high concordance (ICC\u0026thinsp;\u0026gt;\u0026thinsp;0.90 in all tested parameters).\u003c/p\u003e\n\u003ch3\u003eQuantitative Analysis Formulas\u003c/h3\u003e\n\u003cp\u003eFour representative derived indices included Mass-to-Liver SID = (Signal_Lesion\u0026thinsp;\u0026minus;\u0026thinsp;Signal_Liver), Mass-to-Spleen SID = (Signal_Lesion\u0026thinsp;\u0026minus;\u0026thinsp;Signal_Spleen), Mass-to-Liver SIR = (Signal_Lesion\u0026thinsp;\u0026minus;\u0026thinsp;Signal_Liver)/Signal_Liver, and Mass-to-Spleen SIR = (Signal_Lesion\u0026thinsp;\u0026minus;\u0026thinsp;Signal_Spleen)/Signal_Spleen. Further signal-intensity ratios, differences, and logarithmic transforms were computed as dictated by the full variable list.\u003c/p\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n \u003ch2\u003eStatistical Analysis\u003c/h2\u003e\n \u003cp\u003eAll collected data were entered into a secure database and processed in Python. Normality of distribution for continuous variables was assessed with the Kolmogorov\u0026ndash;Smirnov test. Values deemed outliers via interquartile range (IQR) or absolute z-scores greater than 3 were discarded to reduce skew. Missing data were handled by mean (continuous) or mode (categorical) imputation, with complete removal of instances showing excessive missingness. Continuous variables were reported as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation, and categorical data were expressed as frequencies and percentages. Group differences were examined using the independent-sample t-test for continuous outcomes and either chi-square or Fisher\u0026rsquo;s exact tests for categorical comparisons when sample sizes were small.\u003c/p\u003e\n \u003cp\u003eData preprocessing included z-score normalization of all continuous features. The final dataset comprised 197 lesions from 98 patients, with the target label (Diagnosis) encoded as 1 (HCA) or 0 (FNH). Each lesion was analysed independently; mixed-effects logistic regression with patient ID as a random effect confirmed negligible clustering bias (\u0026Delta;AUC\u0026thinsp;\u0026le;\u0026thinsp;0.01) Stratified random splitting assigned 70% of cases to the training set and 30% to the test set while preserving class balance. Feature selection was conducted via recursive feature elimination (RFE) with a logistic regression estimator; each step involved removing the least important predictor and re-evaluating performance through five-fold cross-validation. Recursive feature elimination removed, at each iteration, the predictor with the lowest mean SHAP value across five-fold cross-validation to prioritise features with greatest global importance. Principal component analysis (PCA) was also tested for dimensionality reduction, retaining principal components that captured 95% of variance.\u003c/p\u003e\n \u003cp\u003eEight machine learning algorithms were implemented and tuned using grid search with five-fold cross-validation, including Logistic Regression with PCA, Random Forest, Support Vector Machine (SVM) with RBF kernel, Gradient Boosting Machine (GBM), Ridge Classifier, Perceptron, Linear Discriminant Analysis (LDA), and Stochastic Gradient Descent (SGD) Classifier. The eight classifiers\u0026mdash;encompassing linear, kernel-based, ensemble, and stochastic methods\u0026mdash;were selected based on their prevalence in hepatic radiomics literature, facilitating comprehensive bias\u0026ndash;variance analysis. Key hyperparameters such as regularization strengths (C or \u0026alpha;), tree depth, learning rate, and number of boosting iterations were optimized to maximize generalization. Model performance was evaluated on the training and test subsets based on accuracy, precision, recall (sensitivity), F1-score, and area under the receiver operating characteristic curve (AUC-ROC). Feature importance was determined using Gini impurity for tree-based methods (Random Forest, GBM), standardized coefficients for linear models (Ridge, LDA, Perceptron), permutation-based approaches for SVM, and component loadings for logistic regression when coupled with PCA. All modeling procedures used Scikit-Learn (v1.2.0), with random seeds set to 42 to ensure reproducibility. A full list of the 108 features\u0026mdash;comprising 22 morphological, 41 first-order, 27 textural (GLCM, GLRLM, GLSZM), 10 spleen-referenced, and 8 diffusion-delta metrics\u0026mdash;is provided in Supplementary Appendix A.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eDemographic and Lesion Distribution\u003c/h2\u003e \u003cp\u003eA total of 197 lesions from 98 patients were evaluated (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), comprising 86 females (87.7%) and 12 males (12.4%) with a mean age of 33.95\u0026thinsp;\u0026plusmn;\u0026thinsp;9.28 years (range 15\u0026ndash;66). Among these, 77 lesions (39.0%) were diagnosed as FNH and 120 (61.0%) as HCA, of which 87 were categorized as non-steatotic HCAs. Ten patients presented with multiple FNHs, whereas 82 patients had multiple HCAs. Statistically significant differences were observed in age (p\u0026thinsp;=\u0026thinsp;0.039) but not in sex distribution (p\u0026thinsp;=\u0026thinsp;1.00). Differences in lobulation, intralesional fat, central scarring, cystic/hemorrhagic changes, atoll sign, and capsular features all reached significance between FNHs and HCAs (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, except for scarring at p\u0026thinsp;=\u0026thinsp;0.006). Similar comparisons of FNHs and NSHCAs indicated significant differences in lobulation, intralesional fat or iron, scarring, hypointense arterial rim, atoll sign, and capsule (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for most variables, with p\u0026thinsp;=\u0026thinsp;0.046 for intralesional iron and p\u0026thinsp;=\u0026thinsp;0.003 for arterial rim).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Lesion Characteristics Among FNHs, HCAs, and NSHCAs\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFNHs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHCAs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNSHCAs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ep-value (FNHs vs. HCAs)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep-value (FNHs vs. NSHCAs)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77 (39.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e120 (61.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e197\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eMean diameter (mm\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48.83\u0026thinsp;\u0026plusmn;\u0026thinsp;26.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e37.97\u0026thinsp;\u0026plusmn;\u0026thinsp;30.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e39.27\u0026thinsp;\u0026plusmn;\u0026thinsp;29.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e42.34\u0026thinsp;\u0026plusmn;\u0026thinsp;29.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e.009\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e.027\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGender (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9 (11.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21 (17.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11 (12.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e30 (15.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68 (88.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e99 (82.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e76 (87.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e167 (84.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eMean age (years\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e33.95\u0026thinsp;\u0026plusmn;\u0026thinsp;9.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e38.16\u0026thinsp;\u0026plusmn;\u0026thinsp;9.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e37.34\u0026thinsp;\u0026plusmn;\u0026thinsp;9.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e35.52\u0026thinsp;\u0026plusmn;\u0026thinsp;9.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e.039\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e.118\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eLobulation (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75 (97.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e41 (34.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29 (33.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e116 (58.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eIntralesional Fat (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e36 (30.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13 (14.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e38 (30.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e.007\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eIntralesional Iron (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10 (8.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (8.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e11 (5.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e.053\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e.046\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eCentral scar (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71 (92.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55 (45.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e39 (44.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e132 (67.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eInternal Changes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCystic changes (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNone (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2 (2.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e10 (5.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e.006\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e.137\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHemorrhagic changes (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNone (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (5.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3 (3.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMixed cystic and hemorrhagic changes (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNone (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eHypointense arterial rim (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11 (14.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30 (25.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30 (34.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e42 (20.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e.071\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e.003\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eAtoll sign (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNone (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27 (22.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e27 (31.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e27 (13.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eHigh T2 foci\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (5.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNone (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNone (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4 (2.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e.031\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eRestriction (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48 (62.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e23 (26.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e78 (37.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eCapsule\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (5.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (8.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7 (3.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePseudo (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29 (37.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4 (3.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4 (4.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e33 (16.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypointense (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2 (1.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eThis table summarizes demographic and morphological data for patients with FNH, HCA, and NSHCA, highlighting significant differences in lesion count, diameter, gender, age, lobulation, intralesional components, and capsular status.\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eQualitative Imaging and Morphologic Features\u003c/h3\u003e\n\u003cp\u003eT2- and T1-weighted sequences, precontrast fat-suppressed images, and arterial/portal/delayed phases showed significant qualitative differences (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) when comparing FNHs and HCAs (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). FNHs had notably more frequent visible central scars (92.2%) than HCAs (45.8%, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). On T2-weighted imaging, 74.0% of FNH scars were hyperintense compared to 19.1% in HCAs. T1-weighted sequences identified hypointense scars in 81.1% of FNHs and 18.3% of HCAs (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Comparison with NSHCAs revealed significant distinctions in T2- and T1-weighted and delayed post-contrast sequences (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, 0.014, \u0026lt;\u0026thinsp;0.001, respectively) but no difference in arterial or portal phases (p\u0026thinsp;=\u0026thinsp;0.297, 0.053).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSequence and Signal Intensity Characteristics of FNHs, HCAs, and NSHCAs\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eSequence and signal intensity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFNHs\u003c/p\u003e \u003cp\u003e(77 lesions)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHCAs\u003c/p\u003e \u003cp\u003e(120 lesions)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNSHCAs\u003c/p\u003e \u003cp\u003e(87 lesions)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep-value (FNHs vs. HCAs)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ep-value (FNHs vs. NSHCAs)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eT2-weighted\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75 (97.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e71 (59.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e72 (82.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47 (39.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9 (10.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003ePrecontrast T1 fat suppressed\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (9.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10 8.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e0.014\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70 (90.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100 (83.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e72 (82.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10 (8.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9 (10.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eArterial phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.297\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4 (4.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e76 (98.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e97 (80.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e82 (94.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003ePortal phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24 (20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32 (41.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35 (29.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30 (34.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (58.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61 (50.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e51 (58.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eDelayed phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34 (28.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14 (16.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.0001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e65 (84.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21 (25.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28 (32.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12 (15.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55 (45.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e45 (51.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCentral scars\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eT2-weighted\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (5.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (8.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53 (68.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15 (12.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12 (13.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e57 (74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23 (19.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e20 (23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003ePrecontrast T1 fat suppressed\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 (81.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19 (15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e17 (19.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 (81.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19 (21.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eFIESTA\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (2.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11 (9.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11 (12.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4 (3.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47 (61)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (5.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 ( 8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49 (63.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e18 (20.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eArterial phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60 (77.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e31 (25.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e34 (39.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (4.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5 (5.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (5.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (4.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e64 (83.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e41 (34.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e39 (44.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003ePortal phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNot visible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32 (41.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e107 (89.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e80 (91.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (58.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13 (10.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (8.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eDelayed phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (2.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2 (2.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68 (88.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e33 (27.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e26 (29.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71 (92.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35 (29.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28 (32.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eThis table details T2-weighted, precontrast T1 fat-suppressed, and dynamic post-contrast signal intensities, including the prevalence of scar visibility and lesion enhancement patterns.\u003c/em\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eLesion Size and Quantitative MRI Metrics\u003c/h2\u003e \u003cp\u003eThe mean lesion diameter across all cases was 41.27\u0026thinsp;\u0026plusmn;\u0026thinsp;28.34 mm (range 7\u0026ndash;146 mm) (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). FNHs measured 48.95\u0026thinsp;\u0026plusmn;\u0026thinsp;26.64 mm, significantly larger than HCAs at 36.35\u0026thinsp;\u0026plusmn;\u0026thinsp;28.42 mm (p\u0026thinsp;=\u0026thinsp;0.002). Mean lesion T2-weighted signal intensity (SI), mass-to-spleen T2 signal intensity ratios (SIRs), and mass-to-liver out-of-phase SIDs were significantly different between FNHs and HCAs (p-values ranging from \u0026lt;\u0026thinsp;0.001 to 0.043). When comparing FNHs with NSHCAs, similar metrics were significant, with an additional difference in mass-to-liver T1 SIR under fat-suppressed imaging (p\u0026thinsp;=\u0026thinsp;0.01).\u003c/p\u003e \u003cp\u003eDynamic sequence analysis showed that FNHs had higher mean arterial phase SI (p\u0026thinsp;=\u0026thinsp;0.01) and greater mass-to-liver SID/SIR differences in the arterial phase (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) than HCAs. Portal and delayed phases also exhibited statistically significant SID or SIR differences (p\u0026thinsp;=\u0026thinsp;0.002 and p\u0026thinsp;=\u0026thinsp;0.038). Differences in diffusion-weighted imaging (DWI) were observed primarily in the b-50 and b-400 sequences, with FNHs showing higher lesion SIs (p\u0026thinsp;=\u0026thinsp;0.002). Comparisons of FNHs and NSHCAs indicated significance in multiple DWI-derived parameters, including b-50 mass SI, b-50 mass-to-liver SID, and b-400 mass-to-liver SID (p-values ranging from 0.008 to 0.038).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMRI Signal Intensity Characteristics of FNHs, HCAs, and NSHCAs\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eSequence and signal intensity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFNHs\u003c/p\u003e \u003cp\u003e(77 lesions)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHCAs\u003c/p\u003e \u003cp\u003e(120 lesions)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNSHCAs\u003c/p\u003e \u003cp\u003e(87 lesions)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep-value (FNHs vs. HCAs)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ep-value (FNHs vs. NSHCAs)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eT2-weighted\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75 (97.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e71 (59.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e72 (82.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47 (39.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9 (10.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003ePrecontrast T1 fat suppressed\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (9.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10 8.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e0.014\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70 (90.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100 (83.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e72 (82.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10 (8.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9 (10.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eArterial phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.297\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4 (4.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e76 (98.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e97 (80.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e82 (94.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003ePortal phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24 (20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32 (41.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35 (29.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30 (34.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (58.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61 (50.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e51 (58.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eDelayed phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34 (28.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14 (16.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.0001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e65 (84.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21 (25.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28 (32.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12 (15.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55 (45.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e45 (51.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCentral scars\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eT2-weighted\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (5.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (8.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53 (68.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15 (12.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12 (13.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e57 (74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23 (19.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e20 (23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003ePrecontrast T1 fat suppressed\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 (81.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19 (15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e17 (19.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 (81.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19 (21.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eFIESTA\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (2.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11 (9.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11 (12.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4 (3.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47 (61)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (5.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 ( 8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49 (63.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e18 (20.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eArterial phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60 (77.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e31 (25.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e34 (39.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (4.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5 (5.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (5.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (4.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e64 (83.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e41 (34.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e39 (44.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003ePortal phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNot visible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32 (41.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e107 (89.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e80 (91.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (58.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13 (10.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7 (8.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eDelayed phase\u003c/p\u003e \u003cp\u003eLesion count (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (2.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2 (2.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyper\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68 (88.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e33 (27.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e26 (29.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVisible scar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71 (92.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35 (29.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28 (32.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eThis table reports quantitative signal intensities, signal intensity differences (SID), and signal intensity ratios (SIR) for conventional T1/T2 sequences, dynamic contrast-enhanced phases, and diffusion-weighted imaging.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eHierarchical clustering of correlation data is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, demonstrating grouping among lesion-derived metrics, T2-weighted features, and dynamic contrast parameters.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClassification Model Outcomes\u003c/h2\u003e \u003cp\u003eEight machine learning classifiers (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e)\u0026mdash;Logistic Regression with PCA, Random Forest, SVM (RBF), Gradient Boosting Machine (GBM), Ridge Classifier, Perceptron, Linear Discriminant Analysis (LDA), and Stochastic Gradient Descent (SGD)\u0026mdash;were evaluated for lesion classification. Both GBM and Random Forest showed the highest area under the receiver operating characteristic curve (AUC-ROC) of 0.915, with respective test accuracies of 86.021% and 85.376%. The SVM model reached an AUC-ROC of 0.879, while Logistic Regression with PCA, Perceptron, and LDA achieved similar AUC-ROC values ranging from 0.843 to 0.850. The SGD Classifier obtained the lowest AUC-ROC of 0.830 and a test accuracy of 73.333%. Figures\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e provide visual depictions of model decision boundaries and probability contours in principal component space.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel Performance Metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"10\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrain Accuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTest Accuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTrain Precision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTest Precision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTrain Recall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTest Recall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eTrain F1-score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTest F1-score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eAUC-ROC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression (PCA)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e89.237%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e79.204%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.756%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e81.215%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e88.913%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e77.868%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e89.832%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e79.495%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.850\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e98.521%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e85.376%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e98.885%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e86.234%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e98.149%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e84.215%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e98.516%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e85.206%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.915\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM (RBF Kernel)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e96.872%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83.429%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e97.110%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e84.125%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e96.625%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e82.376%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e96.867%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e83.250%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.879\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting (GBM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e99.085%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e86.021%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e99.340%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87.025%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e98.923%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e85.126%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e99.131%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e86.065%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.915\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRidge Classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e94.891%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e80.000%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e97.500%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87.880%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e93.980%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e78.380%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e95.710%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e82.860%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerceptron\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e96.354%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e78.333%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e97.560%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e83.330%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e96.390%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e81.080%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e96.970%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e82.190%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.850\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLinear Discriminant Analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e94.160%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e80.000%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e97.470%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87.880%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e92.770%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e78.380%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e95.060%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e82.860%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSGD Classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e97.080%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e73.333%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e97.590%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e80.000%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e97.590%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e75.680%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e97.590%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e77.780%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eThis table presents accuracy, precision, recall, F1-score, and AUC-ROC for the training and test sets of each classifier. Metrics capture performance variations in differentiating Hepatocellular Adenoma from FNH..\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eRelative Feature Contributions\u003c/h2\u003e \u003cp\u003eT2-weighted lesion intensity ranked highly across classifiers, along with arterial phase signal ratios, Fiesta-weighted splenic ratios, hypoattenuation in arterial rims, and selected DWI differentials (e.g., b-50 or b-400 signal levels). Linear methods assigned high coefficients to features such as T1 out-of-phase signal ratio or lesion-to-liver T2 ratios, while ensemble approaches (Random Forest, GBM) emphasized broader sets of features with varying Gini importances (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e provides a correlation network among selected imaging attributes, with node size reflecting connectivity and edge thickness representing correlation magnitude.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFeature Importance Metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC-ROC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTop Feature 1 (Value)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTop Feature 2 (Value)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTop Feature 3 (Value)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTop Feature 4 (Value)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTop Feature 5 (Value)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression (PCA)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2-Weighted Central Lesion Intensity (2.992)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eArterial Phase Signal Ratio (2.945)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Ratio (2.937)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHypoattenuation in Arterial Rim (2.912)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eT1 Signal Differential (2.912)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2-Weighted Central Lesion Intensity (0.139)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eArterial Phase Signal Ratio (0.107)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eT1 Signal Differential (0.098)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Intensity (0.066)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDWI B50 Signal Differential (0.063)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM (RBF Kernel)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.879\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2-Weighted Central Lesion Intensity (0.072)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePortal Venous Phase Signal Differential (0.040)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLesion-to-Liver T2 Intensity Ratio (0.035)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Intensity (0.030)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDWI B800 Signal Differential (0.020)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting (GBM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2-Weighted Central Lesion Intensity (0.352)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eArterial Phase Signal Ratio (0.130)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Intensity (0.118)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eT1 Signal Differential (0.087)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDWI B50 Signal Differential (0.057)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRidge Classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDelayed Phase Hepatic Signal Intensity (0.284)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLesion-to-Liver T2 Intensity Ratio (0.246)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Intensity (0.234)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDWI B400 Signal Differential (0.153)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHypoattenuation in Arterial Rim (0.152)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerceptron\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT1 Out-of-Phase Signal Ratio (12.283)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLesion-to-Liver T2 Intensity Ratio (12.259)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDelayed Phase Hepatic Signal Intensity (10.047)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eT1 In-Phase Lesion Signal Intensity (8.251)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eT1 Signal Differential (8.042)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLinear Discriminant Analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDelayed Phase Hepatic Signal Intensity (2.109)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLesion-to-Liver T2 Intensity Ratio (1.740)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Intensity (1.660)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDWI B400 Signal Differential (1.125)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHypoattenuation in Arterial Rim (1.061)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSGD Classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLesion-to-Liver T2 Intensity Ratio (46.771)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eT1 Out-of-Phase Signal Ratio (40.208)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eT1 In-Phase Lesion Signal Intensity (35.286)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDelayed Phase Hepatic Signal Intensity (28.469)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eFiesta-Weighted Splenic Signal Intensity (26.211)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eThis table identifies the top five features per model, with quantitative importance or coefficient values. Methods for determining feature impact include principal component loadings, Gini impurity, permutation importance, or standardized coefficients.\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eRadiomics extracts numerous quantitative imaging descriptors\u0026mdash;intensity ratios, spatial texture, morphological features\u0026mdash;to capture subtle lesion phenotypes beyond human perceptual thresholds [\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. As high-dimensional datasets risk redundancy and noise, techniques optimizing feature subsets improve classification power. Ensemble machine learning methods further mitigate overfitting by amalgamating multiple \u0026ldquo;weak\u0026rdquo; learners into robust predictive frameworks [\u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWithin these radiomic pipelines, T2-weighted central lesion intensity, arterial phase signal ratios, and multi-b-value diffusion profiles constitute pivotal parameters for classifying FNH versus HCA. T2 hyperintensity or hypointensity in central regions may indicate fibrotic or inflammatory changes, while arterial-phase ratios track rapid enhancement patterns characteristic of particular lesion types [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Comparatively, multi-b-value DWI reveals differences in water mobility\u0026mdash;benign lesions often maintain higher ADC values\u0026mdash;highlighting the synergistic effect of combining lower and higher b-values for both detection and differential assessment [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThese advanced radiomic features often align with distinct molecular pathways in HCA subtypes. β-catenin-activated HCA, for instance, shows pronounced arterial hyperenhancement and reduced ADC values, reflecting high cellular density and a propensity for malignant transition [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Inflammatory HCA may overlap with FNH in certain sequences due to persistent enhancement, whereas HNF1A-mutated adenomas display steatosis-linked signal loss on opposed-phase T1 imaging [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Although such overlaps can complicate noninvasive classification, multi-parametric MRI and radiomics significantly narrow the diagnostic gap. To further mitigate over-fitting, future work will employ nested cross-validation and learning-rate decay with sub-sampling.\u003c/p\u003e \u003cp\u003eThe superior performance of ensemble models, including Random Forest (RF) and Gradient Boosting (GBM), is attributable to their capacity to capture non-linear associations and interactions among numerous imaging features [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. By iteratively refining decision boundaries, GBM techniques achieve higher accuracy metrics than single or linear classifiers\u0026mdash;some studies report AUC values exceeding 0.90 when differentiating complex hepatic lesions [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Nonetheless, linear models, while exhibiting lower predictive power, retain the advantage of interpretability, facilitating the elucidation of individual feature contributions in radiomic analyses [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Our contrast-neutral approach achieved AUC 0.915, comparable to hepatobiliary-phase MRI reports of 0.88\u0026ndash;0.91 [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], and remains applicable in patients with contraindications to hepatocyte-specific agents.\u003c/p\u003e \u003cp\u003eSpleen-referenced normalization neutralises hepatic parenchymal heterogeneity and scanner gain drift, improving inter-patient comparability without requiring liver segmentation [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. These normalized signal intensities or ADC values demonstrate improved discriminative power over conventional liver-to-lesion measurements, particularly in advanced disease settings where hepatic texture is highly heterogeneous [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Consequently, splenic normalization offers a strong adjunct to multiparametric approaches, enabling robust cross-patient and cross-institutional comparisons.\u003c/p\u003e \u003cp\u003eFurther refining MRI protocols can amplify their diagnostic yield, with T1 and T2 relaxation mapping providing quantitative tissue characterization, and 3D gradient-echo sequences delivering more consistent volumetric coverage [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Combined with parametric DWI analyses\u0026mdash;encompassing multiple b-values for a comprehensive assessment of lesion diffusivity\u0026mdash;these enhancements may strengthen both the initial detection and subsequent classification of FNH and HCA.\u003c/p\u003e \u003cp\u003eBy reliably distinguishing FNH from HCA, radiomics-based MRI can diminish reliance on invasive biopsy, accelerate patient triage, and slash healthcare costs associated with prolonged diagnostic workups [\u003cspan additionalcitationids=\"CR46\" citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. This precision-based imaging approach is particularly relevant for individuals in whom HCA poses a high hemorrhagic or malignant transformation risk, necessitating closer follow-up or resection.\u003c/p\u003e \u003cp\u003eDespite promising outcomes, radiomic investigations frequently confront retrospective designs, limited sample sizes, and underrepresentation of rarer tumor subtypes, all of which challenge generalizability [\u003cspan additionalcitationids=\"CR49\" citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDeep learning\u0026ndash;based convolutional neural networks (CNNs) have demonstrated superior accuracy in certain hepatic lesion tasks by automatically extracting hierarchical spatial features [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. Hybrid models, blending CNN-derived embeddings with radiomic features, may offer an optimal balance between predictive accuracy and interpretability, especially in multi-center studies that necessitate robust generalization. Future research initiatives should explore prospective trials that harmonize MRI protocols across institutions, employ explainable AI strategies for clinical acceptance, and eventually integrate molecular markers (e.g., CTNNB1) to achieve highly personalized, non-invasive diagnostics [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. By avoiding hepatocyte-specific contrast and reducing biopsy rates, the proposed method could save an estimated \u0026pound;300\u0026ndash;\u0026pound;500 per indeterminate lesion and potentially \u0026gt;\u0026pound;2\u0026nbsp;million annually at a national level.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eLimitations:\u003c/h2\u003e \u003cp\u003e(i) Single-centre design may limit generalizability. (ii) Despite phantom harmonization, residual 1.5 T/3 T effects (\u0026lt;\u0026thinsp;4% CV) may persist. (iii) Free-breathing DWI can introduce blurring, although test\u0026ndash;retest ICCs remained\u0026thinsp;\u0026ge;\u0026thinsp;0.92. (iv) Intentional omission of hepatobiliary-phase imaging tests a universal, contrast-neutral pipeline; future studies will quantify the added value of hepatobiliary sequences.\u003c/p\u003e \u003cp\u003eNext steps include prospective multicentre validation with protocol harmonization, integration of CNN embeddings for feature extraction, and correlation with CTNNB1 and HNF1A genotypes.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe comprehensive exploration of lesion-specific imaging parameters, augmented by robust machine learning algorithms, underscores the synergy between data-driven feature selection and biologically consistent radiological interpretation. The strong performance of ensemble models indicates that identifying pivotal features\u0026mdash;such as T2 central lesion intensity, arterial phase signal ratios, and spleen-based references\u0026mdash;can substantially refine non-invasive differentiation of FNH and HCA. Moreover, the nuanced interactions revealed in various classifiers highlight that integrating clinically relevant radiomic markers with sophisticated pattern-recognition methods yields both actionable insights and heightened diagnostic precision. This approach paves the way for reducing unnecessary biopsies, guiding patient management strategies, and inspiring further research into standardized protocols for benign hepatic lesions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis single-centre retrospective study was approved by the Institutional Review Board of Tehran University of Medical Sciences (IR.TUMS.IKHC.REC.1402.121) with a waiver of written informed consent. All procedures were performed in accordance with the 1964 Declaration of Helsinki and its later amendments.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDe-identified MRI datasets, radiomic feature tables, and all machine-learning analysis scripts that support the findings of this study are not publicly archived but will be supplied by the corresponding author upon reasonable request and can be provided directly to the journal’s editorial team or other qualified researchers for peer-review and academic verification.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research received no external funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA.A. and F.S. conceived and coordinated the study; I.F. and A.K. developed and executed the radiomics and machine-learning pipeline, with methodological support from S.G. Clinical image acquisition and lesion segmentation were performed by A.A., A.G.K. and N.A., while N.F. oversaw patient selection and clinical data curation. F.A.-A. provided histopathological verification. A.A. and S.G. drafted the manuscript; I.F., A.K. and F.S. carried out statistical analysis and interpretation of results; all authors critically revised the work for important intellectual content. F.S. served as senior supervisor and corresponding author. All authors read and approved the final version of the manuscript and agree to be accountable for all aspects of the work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ information (optional)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eRummeny E. Benign focal liver lesions. In: \u003cem\u003eMagnetic Resonance Imaging of the Liver.\u003c/em\u003e edn.; 2000: 21\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaltenbach T, Engler P, Kratzer W, Oeztuerk S, Seufferlein T, Haenle M, Graeter T. Prevalence of benign focal liver lesions: ultrasound investigation of 45,319 hospital patients. Abdom Radiol. 2016;41:25\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSkouras C, Scheuer M. Non-invasive diagnosis and management of common benign liver tumours. BMJ Case Rep 2014.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuido LP, Garcia-Buitrago M. Benign hepatocellular lesions and neoplasms: a comprehensive review. Diagn Histopathology 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRonot M, Vilgrain V. Imaging of benign hepatocellular lesions: current concepts and recent updates. Clin Res Hepatol Gastroenterol. 2014;38(6):681\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan den Esschert JW, van Gulik TV, Phoa S. Imaging modalities for focal nodular hyperplasia and hepatocellular adenoma. Dig Surg. 2010;27:46\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatrich AN, Polshikov SV. The role of ultrasound in the differential diagnosis of liver tumors. Innovative Med Kuban 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCannella R, Borhani AA, Minervini M, Tsung A, Furlan A. Evaluation of texture analysis for the differential diagnosis of focal nodular hyperplasia from hepatocellular adenoma on contrast-enhanced CT images. Abdom Radiol. 2018;44:1323\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrazioli L, Bondioni M, Haradome H, Motosugi U, Tinti R, Frittoli B, Gambarini S, Donato F, Colagrande S. Hepatocellular adenoma and focal nodular hyperplasia: value of gadoxetic acid-enhanced MR imaging in differential diagnosis. Radiology. 2012;262(2):520\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo Y, Li W, Cai W, Zhang Y, Fang Y, Hong G. Diagnostic value of gadoxetic acid-enhanced MR imaging to distinguish HCA and its subtype from FNH: A systematic review. Int J Med Sci. 2017;14:668\u0026ndash;74.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBieze M, Bennink R, El-Massoudi Y, Phoa S, Verheij J, Beuers U, van Gulik TV. The use of 18F-fluoromethylcholine PET/CT in differentiating focal nodular hyperplasia from hepatocellular adenoma: a prospective study of diagnostic accuracy. Nucl Med Commun. 2012;34:146\u0026ndash;54.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDonato H, Fran\u0026ccedil;a M, Candel\u0026aacute;ria I, Caseiro-Alves F. Liver MRI: From basic protocol to advanced techniques. Eur J Radiol. 2017;93:30\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark SH, Kim B, Kim S, Choi S, Huh J, Kim H, Kim K, Lee SS. Characterizing computed tomography-detected arterial hyperenhancing-only lesions in patients at risk of hepatocellular carcinoma: Can non-contrast magnetic resonance imaging be used for sequential imaging? Korean J Radiol. 2020;21:280\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSinitsyn V, Chueva NA. Differential diagnosis of liver focal formations at diffusion MRI. Experimental Clin Gastroenterol. 2010;7:131\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGupta S, Choh N, Bhatt MG, Wani R, Rasool Z, Rasool SR. Diffusion-weighted imaging: An exciting and problem-solving tool in patients with hepatic metastases. Indian J Radiol Imaging. 2021;31:86\u0026ndash;90.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoshi AA, Kulkarni S, Shah A. Role of MRI in evaluation of spectrum of liver lesions in cirrhotic patients. J Assoc Phys India. 2020;68(4):32\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoussef A, Elkharbotly A, Khater H, Awad MA. State of the art of magnetic resonance imaging (MRI) in evaluation of focal hepatic lesions. Benha Med J 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBioulac-Sage P, Sempoux C, Balabaud C. Hepatocellular adenoma: Classification, variants and clinical relevance. Semin Diagn Pathol. 2017;34(2):112\u0026ndash;25.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNault J, Couchy G, Balabaud C, Morcrette G, Caruso S, Blanc J, Bacq Y, Calderaro J, Paradis V, Ramos J, et al. Molecular classification of hepatocellular adenoma associates with risk factors, bleeding, and malignant transformation. Gastroenterology. 2017;152(4):880\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatabathina V, Menias C, Shanbhogue A, Jagirdar J, Paspulati R, Prasad S. Genetics and imaging of hepatocellular adenomas: 2011 update. Radiographics. 2011;31(6):1529\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParadis V. Hepatocellular adenomas: WHO classification and immunohistochemical workup. Surg Pathol Clin. 2013;6(2):311\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTse J, Felker E, Naini B, Shen L, Shen J, Lu D, Kamaya A, Raman S. Hepatocellular adenomas: Molecular basis and multimodality imaging update. Radiographics 2023, 43(3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen K, Dong Y, Zhang W, Han H, Mao F, Zhang Q, Zheng Z, He W, Wang W. Analysis of contrast-enhanced ultrasound features of hepatocellular adenoma according to different pathological molecular classifications. Clin Hemorheol Microcirc 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZossou V, Gnangnon F, Biaou O, de Vathaire F, Allodji RS, Ezin EC. Radiomics-based classification of tumor and healthy liver on computed tomography images. Cancers 2024, 16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHardie A, Chamberlin J, Boyum JH, Sharbidre K, Petrocelli R, Flemming B, Zahid M, Venkatesh S, Mruthyunjayappa S, Hajdu C et al. Multi-center follow-up study to develop a classification system which differentiates mucinous cystic neoplasm of the liver and benign hepatic cyst using machine learning. \u003cem\u003eAcademic Radiology\u003c/em\u003e 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBae H, Lee HS, Kim S, Han K, Rhee H, Kim D, Kwon H, Hong H, Lim J. Radiomics analysis of contrast-enhanced CT for classification of hepatic focal lesions in colorectal cancer patients: Its limitations compared to radiologists. Eur Radiol. 2021;31:8786\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng R, Shi C, Wang C, Shi N, Qiu T, Chen W, Shi Y, Wang H. Imaging-based staging of hepatic fibrosis in patients with hepatitis B: A dynamic radiomics model based on Gd-EOB-DTPA-enhanced MRI. Biomolecules 2021, 11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOyama A, Hiraoka Y, Obayashi I, Saikawa Y, Furui S, Shiraishi K, Kumagai S, Hayashi T, Kotoku J. Hepatic tumor classification using texture and topology analysis of non-contrast-enhanced three-dimensional T1-weighted MR images with a radiomics approach. Sci Rep 2019, 9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHomayounieh F, Singh R, Nitiwarangkul C, Lades F, Schmidt B, Sedlmair M, Saini S, Kalra M. Semiautomatic segmentation and radiomics for dual-energy CT: A pilot study to differentiate benign and malignant hepatic lesions. AJR Am J Roentgenol 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIchikawa S, Motosugi U, Suzuki T, Shimizu T, Onishi H. Imaging features of hepatic inflammatory pseudotumor: Distinction from colorectal liver metastasis using gadoxetate disodium-enhanced magnetic resonance imaging. Abdom Radiol. 2020;45:2400\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang H, Guo S. Solitary hypovascular hepatic nodules: Comparison and differentiation between peripheral nodular cholangiocarcinoma and atypical liver hemangioma with MRI. Translational Cancer Res. 2023;12:2308\u0026ndash;18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuofa S. Study of diagnostic values of dynamic MRI and diffusion weighted imaging in small hepatic hemangiomas (\u0026le;\u0026thinsp;2 cm). J Clin Radiol 2013.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLebert P, Adens-Fauquembergue M, Azahaf M, Gnemmi V, Behal H, Luciani A, Ernst O. MRI for characterization of benign hepatocellular tumors on hepatobiliary phase: The added value of in-phase imaging and lesion-to-liver visual signal intensity ratio. Eur Radiol 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNi T, Shang X, Wang W, Hu X, Zeng M, Rao S. Different MR features for differentiation of intrahepatic mass-forming cholangiocarcinoma from hepatocellular carcinoma according to tumor size. Br J Radiol 2018, 91.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu H, Chen M, Huang Y, Wang P, Li J, Li Q, Jiang R. Differential diagnosis of intracranial malignant tumors using MRI based on morphological features and signal intensity ratio of lesions. Altern Ther Health Med 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePiao Y, Wang H, Ma L, Yu G, Du G, Ye H. MRI features of primary hepatic neuroendocrine tumor. Chin Med J. 2018;52:125\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeshmukh P, Patil H. Depression prediction model based on ensemble learning classifier. Indian J Sci Technol 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Z, Zhu X, Liu D. Model of gradient boosting random forest prediction. In: \u003cem\u003e2022\u003c/em\u003e; 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVamvakas A, Tsivaka D, Logothetis A, Vassiou K, Tsougos I. Breast cancer classification on multiparametric MRI \u0026ndash; Increased performance of boosting ensemble methods. Technol Cancer Res Treat 2022, 21.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrishnamoorthy S. Interpretable classifier models for decision support using high utility gain patterns. IEEE Access. 2024;12:126088\u0026ndash;107.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElHariri MA, Ali TF, Hussien HI. Apparent diffusion coefficient (ADC) in liver fibrosis: Usefulness of normalized ADC using the spleen as reference organ. Egypt J Radiol Nuclear Med. 2013;44:441\u0026ndash;51.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKamal OA, Ahmed AY, Ali M, Mostafa MA. The diagnostic value of normalized ADC using spleen as a reference organ in assessment of liver fibrosis. QJM: Int J Med 2021, 113.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen X, Chen G, Zhou P, Li H. Association of the liver and spleen signal intensity on MRI with anemia in gynecological cancer. Curr Med Imaging 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIshii S, Hara T, Nanbu T, Suenaga H, Sugawara S, Kuroiwa D, Sekino H, Miyajima M, Kubo H, Oriuchi N, et al. Optimized workflow and imaging protocols for whole-body oncologic PET/MRI. Japanese J Radiol. 2016;34:754\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiuqua C. Efficacy of a multiparametric MRI protocol for BI-RADS categorization of breast lesions. Chin J Med Imaging 2015.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JY, Huo E, Weinstein S, Santos C, Monto A, Corvera C, Yee J, Hope T. Evaluation of an abbreviated screening MRI protocol for patients at risk for hepatocellular carcinoma. Abdom Radiol. 2018;43:1627\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCanellas R, Patel MJ, Agarwal S, Sahani D. Lesion detection performance of an abbreviated gadoxetic acid\u0026ndash;enhanced MRI protocol for colorectal liver metastasis surveillance. Eur Radiol 2019:1\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGranata V, Fusco R, De Muzio F, Cutolo C, Mattace Raso M, Gabelloni M, Avallone A, Ottaiano A, Tatangelo F, Brunese MC et al. Radiomics and machine learning analysis based on magnetic resonance imaging in the assessment of colorectal liver metastases growth pattern. Diagnostics 2022, 12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGiannini V, Rosati S, Defeudis A, Balestra G, Vassallo L, Cappello G, Mazzetti S, De Mattia C, Rizzetto F, Torresin A, et al. Radiomics predicts response of individual HER2-amplified colorectal cancer liver metastases in patients treated with HER2‐targeted therapy. Int J Cancer. 2020;147:3215\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGovardhanam V, Zeghal M, Cheung A. A287 leveraging machine learning to improve the diagnostic accuracy of ultrasound screening for hepatocellular carcinoma. J Can Association Gastroenterol. 2024;7:231\u0026ndash;2.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSailer M, Schiller F, Falk T, Jud A, Lang SA, Ruf J, Mix M. Prediction of the histopathological tumor type of newly diagnosed liver lesions from standard abdominal computer tomography with a machine-learning classifier based on convolutional neural networks. Curr Dir Biomedical Eng. 2021;7:150\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaeem S, Ali A, Qadri S, Mashwani WK, Tairan N, Shah H, Fayaz M, Jamal F, Chesneau C, Anam S. Machine-learning based hybrid-feature analysis for liver cancer classification using fused (MR and CT) images. \u003cem\u003eApplied Sciences\u003c/em\u003e 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrooks JA, Kallenbach M, Radu I, Berzigotti A, Dietrich CF, Kather JN, Luedde T, Seraphin TP. Artificial intelligence for contrast-enhanced ultrasound of the liver: A systematic review. Digestion; 2024.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Focal Nodular Hyperplasia, Hepatocellular Adenoma, Magnetic Resonance Imaging, Radiomics, Machine Learning","lastPublishedDoi":"10.21203/rs.3.rs-6631565/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6631565/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eFocal Nodular Hyperplasia (FNH) and Hepatocellular Adenoma (HCA) often present overlapping imaging features on MRI. Recent radiomics and machine learning (ML) strategies may enhance differentiation accuracy.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis retrospective study involved 197 lesions (77 FNH; 120 HCA, including 87 non-steatotic HCAs) from 98 patients (mean age 33.95\u0026thinsp;\u0026plusmn;\u0026thinsp;9.28 years, 87.7% female). Multiparametric MRI included T2-weighted, arterial-phase, and diffusion-weighted (b-50, b-400, b-800) sequences. Over 100 features, including mass-to-spleen T2 ratios and spleen-referenced diffusion metrics, were extracted. Recursive feature elimination identified the most discriminative variables. Eight ML models were trained and evaluated using accuracy, F1-score, and AUC-ROC on separate training/testing sets.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFNH lesions were significantly larger than HCA (48.95\u0026thinsp;\u0026plusmn;\u0026thinsp;26.64 mm vs. 36.35\u0026thinsp;\u0026plusmn;\u0026thinsp;28.42 mm; p\u0026thinsp;=\u0026thinsp;0.002). The T2-weighted mass-to-spleen signal ratio effectively distinguished FNH from HCA/NSHCA (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). DWI revealed higher signal intensities in FNH at b-50 and b-400 (402\u0026thinsp;\u0026plusmn;\u0026thinsp;275, 222\u0026thinsp;\u0026plusmn;\u0026thinsp;139) compared to HCA (273\u0026thinsp;\u0026plusmn;\u0026thinsp;265, 166\u0026thinsp;\u0026plusmn;\u0026thinsp;166; p\u0026thinsp;=\u0026thinsp;0.002), and consistently greater mean mass-to-liver SIDs for b-50, b-400, and b-800 (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Gradient Boosting Machine (GBM) and Random Forest demonstrated the highest test-set performance (AUC-ROC\u0026thinsp;=\u0026thinsp;0.915), with GBM achieving 86.0% accuracy; a standalone decision tree classifier achieved 87.5% accuracy on training and 81.1% on testing sets. Across models, T2 lesion intensity, arterial phase signal ratio, and spleen-referenced DWI metrics ranked as the top three predictive features.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eIntegrating T2 mass-to-spleen ratios, arterial-phase enhancement, and advanced DWI significantly improves the differentiation of FNH from HCA, including non-steatotic variants. Ensemble ML methods outperformed simpler decision tree classifiers and underscore the importance of biologically aligned feature selection for robust lesion characterization.\u003c/p\u003e","manuscriptTitle":"MRI-Based T2 and Arterial Radiomics to Differentiate Focal Nodular Hyperplasia (FNH) from Hepatocellular Adenoma (HCA): A High-Accuracy Spleen-Referenced Approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-09 10:33:53","doi":"10.21203/rs.3.rs-6631565/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"08dd0886-4849-4195-8d43-296432897bdd","owner":[],"postedDate":"June 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-08T20:53:16+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-09 10:33:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6631565","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6631565","identity":"rs-6631565","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}