Machine Learning Prediction of Carotid Intraplaque Hemorrhage: Fusing CT Radiomics and Clinical Data

Machine Learning Prediction of Carotid Intraplaque Hemorrhage: Fusing CT Radiomics and Clinical Data | 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 Article Machine Learning Prediction of Carotid Intraplaque Hemorrhage: Fusing CT Radiomics and Clinical Data Meilan Zhang, Jing Li, Yan Wang, Yani Zhao, Wanting Xie, Fang Zhang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7592769/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 Objective Develop and validate a machine learning model integrating CT-based radiomics with clinical and imaging data to diagnose carotid intraplaque hemorrhage (IPH), aiding clinical decision-making. Method This retrospective study analyzed 127 carotid plaques from 66 patients undergoing concurrent head/neck CTA and MRI at Ordos Central Hospital (April 2023–September 2024). Based on MRI results, plaques were categorized as IPH-positive (n = 41) or IPH-negative (n = 86). Radiomics features derived from CTA images were combined with clinical baseline characteristics. Three machine learning models were developed: 1) clinical baseline, 2) radiomics, and 3) integrated. Model performance was evaluated using AUC-ROC, calibration curves, and DCA. Results The logistic regression (LR) algorithm demonstrated superior diagnostic performance in both radiomics and integrated models, achieving test set AUCs of 0.804 (95% CI: 0.651–0.958) and 0.824 (95% CI: 0.682–0.965), respectively. The integrated model exhibited enhanced calibration and clinical utility via decision curve analysis. Plaque surface morphology was identified as a significant independent predictor of IPH. Conclusion The machine learning model combining CT-based radiomics features with clinical and imaging characteristics effectively diagnoses carotid artery IPH. This integrated approach provides valuable support for risk stratification and clinical decision-making in carotid plaque management, demonstrating significant potential for clinical application. Health sciences/Biomarkers Biological sciences/Computational biology and bioinformatics Health sciences/Diseases Health sciences/Medical research Health sciences/Risk factors Carotid plaque Radiomics༛Machine learning༛Intraplaque hemorrhage Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction Ischemic stroke is frequently linked to atherosclerotic plaque as a primary contributor.Historically, stroke prevention in patients with carotid plaque relied heavily on assessing the severity of luminal stenosis, as reflected in clinical guidelines and management strategies( 1 , 2 ). However, advances in imaging technology have increasingly demonstrated( 3 ) that evaluating stroke risk based solely on stenosis has significant limitations. While luminal stenosis remains a widely used clinical metric—primarily reflecting mechanical obstruction of blood flow—it fails to fully capture the biological characteristics of plaques or their potential to trigger acute cerebrovascular events.Critically, vulnerable plaques with intraplaque hemorrhage (IPH), thin fibrous caps, large lipid cores, or inflammatory infiltration can trigger acute ischemic stroke, even with mild luminal stenosis ( 4 ). In recent years, the focus of research has shifted toward identifying these high-risk plaque characteristics beyond stenosis, along with developing advanced imaging techniques for their detection( 5 ). Among these features, IPH is considered particularly critical, as it not only accelerates plaque progression but also exacerbates instability by promoting inflammation and neovascularization. Strong evidence supports the association between IPH and both initial stroke occurrence ( 6 , 7 ) and stroke recurrence( 8 ), underscoring its importance in risk stratification and preventive strategies. Studies have demonstrated( 9 ) that multiple imaging modalities can serve as noninvasive tools for assessing plaque vulnerability. Computed tomography angiography (CTA) is rapid acquisition and high spatial resolution make it essential for pretreatment atherosclerotic plaque assessment( 10 ). Beyond accurately quantifying luminal stenosis, CTA also provides detailed insights into plaque morphology. However, its diagnostic utility largely depends on subjective physician interpretation, and it remains limited in detecting IPH.Magnetic resonance imaging (MRI) maintains its position as the benchmark modality for intraplaque hemorrhage (IPH) detection, attributable to superior diagnostic accuracy in sensitivity and specificity( 11 ). High-resolution MRI can distinctly visualize plaque components, particularly IPH, with T1-weighted imaging (T1WI) demonstrating superior detection capability. Nevertheless, MRI has notable drawbacks, including prolonged scan times, high costs, and incompatibility for patients with claustrophobia or metallic implants.Radiomics, an emerging multidisciplinary field, presents new opportunities for precise carotid plaque characterization. By extracting quantitative features imperceptible to the human eye, radiomics enables objective, data-driven analysis for applications such as disease diagnosis, treatment monitoring, and prognosis prediction ( 12 ), thereby enhancing clinical decision-making.Given these advancements, this study focuses on leveraging radiomic features to identify IPH in CTA images, a challenge for conventional diagnostic approaches. While prior research has explored MRI-based radiomics ( 13 )and CT texture analysis for vulnerable plaque detection, studies specifically targeting IPH using CT radiomics remain scarce. This study aims to develop a CT-based carotid plaque imaging model integrated with clinical and imaging data to noninvasively assess intraplaque hemorrhage (IPH) during routine CTA examinations, enabling enhanced clinical decision support. 2. Materials and Methods 2.1 Study population Ethical approval was granted by the Ordos Central Hospital Ethics Committee (2025-078) in compliance with the Declaration of Helsinki.We retrospectively enrolled patients undergoing head/neck CTA and MRI at Ordos Central Hospital (April 2023–September 2024), with informed consent waived per institutional policy.Inclusion criteria: ① CTA-confirmed carotid plaque diagnosis ② Complete clinical data availability ③ ≤1-week interval between CTA and MRI exams. Exclusion criteria: ① Prior carotid revascularization (stenting or endarterectomy) ② Documented carotid occlusion ③ Non-diagnostic imaging quality or incomplete MRI acquisition.The exclusion criteria are shown in Fig.1. Each plaque of each patient enrolled was measured and outlined, and 66 patients were finally included, 57 males and 9 females, with a total number of 127 plaques. Plaques were stratified into IPH+ (n=41) and IPH- (n=86) cohorts based on TOF-MRI criteria: intraplaque signal intensity >150% of adjacent sternocleidomastoid muscle on T1-weighted imaging(8) (Fig.2-3). Patients were randomized 7:3 to training/test sets. Clinical variables (age, gender, smoking history, diabetes, hypertension, dyslipidemia, BMI) and imaging parameters (maximum plaque thickness(14), plaque CT attenuation (15), surface morphology(3), calcification burden(16), stenosis severity(17), plaque volume) were analyzed. Maximum plaque thickness represented the maximal cross-sectional dimension; plaque CT attenuation was measured in the lowest-density region (ROI ≥1 mm²); surface morphology was categorized as ulcerated (contrast entry ≥1 mm), irregular (contrast entry 0.3-0.9 mm), or smooth (no significant penetration); calcification burden included voxels >130 HU; stenosis severity was quantified per NASCET criteria: Stenosis(%) = (normal diameter of the distal segment of the stenosis - minimum residual diameter of the stenotic segment ) / normal diameter of the distal segment of stenosis×100%, graded mild (<30%), moderate (30-69%), or severe (70-99%); plaque volume was automatically calculated after semi-automated segmentation (ITK-SNAP). 2.2 Instruments and methods Contrast-enhanced CT scanning was performed using a GE Revolution 256-slice scanner with patients in supine position (aortic arch to cranial vault). Venous patency was confirmed via antecubital saline test (20 mL), followed by iohexol contrast injection (40-50 mL, 4.0-5.0 mL/s) and saline flush (20 mL). Automated bolus tracking triggered acquisition 2s after descending aorta attenuation reached 245 HU. Scanning parameters: 100 kVp tube voltage, automatic tube current (100-350 mA), 512×512 matrix, 40 mm detector width, 25.0 cm SFOV, 0.625 mm slice thickness/reconstruction interval. Images were reconstructed on AW 4.7 workstation. The MRI scanning was performed using a SIEMENS Vida 3.0T magnetic resonance scanner with a Zongzhi 32-channel head and neck coil, and the patient was assisted to assume a supine position with the head advanced, and was instructed to remain immobile and to minimize swallowing movements. MRI protocols comprised 3D-TOF MRA, axial T1-weighted (T1WI), axial T2 fat-suppressed (T2-FS), and sagittal 3D T1 SPACE (pre-/post-contrast). Gadopentetate dextran (0.1 mmol/kg) was administered intravenously followed by a 120-second delay prior to enhanced acquisition. Standard parameters included: FOV 140mm (3D-TOF) and 160mm (T1WI/T2-FS/T1 SPACE), uniform matrix 256×256, NEX=1 (number of excitations). Sequence-specific parameters were: 3D-TOF (TR/TE=20/4.9 ms, FA=20°, thickness=1.0 mm); T1WI (TR/TE=550/18 ms, thickness=2 mm); T2-FS (TR/TE=3000/93 ms, thickness=2 mm); T1 SPACE±C (TR/TE=900/20 ms, thickness=0.63 mm). All images were post-processed using syngo.via workstation. 2.3 Model construction The collected clinical baseline features were statistically analyzed, and the meaningful features were included to construct the clinical baseline model (CM) using four machine learning models, namely K-nearst neighbors (KNN), Logistic Regression (LR), Multilayer Perceptron (MLP), Support Vector Machine (SVM), and named as KNN-CM, LR-CM, MR-CM, and Support Vector Machine (SVM), respectively. Support Vector Machine (SVM) four machine learning models to construct the Clinical Model (CM), and named KNN-CM, LR-CM, MLP-CM, SVM-CM. each plaque was outlined and saved, plaque outlining and image histological features extraction, screening as in the first part. The screened imaging histology features were used to build an imaging histology model (Radiomics Model, RM) using the four machine learning models mentioned above and named as KNN-RM, LR-RM, MLP-RM, SVM-RM. Clinical and radiomic features were integrated to develop the Combined Model (CombM), with subsequent ROC curve analysis quantifying discriminative ability via AUC, sensitivity, and specificity metrics. 2.4 Statistical analysis Statistical analyses employed SPSS 26.0 and Python 3.9. Categorical data were summarized as frequencies with intergroup comparisons using χ² tests. Continuous variables underwent normality testing, with normally distributed data expressed as mean±SD and compared via independent t-tests; non-normally distributed variables used Mann-Whitney U tests. Statistical significance was defined as two-tailed p<0.05. 3. Results 3.1 General information The study analyzed 127 carotid plaques from 66 patients. As detailed in Table 1, maximum plaque thickness, volume, luminal stenosis degree, and surface morphology differed significantly between groups (P<0.05). The maximum plaque thickness (5.57±6.59 mm) and plaque volume (978.06±806.32 mm 3 ) in the group with intra-plaque hemorrhage were significantly greater than those in the group without intra-plaque hemorrhage (3.39±1.31 mm and 600.41±684.75 mm 3 ); Hemorrhage group demonstrated significantly higher moderate-to-severe stenosis prevalence than non-hemorrhage group (73.18% vs 37.80%). In addition, plaque surface morphology was significantly higher in the proportion of ulcerated plaques in the group with hemorrhage (24.39%) than in the group without hemorrhage (3.66%). While demographic variables (gender, age, BMI) and comorbidities (hypertension, diabetes, hyperlipidemia) showed no statistical significance (P>0.05), Spearman-LASSO feature selection identified maximum plaque thickness, volume, and surface morphology as optimal predictors. These features were incorporated into KNN, LR, MLP, and SVM algorithms to construct the Clinical Model (designated KNN-CM, LR-CM, etc.). The LR-CM demonstrated superior and more stable diagnostic performance across training/testing cohorts (Fig.4). Table 1 General statistical results No intraplaque hemorrhage group Group with intraplaque hemorrhage p-value Age（years,x̄ ±s） a 66.98±8.69 68.83±8.53 0.115 BMI（ ± s ） a 24.80±2.92 24.61±2.69 0.723 Maximum Plaque Thickness（mm， ± s ） a 3.39±1.31 5.57±6.59 <0.001 CT Value（HU， ± s ） a 72.88±178.64 41.83±19.27 0.277 Plaque Volume（mm 3 ， ± s ） a 600.41±684.75 978.06±806.32 0.002 The degree of stenosis, n（%） b <0.001 Mild 51(62.20) 11(26.83) Moderate 18(21.95) 15(36.59) Severe 13(15.85) 15(36.59) Position, n（%） b 0.747 left side 46(56.10) 25(60.98) Right side 36(43.90) 16(39.02) Gender, n（%） b 1.00 Female 10(12.20) 5(12.20) Male 72(87.80) 36(87.80) Smoking, n（%） b 0.517 0 24(29.27) 9(21.95) 1 58(70.73) 32(78.05) Hypertension, n（%） b 0.735 0 26(31.71) 15(36.59) 1 56(68.29) 26(63.41) hyperlipidemian, n（%） b 1.00 0 49(59.76) 24(58.54) 1 33(40.24) 17(41.46) Diabetes, n（%） b 0.735 0 54(65.85) 25(60.98) 1 28(34.15) 16(39.02) Calcification, n（%） b 0.817 0 19(23.17) 8(19.51) 1 63(76.83) 33(80.49) Morphology of the plaque surface, n（%） b 0.001 Smooth 77(93.90) 29(70.73) irregular 2(2.44) 2(4.88) ulcerated 3(3.66) 10(24.39) Note: a Continuous variables are expressed as mean ± standard deviation； b Categorical variables are expressed as frequency and percentage. 3.2 Image histology machine learning model construction and model effectiveness comparison From 1834 initially extracted radiomic features in the training set, stability analysis (ICC) and Mann-Whitney U testing (P<0.05) retained 543 features. Subsequent Spearman correlation eliminated redundant features (134 remaining), with LASSO regression selecting the 10 most predictive features for machine learning model construction (KNN, LR, MLP, SVM). All four radiomics models demonstrated strong diagnostic performance in training and test sets (Fig.5). Test set AUCs ranged 0.683-0.804, with the Logistic Regression Radiomics Model (LR-RM) showing optimal stability and performance: AUC 0.804 (95%CI:0.651-0.958), sensitivity 0.769, specificity 0.792 (Table 2). Support Vector Machine (SVM-RM) exhibited significant performance degradation (training AUC 0.968 vs test 0.683), indicating potential overfitting. KNN model demonstrated suboptimal accuracy and sensitivity.The LR model is a classical linear classification model, and compared to other complex machine learning models (e.g., SVM, MLP), the LR model's structure is relatively simple and has fewer parameters, so the LR model is not prone to overfitting phenomenon when the amount of training data is limited. Table 2 Diagnostic Performance of Radiomics Machine Learning Models in the Training and Testing Sets Accuracy AUC 95% CI Sensitivity Specificity PPV NPV Set LR 0.872 0.902 0.8252 - 0.9789 0.857 0.879 0.774 0.927 train LR 0.784 0.804 0.6514 - 0.9575 0.769 0.792 0.667 0.864 test SVM 0.942 0.968 0.9197 - 1.0000 0.857 0.983 0.960 0.934 train SVM 0.649 0.683 0.4972 - 0.8682 0.769 0.583 0.500 0.824 test KNN 0.814 0.882 0.8145 - 0.9496 0.536 0.948 0.833 0.809 train KNN 0.622 0.670 0.4943 - 0.8454 0.385 0.750 0.455 0.692 test MLP 0.872 0.904 0.8288 - 0.9791 0.714 0.948 0.870 0.873 train MLP 0.730 0.756 0.5936 - 0.9192 0.615 0.792 0.615 0.792 test 3.3 Evaluation of the efficacy of the combined model of clinical and imaging signs and imaging features Eleven features integrating clinical characteristics and radiomic signatures comprised the Joint Model. Baseline features included plaque surface morphology and maximum plaque thickness. Radiomic features involved: wavelet_HHH_firstorder_Kurtosis wavelet_LLL_ngtdm_Strength wavelet_HLL_glszm_GrayLevelNonUniformity lbp_3D_k_gldm_LargeDependenceHighGrayLevelEmphasis lbp_3D_m2_ngtdm_Busyness wavelet_LHH_ngtdm_Strength square_glszm_ZonePercentage lbp_3D_k_ngtdm_Busyness lbp_3D_k_ngtdm_Coarseness Plaque surface morphology demonstrated the highest feature weight (Fig 6), particularly ulcerative changes, indicating significant diagnostic value for IPH detection. In addition, multiple wavelet transform-based texture features occupy an important position in the model, reflecting the contribution of the internal heterogeneity of the plaque to the diagnosis of IPH. The joint model (LR-CombM) demonstrated optimal diagnostic performance with a test set AUC of 0.824 (95% CI: 0.682-0.965), significantly outperforming other models (Fig.7). Calibration analysis revealed minimal deviations between predicted and observed IPH probabilities (Fig.8a), while decision curve analysis confirmed substantial net clinical benefit across probability thresholds, indicating robust clinical utility (Fig 8b). 4. Discussion This study developed and validated a CT-based radiomics model incorporating clinical and CTA features for non-invasive detection of carotid intraplaque hemorrhage (IPH). Multiple machine learning algorithms demonstrated robust diagnostic performance, with logistic regression (LR) achieving optimal results in both radiomics (AUC = 0.804) and integrated models (AUC = 0.824). Plaque surface morphology emerged as a significant predictor in clinical baseline features. The findings establish CT radiomics with clinical integration as an effective approach for IPH identification, enabling enhanced risk stratification and clinical decision-making for carotid plaque patients. 4.1 Clinical significance of carotid plaque identification of IPH Intraplaque hemorrhage (IPH) is a recognized biomarker of carotid plaque vulnerability, strongly associated with ischemic stroke incidence and recurrence. Mechanistically, IPH increases cerebrovascular event risk by promoting intraplaque neovascular rupture and inflammation, accelerating plaque instability( 4 ). Multiple studies confirm IPH as a key predictor of stroke outcomes( 18 , 19 ), with Che et al. reporting a striking hazard ratio (HR = 6.64; 95%CI 2.84–15.54; P < 0.001) for ipsilateral stroke prediction( 8 ).Therefore, identifying IPH not only helps to predict stroke risk, but also provides clinicians with an important basis for treatment decisions. In addition, treatment needs to be altered for patients with IPH, and some studies have suggested that vitamin K antagonists (VKA) and antiplatelet agents need to be used with caution when IPH is present ( 20 , 21 ). In conclusion, the diagnosis of IPH not only helps to identify high-risk patients, but also provides an important basis for individualized treatment to avoid unnecessary drug use and potential risks. 4.2 Clinical imaging features associated with carotid plaque IPH Maximum plaque thickness, volume, stenosis degree, and ulceration were identified as significant IPH risk factors, aligning with Larson et al.'s findings of increased stenosis severity in IPH + versus IPH- plaques ( 22 ). This correlation likely reflects hemodynamically mediated alterations in intraplaque neovascularization. This is also similar to our results, although the degree of plaque stenosis was ultimately removed during LASS screening, it can be found that the percentage of plaques with severe luminal stenosis was also higher in the group without intra-plaque hemorrhage than in the group with intra-plaque hemorrhage (15.85% vs. 36.59%). Plaque volume was significantly elevated in IPH+ (978.06 ± 806.32 mm³) versus IPH- plaques (600.41 ± 684.75 mm³). This likely reflects greater overall plaque burden - encompassing lipid cores, fibrous caps, calcified components, and hemorrhagic areas - as measured by plaque volume and thickness ( 23 ). Plaque surface morphology (especially ulcerative changes) had the highest feature weighting in the combined model, indicating its importance in the diagnosis of IPH. Ulcerated plaques with broken endothelial layer on the surface and exposed plaque cores rich in lipids and inflammatory cells are prone to rupture and hemorrhage( 24 ), which is consistent with the findings of Frink RJ et al ( 25 ). 4.3 Ability of CTA Imaging Histology Machine Learning Models in Diagnosing IPH CTA is the predominant imaging modality for carotid plaque assessment.CT-based IPH diagnosis yields inconsistent results across studies. Notably, some investigations demonstrate a 25 Hounsfield Unit (HU) threshold provides high sensitivity and specificity for discriminating IPH from lipid necrotic cores ( 26 ); others have taken a different view, arguing that the diagnostic efficacy of CT in distinguishing IPH cannot be demonstrated due to the overlap in the distribution of CT values( 27 , 28 ).MRI remains the gold standard for IPH diagnosis( 3 , 29 ), demonstrating high sensitivity and specificity.However, MRI is not as widely used as CTA due to its long scanning time, high price and many contraindications. Radiomics has garnered substantial attention in biomedical research as an emerging analytical approach.With the advantages of non-invasiveness, high precision, and multimodal integration, it plays an important role in disease diagnosis, treatment prediction, disease monitoring and prognosis assessment, etc. Moreover, it can be quantitatively analyzed, which can more accurately reflect the subtle changes of lesions, and help to realize personalized medicine. For example, the wavelet_HHH_firstorder_Kurtosis feature reflects the higher-order statistical properties of intra-plaque signals and can identify signal inhomogeneity caused by intra-plaque hemorrhage; wavelet_LLL_ngtdm_Strength and wavelet_HLL_glszm_GrayLevelNonUniformity, and other wavelet transform-based features, on the other hand, are able to capture texture differences in different regions within the plaque, which may be related to local signal variations caused by IPH. Therefore, this study diagnosed IPH based on CT imaging histology and obtained good results (rad model AUC reached 0.670–0.804 in the test set). Among the multiple machine learning models selected for this study, the LR model has the best diagnostic efficacy, and the DAC and fitting curves also show that it has good stability. This may be due to the fact that it uses a logistic function to map the results of linear regression to the range of 0 to 1, converting the output of linear regression to probability ( 30 ), which can directly reflect the likelihood of the sample belonging to a certain category, and has a high degree of interpretability. In contrast, the outputs of models such as KNN and SVM are category labels or distance measures, which are less interpretable. There have been studies( 31 ) to diagnose vulnerable plaques by constructing models using LR, and good diagnostic efficacy has been obtained. Zhang et al. ( 13 ) developed an MRI-based radiomics LR model (AUC = 0.85) identifying high-risk carotid plaques, while Yan et al. ( 32 ) achieved comparable accuracy (AUC = 0.82) using multimodal ultrasound and clinical risk factors. These studies consistently demonstrate LR's efficacy in vulnerable plaque diagnosis. 4.4 Limitations of this study This study has limitations: First, the modest cohort size may introduce bias despite tenfold cross-validation mitigating overfitting, necessitating prospective validation in larger populations for model stability. Second, while MRI served as the IPH diagnostic reference (current clinical standard ( 28 )), pathological confirmation remains the gold standard, with verification available for only select cases. Third, the single-center design without external validation constrains generalizability, warranting multicenter trials with independent cohorts. Future investigations should explore CT/US multimodal integration to leverage complementary imaging advantages for enhanced diagnostic precision. 5. Conclusion This study developed a CT-based radiomics LR model for IPH detection, demonstrating clinically applicable diagnostic efficacy to support decision-making and refine risk stratification in carotid plaque management. In the future, the performance of the model can be further optimized through the application of joint deep learning technology and biomarker combination to promote its wide application in clinical practice. Declarations Funding ： This study was supported by the Ordos Science and Technology Plan Project (No. 2019501). Competing interests The authors declare no competing interests. Author Contribution Meilan Zhang and Jing Li: contributed equally to this study, being responsible for the study design, data collection and analysis, as well as manuscript drafting and revisions. Yan Wang and Yani Zhao and Wanting Xie: assisted with data processing and experimental analysis. Hai Du: provided technical support and suggestions on the research direction and reviewed the manuscript. Hai Du and Fang Zhang: as the corresponding authors, supervised the overall study, provided critical guidance, and finalized the manuscript revisions. Meilan Zhang and Jing Li are co-first authors. All authors read and approved the final manuscript. Data Availability The datasets generated and analysed during the current study are not publicly available due patient privacy but are available from the corresponding author on reasonable request. References Song P, Fang Z, Wang H, et al. Global and regional prevalence, burden, and risk factors for carotid atherosclerosis: A systematic review, meta-analysis, and modelling study. Lancet Glob Health . 2020;8(5):e721-e729. doi:10.1016/S2214-109X(20)30117-0 Aboyans V, Ricco JB. The “ten commandments” of 2017 ESC guidelines on the diagnosis and treatment of peripheral arterial diseases. Eur Heart J . 2018;39(9):722. doi:10.1093/eurheartj/ehy045 Saba L, Saam T, Jäger HR, et al. Imaging biomarkers of vulnerable carotid plaques for stroke risk prediction and their potential clinical implications. Lancet Neurol . 2019;18(6):559-572. doi:10.1016/S1474-4422(19)30035-3 Saam T, Hetterich H, Hoffmann V, et al. Meta-analysis and systematic review of the predictive value of carotid plaque hemorrhage on cerebrovascular events by magnetic resonance imaging. J Am Coll Cardiol . 2013;62(12):1081-1091. doi:10.1016/j.jacc.2013.06.015 Kramer CM, Treiman GS. Vulnerable plaque in carotid arteries without “significant” stenosis: Unmasking the hidden links to stroke. J Am Coll Cardiol . 2020;76(19):2223-2225. doi:10.1016/j.jacc.2020.09.531 Larson AS, Nasr DM, Rizvi A, et al. Embolic stroke of undetermined source: The association with carotid intraplaque hemorrhage. JACC Cardiovasc Imaging . 2021;14(2):506-508. doi:10.1016/j.jcmg.2020.08.007 Qi H, Sun J, Qiao H, et al. Carotid intraplaque hemorrhage imaging with quantitative vessel wall T1 mapping: Technical development and initial experience. Radiology . 2018;287(1):276-284. doi:10.1148/radiol.2017170526 Che F, Mi D, Wang A, et al. Extracranial carotid plaque hemorrhage predicts ipsilateral stroke recurrence in patients with carotid atherosclerosis - a study based on high-resolution vessel wall imaging MRI. BMC Neurol . 2022;22(1):237. doi:10.1186/s12883-022-02758-3 Saba L, Yuan C, Hatsukami TS, et al. Carotid artery wall imaging: perspective and guidelines from the ASNR vessel wall imaging study group and expert consensus recommendations of the american society of neuroradiology. Am J Neuroradiol . 2018;39(2):E9-E31. doi:10.3174/ajnr.A5488 Rajiah P. Updates in vascular computed tomography. Radiol Clin North Am . 2020;58(4):671-691. doi:10.1016/j.rcl.2020.02.011 Zhou T, Jia S, Wang X, et al. Diagnostic performance of MRI for detecting intraplaque hemorrhage in the carotid arteries: A meta-analysis. Eur Radiol . 2019;29(10):5129-5138. doi:10.1007/s00330-019-06053-7 Zhang L, Wang Y, Peng Z, et al. The progress of multimodal imaging combination and subregion based radiomics research of cancers. Int J Biol Sci . 2022;18(8):3458-3469. doi:10.7150/ijbs.71046 Zhang R, Zhang Q, Ji A, et al. Identification of high-risk carotid plaque with MRI-based radiomics and machine learning. Eur Radiol . 2021;31(5):3116-3126. doi:10.1007/s00330-020-07361-z Zhao X, Hippe DS, Li R, et al. Prevalence and characteristics of carotid artery high-risk atherosclerotic plaques in chinese patients with cerebrovascular symptoms: a chinese atherosclerosis risk evaluation II study. J Am Heart Assoc . 2017;6(8):e005831. doi:10.1161/JAHA.117.005831 Gupta A, Gialdini G, Lerario MP, et al. Magnetic resonance angiography detection of abnormal carotid artery plaque in patients with cryptogenic stroke. J Am Heart Assoc . 2015;4(6):e002012. doi:10.1161/JAHA.115.002012 Saba L, Caddeo G, Sanfilippo R, Montisci R, Mallarini G. CT and Ultrasound in the Study of Ulcerated Carotid Plaque Compared with Surgical Results: Potentialities and Advantages of Multidetector Row CT Angiography. Am J Neuroradiol . 2007;28:1061-1066. doi:10.3174/ajnr.A0486 North american symptomatic carotid endarterectomy trial. Methods, patient characteristics, and progress. Stroke . 1991;22(6):711-720. doi:10.1161/01.STR.22.6.711 Dam-Nolen DV van, Truijman M, Kolk AG van der, et al. Carotid Plaque Characteristics Predict Recurrent Ischemic Stroke and TIA: The PARISK (Plaque At RISK) Study. JACC Cardiovasc Imaging . 2022;15 10:1715-1726. doi:10.1016/j.jcmg.2022.04.003 Lin R, Chen S, Liu G, Xue Y, Zhao X. Association between carotid atherosclerotic plaque calcification and intraplaque hemorrhage: A magnetic resonance imaging study. Arterioscler Thromb Vasc Biol . 2017;37(6):1228-1233. doi:10.1161/ATVBAHA.116.308360 Tran KA, Kondrashova O, Bradley A, Williams ED, Pearson JV, Waddell N. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med . 2021;13(1):152. doi:10.1186/s13073-021-00968-x Mujaj B, Bos D, Muka T, et al. Antithrombotic treatment is associated with intraplaque haemorrhage in the atherosclerotic carotid artery: A cross-sectional analysis of the rotterdam study. Eur Heart J . 2018;39(36):3369-3376. doi:10.1093/eurheartj/ehy433 Larson AS, Brinjikji W, Savastano LE, Huston Iii J, Benson JC. Carotid intraplaque hemorrhage is associated with cardiovascular risk factors. Cerebrovasc Dis Basel Switz . 2020;49(4):355-360. doi:10.1159/000508733 Zhu C, Tian X, Degnan AJ, et al. Clinical significance of intraplaque hemorrhage in low- and high-grade basilar artery stenosis on high-resolution MRI. Am J Neuroradiol . 2018;39(7):1286-1292. doi:10.3174/ajnr.A5676 Chen X, Meschia JF, Huang J, et al. Intraplaque hemorrhage and plaque ulceration are more likely in patients with symptomatic mild-to-moderate carotid artery stenosis than in symptomatic and asymptomatic high-grade stenosis: a retrospective cohort study. Ann Vasc Surg . 2025;112:82-92. doi:10.1016/j.avsg.2024.12.004 Frink RJ. Inflammation, chronic ulcerated plaques, and unstable coronary syndromes. Cardiol Rev . 1998;6(5):302-311. doi:10.1097/00045415-199809000-00012 Saba L, Francone M, Bassareo PP, et al. CT Attenuation Analysis of Carotid Intraplaque Hemorrhage. Am J Neuroradiol . 2018;39:131-137. doi:10.3174/ajnr.A5461 U-King-Im JM, Fox AJ, Aviv RI, et al. Characterization of carotid plaque hemorrhage: A CT angiography and MR intraplaque hemorrhage study. Stroke . 2010;41(8):1623-1629. doi:10.1161/STROKEAHA.110.579474 Zhang S, Gao L, Kang B, Yu X, Zhang R, Wang X. Radiomics assessment of carotid intraplaque hemorrhage: detecting the vulnerable patients. Insights Imaging . 2022;13(1):200. doi:10.1186/s13244-022-01324-2 Mark IT, Nasr DM, Huston J, et al. Embolic stroke of undetermined source and carotid intraplaque hemorrhage on MRI : A systemic review and meta-analysis. Clin Neuroradiol . 2021;31(2):307-313. doi:10.1007/s00062-020-00921-2 Devi S, Gaikwad SR, R H. Prediction and detection of cervical malignancy using machine learning models. Asian Pac J Cancer Prev APJCP . 2023;24(4):1419-1433. doi:10.31557/APJCP.2023.24.4.1419 Curcio N, Rosato A, Mazzaccaro D, Nano G, Conti M, Matrone G. 3D patient-specific modeling and structural finite element analysis of atherosclerotic carotid artery based on computed tomography angiography. Sci Rep . 2023;13(1):19911. doi:10.1038/s41598-023-46949-5 Yan L, Ye X, Fu L, Hou W, Lin S, Su H. Construction of vulnerable plaque prediction model based on multimodal vascular ultrasound parameters and clinical risk factors. Sci Rep . 2024;14(1):24255. doi:10.1038/s41598-024-75375-4 Additional Declarations No competing interests reported. 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-7592769","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":522382705,"identity":"5fe4f592-fed1-4a2f-adde-4025bf787648","order_by":0,"name":"Meilan Zhang","email":"","orcid":"","institution":"Baotou Medical College","correspondingAuthor":false,"prefix":"","firstName":"Meilan","middleName":"","lastName":"Zhang","suffix":""},{"id":522382706,"identity":"9c562ba9-9367-40c4-954e-9e12201ee398","order_by":1,"name":"Jing Li","email":"","orcid":"","institution":"Baotou Medical College","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Li","suffix":""},{"id":522382707,"identity":"0ee713e1-65a2-4cb9-8f32-d1b31cd6ede0","order_by":2,"name":"Yan Wang","email":"","orcid":"","institution":"Ordos Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yan","middleName":"","lastName":"Wang","suffix":""},{"id":522382708,"identity":"5b80060b-e282-45a1-8e51-e060137aae2a","order_by":3,"name":"Yani Zhao","email":"","orcid":"","institution":"Baotou Medical College","correspondingAuthor":false,"prefix":"","firstName":"Yani","middleName":"","lastName":"Zhao","suffix":""},{"id":522382709,"identity":"4fc68687-fa1d-4977-8ae1-306c7a7acd33","order_by":4,"name":"Wanting Xie","email":"","orcid":"","institution":"Baotou Medical College","correspondingAuthor":false,"prefix":"","firstName":"Wanting","middleName":"","lastName":"Xie","suffix":""},{"id":522382710,"identity":"574bbd7f-d99d-414d-b4b8-4127af55e90d","order_by":5,"name":"Fang Zhang","email":"","orcid":"","institution":"Ordos Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Fang","middleName":"","lastName":"Zhang","suffix":""},{"id":522382711,"identity":"f5d650c7-4c39-4f89-9ed0-34bb6c9516df","order_by":6,"name":"Hai Du","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAtElEQVRIiWNgGAWjYFACHgaGBAYGOTb25gNE6mADaUlgMObjOZZAghagNYnzJHIUiNNhLt978MPDH4fT2xhyGBh+VGwjrMWyjS9ZIiHhcG4bw9kDjD1nbhPWYnCMxwCihbEvgZmxjTgtxj+AWtLZmHkMiNZiBrIlgY2NeC05ZhYJaemGbTxsCQeJ88vhM8Y3f9hYy8vPf3zwwY8KIrRAQTOYPEC0eiCoI0XxKBgFo2AUjDQAAESxOdmc7Z+7AAAAAElFTkSuQmCC","orcid":"","institution":"Ordos Central Hospital","correspondingAuthor":true,"prefix":"","firstName":"Hai","middleName":"","lastName":"Du","suffix":""}],"badges":[],"createdAt":"2025-09-11 14:08:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7592769/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7592769/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92680411,"identity":"c2034bd4-eaf4-442b-bc26-ff47888cd7cf","added_by":"auto","created_at":"2025-10-03 01:05:11","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2434032,"visible":true,"origin":"","legend":"","description":"","filename":"manuscriptX221.docx","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/88ea153dfcd9af646181558f.docx"},{"id":92682341,"identity":"11b7e20c-a955-47bb-b09e-3228dcb13b8b","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8436,"visible":true,"origin":"","legend":"","description":"","filename":"d82ed6e9655c49ccb9adea10a60aa49d.json","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/ca91cb97576b7034db40c0f3.json"},{"id":92680387,"identity":"54fcf84e-c2e8-4f82-8b9f-bbc802dedf61","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110360,"visible":true,"origin":"","legend":"","description":"","filename":"d82ed6e9655c49ccb9adea10a60aa49d1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/c899dd3602971d0e0e988e72.xml"},{"id":92680396,"identity":"c6fd3c5b-a704-4699-bb2a-814bb4019d7c","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114151,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/1bdf2e195e0f443b52c0388a.png"},{"id":92682344,"identity":"45b2cef2-5877-410e-9efb-cab1f0436cc5","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":849713,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/5904768328fb5d5d363303e5.png"},{"id":92682343,"identity":"1d32ae07-b762-4bb6-a225-76ef91be655d","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":742547,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/1832421d2e5a560f6ffa6d04.png"},{"id":92680405,"identity":"c22710bc-272c-40ad-8bf4-91101cd5a264","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139055,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/fe53e5173eda28b43af514f8.png"},{"id":92680413,"identity":"4c323ca8-f78c-4eb4-8aff-9bdb05653a4c","added_by":"auto","created_at":"2025-10-03 01:05:11","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139326,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/1b69d69c0680b7bc836ae69d.png"},{"id":92680394,"identity":"f8ff735b-4aee-4a5b-8268-3e416c40bb02","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":97957,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/e4fd9fe10c33d768f9f96bad.png"},{"id":92680400,"identity":"86a5ee96-cc67-4ae3-ac81-2b7eed33e49a","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":142741,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/d3fbf058fbcf218b4d5908a1.png"},{"id":92680402,"identity":"1aea1434-d4c3-4688-b833-6d8ac4633b05","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":109784,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/7553732333b750b631dfa47b.png"},{"id":92680407,"identity":"1e68d9cb-9333-4695-9a3d-15018e3c7d42","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"wmf","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":380,"visible":true,"origin":"","legend":"","description":"","filename":"image4.wmf","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/d9035a67d30ea3f377c32bdc.wmf"},{"id":92680392,"identity":"ad133a79-75b5-4b9f-b3ce-1cd6ff86be7b","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29026,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/636260db94954135bec332a6.png"},{"id":92680406,"identity":"9047b84e-c317-4588-bfa7-9019113756db","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":192599,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/8624d62639368bd447b4b078.png"},{"id":92682686,"identity":"16e0b8fb-52a8-49d5-9fa8-64127bebadd7","added_by":"auto","created_at":"2025-10-03 01:21:10","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":158231,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/c7b5e6360dbb1ccb29e15bff.png"},{"id":92680404,"identity":"9b5b66d6-9b1a-42de-973c-1165554572e6","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28328,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/974c3748cd7bdad4a4026dc4.png"},{"id":92682348,"identity":"96ced2c4-112e-40d2-a390-b3764bb5706d","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":27321,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/559a2238a3be6837ecab70cd.png"},{"id":92680410,"identity":"02c8e40b-3625-414e-b189-ad8848db1fb5","added_by":"auto","created_at":"2025-10-03 01:05:11","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":22785,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/1157f65cf0dd1bf6b1707c69.png"},{"id":92682349,"identity":"6dbb32c7-4d84-453c-bd3e-f184de3742cb","added_by":"auto","created_at":"2025-10-03 01:13:11","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":27813,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/6db9d3c79a2cba4c5f9a2ff8.png"},{"id":92680408,"identity":"12dc567b-83a3-4919-bac0-7e506f562112","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29673,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/5f2d4b0e760face7b8479603.png"},{"id":92680398,"identity":"e4615b89-846a-4634-85aa-b1b1163ba6fc","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":240,"visible":true,"origin":"","legend":"","description":"","filename":"Onlineimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/f82aab1c3b1cf7e87a9152ee.png"},{"id":92682346,"identity":"9245db75-157a-442f-976d-2f14345bf357","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"xml","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":109059,"visible":true,"origin":"","legend":"","description":"","filename":"d82ed6e9655c49ccb9adea10a60aa49d1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/5982938570e821509884bb07.xml"},{"id":92680414,"identity":"09f91ca1-85d2-41bd-bd01-f4ae7c5224ef","added_by":"auto","created_at":"2025-10-03 01:05:11","extension":"html","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":121518,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/ecf592416aa6ab21ed04cd88.html"},{"id":92680384,"identity":"7267e1cd-6470-4410-990a-e0421a3663fa","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":114151,"visible":true,"origin":"","legend":"\u003cp\u003eStandard flow chart for the exclusion of patients with intraplaque hemorrhage\u003c/p\u003e\n\u003cp\u003eNote: n represents the number of patients and N represents the number of plaques\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/97638af7dc451436d497f2d2.png"},{"id":92680385,"identity":"3b85b1dc-02ac-4bb9-ae79-16c02631364d","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":968790,"visible":true,"origin":"","legend":"\u003cp\u003eExample of a patient with no bleeding within the plaque\u003c/p\u003e\n\u003cp\u003eNote; The figure shows a 72-year-old male patient with carotid plaque, head and neck CTA reconstruction image (Figure a), showing a mixed plaque at the beginning of the left internal carotid artery visible at the point indicated by the arrow, with mild luminal narrowing, and smooth surface morphology of the plaque; the plaque at the beginning of the left internal carotid artery was seen in isotropic T1 signal without IPH on the T1WI image (Figure b), T2-FS image (Figure c), and T1 SPACE (Figure d). Enhanced scan (Figure e) showed an intact fiber cap.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/95de317956345f88648dc7b6.png"},{"id":92680388,"identity":"a151564b-092e-4286-8522-d520ec554df2","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":833459,"visible":true,"origin":"","legend":"\u003cp\u003eExample of a patient with intraplaque hemorrhage\u003c/p\u003e\n\u003cp\u003eNote: The figure shows an 80-year-old male patient with carotid plaque, head and neck CTA reconstruction image (Figure a), showing a mixed plaque at the beginning of the left internal carotid artery visible at the point indicated by the arrow, with severe lumen stenosis, and irregular morphology on the plaque surface; short T1 signal was seen within the plaque at the beginning of the left internal carotid artery on the T1WI image (Figure b), the T2-FS image (Figure c), and the T1 SPACE (Figure d), and the enhanced Scan (Figure e) did not show significant enhancement, proving that there was IPH.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/b36505d1e56b49a4e8329546.png"},{"id":92682342,"identity":"490d4a66-f6d1-41eb-951b-67bb4c5956c5","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":139055,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of the clinical model in the training and test sets\u003c/p\u003e\n\u003cp\u003eNote: Figure a shows the training set and figure b shows the test set.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/800ebfd3c86d15458c7dc2c9.png"},{"id":92680390,"identity":"08c2168f-f29a-4a7a-9ed9-6716d02a1211","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":139326,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of the four imaging histology machine learning models in the training and test sets\u003c/p\u003e\n\u003cp\u003eNote: Figure a is the training set and figure b is the test set.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/fcd08ffb5ffc878b740318f1.png"},{"id":92680397,"identity":"2b66d115-8f26-4786-82a9-cee7c3e80f11","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":97957,"visible":true,"origin":"","legend":"\u003cp\u003eCharacteristic weights map for constructing the joint model\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/e1a775434db5899cf916276a.png"},{"id":92680409,"identity":"c506eb57-1dc4-4c75-a710-c042f75e2072","added_by":"auto","created_at":"2025-10-03 01:05:10","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":142741,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of the joint model model in the training and test sets\u003c/p\u003e\n\u003cp\u003eNote: Figure a shows the training set and figure b shows the test set.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/e6d6b6b9a7a888a4a66adbaa.png"},{"id":92682345,"identity":"2f537d34-3cd5-4385-8422-5ea5cf074221","added_by":"auto","created_at":"2025-10-03 01:13:10","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":111874,"visible":true,"origin":"","legend":"\u003cp\u003eLR-CombM model calibration curve (a) and decision analysis curve (b)\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/5a36029d7af8e01984071951.png"},{"id":106723826,"identity":"585bcc56-f4cf-4fe2-b063-42c20d4b9b7a","added_by":"auto","created_at":"2026-04-12 18:16:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3444592,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7592769/v1/5d0ca24f-c1a7-41e5-81f2-2953c873256c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Prediction of Carotid Intraplaque Hemorrhage: Fusing CT Radiomics and Clinical Data","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIschemic stroke is frequently linked to atherosclerotic plaque as a primary contributor.Historically, stroke prevention in patients with carotid plaque relied heavily on assessing the severity of luminal stenosis, as reflected in clinical guidelines and management strategies(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). However, advances in imaging technology have increasingly demonstrated(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) that evaluating stroke risk based solely on stenosis has significant limitations. While luminal stenosis remains a widely used clinical metric\u0026mdash;primarily reflecting mechanical obstruction of blood flow\u0026mdash;it fails to fully capture the biological characteristics of plaques or their potential to trigger acute cerebrovascular events.Critically, vulnerable plaques with intraplaque hemorrhage (IPH), thin fibrous caps, large lipid cores, or inflammatory infiltration can trigger acute ischemic stroke, even with mild luminal stenosis (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). In recent years, the focus of research has shifted toward identifying these high-risk plaque characteristics beyond stenosis, along with developing advanced imaging techniques for their detection(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). Among these features, IPH is considered particularly critical, as it not only accelerates plaque progression but also exacerbates instability by promoting inflammation and neovascularization. Strong evidence supports the association between IPH and both initial stroke occurrence (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e) and stroke recurrence(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e), underscoring its importance in risk stratification and preventive strategies.\u003c/p\u003e\u003cp\u003eStudies have demonstrated(\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e) that multiple imaging modalities can serve as noninvasive tools for assessing plaque vulnerability. Computed tomography angiography (CTA) is rapid acquisition and high spatial resolution make it essential for pretreatment atherosclerotic plaque assessment(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e). Beyond accurately quantifying luminal stenosis, CTA also provides detailed insights into plaque morphology. However, its diagnostic utility largely depends on subjective physician interpretation, and it remains limited in detecting IPH.Magnetic resonance imaging (MRI) maintains its position as the benchmark modality for intraplaque hemorrhage (IPH) detection, attributable to superior diagnostic accuracy in sensitivity and specificity(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). High-resolution MRI can distinctly visualize plaque components, particularly IPH, with T1-weighted imaging (T1WI) demonstrating superior detection capability. Nevertheless, MRI has notable drawbacks, including prolonged scan times, high costs, and incompatibility for patients with claustrophobia or metallic implants.Radiomics, an emerging multidisciplinary field, presents new opportunities for precise carotid plaque characterization. By extracting quantitative features imperceptible to the human eye, radiomics enables objective, data-driven analysis for applications such as disease diagnosis, treatment monitoring, and prognosis prediction (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e), thereby enhancing clinical decision-making.Given these advancements, this study focuses on leveraging radiomic features to identify IPH in CTA images, a challenge for conventional diagnostic approaches. While prior research has explored MRI-based radiomics (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e)and CT texture analysis for vulnerable plaque detection, studies specifically targeting IPH using CT radiomics remain scarce.\u003c/p\u003e\u003cp\u003eThis study aims to develop a CT-based carotid plaque imaging model integrated with clinical and imaging data to noninvasively assess intraplaque hemorrhage (IPH) during routine CTA examinations, enabling enhanced clinical decision support.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003e2.1 Study population\u003c/p\u003e\n\u003cp\u003eEthical approval was granted by the Ordos Central Hospital Ethics Committee (2025-078) in compliance with the Declaration of Helsinki.We retrospectively enrolled patients undergoing head/neck CTA and MRI at Ordos Central Hospital (April 2023\u0026ndash;September 2024), with informed consent waived per institutional policy.Inclusion criteria:\u0026nbsp;①\u0026nbsp;CTA-confirmed carotid plaque diagnosis\u0026nbsp;②\u0026nbsp;Complete clinical data availability\u0026nbsp;③\u0026nbsp;\u0026le;1-week interval between CTA and MRI exams. Exclusion criteria:\u0026nbsp;①\u0026nbsp;Prior carotid revascularization (stenting or endarterectomy)\u0026nbsp;②\u0026nbsp;Documented carotid occlusion\u0026nbsp;③\u0026nbsp;Non-diagnostic imaging quality or incomplete MRI acquisition.The exclusion criteria are shown in Fig.1. Each plaque of each patient enrolled was measured and outlined, and 66 patients were finally included, 57 males and 9 females, with a total number of 127 plaques. Plaques were stratified into IPH+ (n=41) and IPH- (n=86) cohorts based on TOF-MRI criteria: intraplaque signal intensity \u0026gt;150% of adjacent sternocleidomastoid muscle on T1-weighted imaging(8) (Fig.2-3). Patients were randomized 7:3 to training/test sets.\u003c/p\u003e\n\u003cp\u003eClinical variables (age, gender, smoking history, diabetes, hypertension, dyslipidemia, BMI) and imaging parameters (maximum plaque thickness(14), plaque CT attenuation (15), surface morphology(3), calcification burden(16), stenosis severity(17), plaque volume) were analyzed. Maximum plaque thickness represented the maximal cross-sectional dimension; plaque CT attenuation was measured in the lowest-density region (ROI \u0026ge;1 mm\u0026sup2;); surface morphology was categorized as ulcerated (contrast entry \u0026ge;1 mm), irregular (contrast entry 0.3-0.9 mm), or smooth (no significant penetration); calcification burden included voxels \u0026gt;130 HU; stenosis severity was quantified per NASCET criteria: Stenosis(%) = (normal diameter of the distal segment of the stenosis - minimum residual diameter of the stenotic segment ) / normal diameter of the distal segment of stenosis\u0026times;100%, graded mild (\u0026lt;30%), moderate (30-69%), or severe (70-99%); plaque volume was automatically calculated after semi-automated segmentation (ITK-SNAP).\u003c/p\u003e\n\u003cp\u003e2.2 Instruments and methods\u003c/p\u003e\n\u003cp\u003eContrast-enhanced CT scanning was performed using a GE Revolution 256-slice scanner with patients in supine position (aortic arch to cranial vault). Venous patency was confirmed via antecubital saline test (20 mL), followed by iohexol contrast injection (40-50 mL, 4.0-5.0 mL/s) and saline flush (20 mL). Automated bolus tracking triggered acquisition 2s after descending aorta attenuation reached 245 HU. Scanning parameters: 100 kVp tube voltage, automatic tube current (100-350 mA), 512\u0026times;512 matrix, 40 mm detector width, 25.0 cm SFOV, 0.625 mm slice thickness/reconstruction interval. Images were reconstructed on AW 4.7 workstation.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;The MRI scanning was performed using a SIEMENS Vida 3.0T magnetic resonance scanner with a Zongzhi 32-channel head and neck coil, and the patient was assisted to assume a supine position with the head advanced, and was instructed to remain immobile and to minimize swallowing movements. MRI protocols comprised 3D-TOF MRA, axial T1-weighted (T1WI), axial T2 fat-suppressed (T2-FS), and sagittal 3D T1 SPACE (pre-/post-contrast). Gadopentetate dextran (0.1 mmol/kg) was administered intravenously followed by a 120-second delay prior to enhanced acquisition. Standard parameters included: FOV 140mm (3D-TOF) and 160mm (T1WI/T2-FS/T1 SPACE), uniform matrix 256\u0026times;256, NEX=1 (number of excitations). Sequence-specific parameters were: 3D-TOF (TR/TE=20/4.9 ms, FA=20\u0026deg;, thickness=1.0 mm); T1WI (TR/TE=550/18 ms, thickness=2 mm); T2-FS (TR/TE=3000/93 ms, thickness=2 mm); T1 SPACE\u0026plusmn;C (TR/TE=900/20 ms, thickness=0.63 mm). All images were post-processed using syngo.via workstation.\u003c/p\u003e\n\u003cp\u003e2.3 Model construction\u003c/p\u003e\n\u003cp\u003eThe collected clinical baseline features were statistically analyzed, and the meaningful features were included to construct the clinical baseline model (CM) using four machine learning models, namely K-nearst neighbors (KNN), Logistic Regression (LR), Multilayer Perceptron (MLP), Support Vector Machine (SVM), and named as KNN-CM, LR-CM, MR-CM, and Support Vector Machine (SVM), respectively. Support Vector Machine (SVM) four machine learning models to construct the Clinical Model (CM), and named KNN-CM, LR-CM, MLP-CM, SVM-CM. each plaque was outlined and saved, plaque outlining and image histological features extraction, screening as in the first part. The screened imaging histology features were used to build an imaging histology model (Radiomics Model, RM) using the four machine learning models mentioned above and named as KNN-RM, LR-RM, MLP-RM, SVM-RM.\u003c/p\u003e\n\u003cp\u003eClinical and radiomic features were integrated to develop the Combined Model (CombM), with subsequent ROC curve analysis quantifying discriminative ability via AUC, sensitivity, and specificity metrics.\u003c/p\u003e\n\u003cp\u003e2.4 Statistical analysis\u003c/p\u003e\n\u003cp\u003eStatistical analyses employed SPSS 26.0 and Python 3.9. Categorical data were summarized as frequencies with intergroup comparisons using \u0026chi;\u0026sup2; tests. Continuous variables underwent normality testing, with normally distributed data expressed as mean\u0026plusmn;SD and compared via independent t-tests; non-normally distributed variables used Mann-Whitney U tests. Statistical significance was defined as two-tailed p\u0026lt;0.05.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e3.1 General information\u003c/p\u003e\n\u003cp\u003eThe study analyzed 127 carotid plaques from 66 patients. As detailed in Table 1, maximum plaque thickness, volume, luminal stenosis degree, and surface morphology differed significantly between groups (P\u0026lt;0.05). The maximum plaque thickness (5.57\u0026plusmn;6.59 mm) and plaque volume (978.06\u0026plusmn;806.32 mm\u003csup\u003e3\u003c/sup\u003e) in the group with intra-plaque hemorrhage were significantly greater than those in the group without intra-plaque hemorrhage (3.39\u0026plusmn;1.31 mm and 600.41\u0026plusmn;684.75 mm\u003csup\u003e3\u003c/sup\u003e); Hemorrhage group demonstrated significantly higher moderate-to-severe stenosis prevalence than non-hemorrhage group (73.18% vs 37.80%). In addition, plaque surface morphology was significantly higher in the proportion of ulcerated plaques in the group with hemorrhage (24.39%) than in the group without hemorrhage (3.66%). While demographic variables (gender, age, BMI) and comorbidities (hypertension, diabetes, hyperlipidemia) showed no statistical significance (P\u0026gt;0.05), Spearman-LASSO feature selection identified maximum plaque thickness, volume, and surface morphology as optimal predictors. These features were incorporated into KNN, LR, MLP, and SVM algorithms to construct the Clinical Model (designated KNN-CM, LR-CM, etc.). The LR-CM demonstrated superior and more stable diagnostic performance across training/testing cohorts (Fig.4).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 1 \u0026nbsp;General statistical results\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"534\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003eNo intraplaque hemorrhage group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003eGroup with intraplaque hemorrhage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eAge（years,x̄ \u0026plusmn;s）\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e66.98\u0026plusmn;8.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e68.83\u0026plusmn;8.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.115\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eBMI（\u003cstrong\u003e\u003cimg width=\"16\" height=\"15\" src=\"data:image/wmf;base64,R0lGODlhGAAWAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAQABAAOAA0AhAAAAAAAAB0AAAAAHQAdMgAcSAAzWh0zWjQdHTMzWzVbbjNbgEgcAFozAFszM0gzM1tISEhuf1luf2xGHW5GM25bNW5bSG5uWX9/XW6AbmaIiIBbM4iIZgECAwECAwECAwUvIAAEZGmSYqqubOu+AJJSQsGSA+DAlUDAIoYNCMgBHwIgBMBwTVa2BMsI2AgOrBAAOw==\" alt=\"image\"\u003e\u003c/strong\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e24.80\u0026plusmn;2.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e24.61\u0026plusmn;2.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.723\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eMaximum Plaque Thickness（mm，\u003cstrong\u003e\u003cimg width=\"16\" height=\"15\" src=\"data:image/wmf;base64,R0lGODlhGAAWAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAQABAAOAA0AhAAAAAAAAB0AAAAAHQAdMgAcSAAzWh0zWjQdHTMzWzVbbjNbgEgcAFozAFszM0gzM1tISEhuf1luf2xGHW5GM25bNW5bSG5uWX9/XW6AbmaIiIBbM4iIZgECAwECAwECAwUvIAAEZGmSYqqubOu+AJJSQsGSA+DAlUDAIoYNCMgBHwIgBMBwTVa2BMsI2AgOrBAAOw==\" alt=\"image\"\u003e\u003c/strong\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e3.39\u0026plusmn;1.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e5.57\u0026plusmn;6.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eCT Value（HU，\u003cstrong\u003e\u003cimg width=\"16\" height=\"15\" src=\"data:image/wmf;base64,R0lGODlhGAAWAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAQABAAOAA0AhAAAAAAAAB0AAAAAHQAdMgAcSAAzWh0zWjQdHTMzWzVbbjNbgEgcAFozAFszM0gzM1tISEhuf1luf2xGHW5GM25bNW5bSG5uWX9/XW6AbmaIiIBbM4iIZgECAwECAwECAwUvIAAEZGmSYqqubOu+AJJSQsGSA+DAlUDAIoYNCMgBHwIgBMBwTVa2BMsI2AgOrBAAOw==\" alt=\"image\"\u003e\u003c/strong\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e72.88\u0026plusmn;178.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e41.83\u0026plusmn;19.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.277\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003ePlaque Volume（mm\u003csup\u003e3\u003c/sup\u003e，\u003cstrong\u003e\u003cimg width=\"16\" height=\"15\" src=\"data:image/wmf;base64,R0lGODlhGAAWAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAQABAAOAA0AhAAAAAAAAB0AAAAAHQAdMgAcSAAzWh0zWjQdHTMzWzVbbjNbgEgcAFozAFszM0gzM1tISEhuf1luf2xGHW5GM25bNW5bSG5uWX9/XW6AbmaIiIBbM4iIZgECAwECAwECAwUvIAAEZGmSYqqubOu+AJJSQsGSA+DAlUDAIoYNCMgBHwIgBMBwTVa2BMsI2AgOrBAAOw==\" alt=\"image\"\u003e\u003c/strong\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e600.41\u0026plusmn;684.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e978.06\u0026plusmn;806.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eThe degree of stenosis, n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eMild\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e51(62.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e11(26.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eModerate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e18(21.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e15(36.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eSevere\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e13(15.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e15(36.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003ePosition,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.747\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eleft side\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e46(56.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e25(60.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eRight side\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e36(43.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e16(39.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eGender,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e10(12.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e5(12.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e72(87.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e36(87.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eSmoking,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.517\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e24(29.27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e9(21.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e58(70.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e32(78.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eHypertension,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.735\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e26(31.71)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e15(36.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e56(68.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e26(63.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003ehyperlipidemian,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e49(59.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e24(58.54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e33(40.24)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e17(41.46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eDiabetes,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.735\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e54(65.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e25(60.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e28(34.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e16(39.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eCalcification,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.817\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e19(23.17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e8(19.51)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e63(76.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e33(80.49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eMorphology of the plaque surface,\u0026nbsp;n（%）\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eSmooth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e77(93.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e29(70.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eirregular\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e2(2.44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e2(4.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.0824%;\"\u003e\n \u003cp\u003eulcerated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 23.0337%;\"\u003e\n \u003cp\u003e3(3.66)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.7191%;\"\u003e\n \u003cp\u003e10(24.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.1648%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eNote:\u003csup\u003ea\u003c/sup\u003eContinuous variables are expressed as mean\u0026nbsp;\u0026plusmn;\u0026nbsp;standard deviation；\u003csup\u003eb\u003c/sup\u003eCategorical variables are expressed as frequency and percentage.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;3.2 Image histology machine learning model construction and model effectiveness comparison\u003c/p\u003e\n\u003cp\u003eFrom 1834 initially extracted radiomic features in the training set, stability analysis (ICC) and Mann-Whitney U testing (P\u0026lt;0.05) retained 543 features. Subsequent Spearman correlation eliminated redundant features (134 remaining), with LASSO regression selecting the 10 most predictive features for machine learning model construction (KNN, LR, MLP, SVM).\u003c/p\u003e\n\u003cp\u003eAll four radiomics models demonstrated strong diagnostic performance in training and test sets (Fig.5). Test set AUCs ranged 0.683-0.804, with the Logistic Regression Radiomics Model (LR-RM) showing optimal stability and performance: AUC 0.804 (95%CI:0.651-0.958), sensitivity 0.769, specificity 0.792 (Table 2). Support Vector Machine (SVM-RM) exhibited significant performance degradation (training AUC 0.968 vs test 0.683), indicating potential overfitting. KNN model demonstrated suboptimal accuracy and sensitivity.The LR model is a classical linear classification model, and compared to other complex machine learning models (e.g., SVM, MLP), the LR model\u0026apos;s structure is relatively simple and has fewer parameters, so the LR model is not prone to overfitting phenomenon when the amount of training data is limited.\u003c/p\u003e\n\u003cp\u003eTable 2 Diagnostic Performance of Radiomics Machine Learning Models in the Training and Testing Sets\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\u0026nbsp;\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"97%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003eSet\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.902\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.8252 - 0.9789\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.879\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.774\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.927\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.804\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.6514 - 0.9575\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.792\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.864\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.942\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.968\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.9197 - 1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.983\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.960\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.934\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.649\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.683\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.4972 - 0.8682\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.583\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.824\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.814\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.882\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.8145 - 0.9496\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.536\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.948\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.809\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.622\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.670\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.4943 - 0.8454\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.385\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.455\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.692\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eMLP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.904\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.8288 - 0.9791\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.948\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.870\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.873\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 6.31579%;\"\u003e\n \u003cp\u003eMLP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003e0.730\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.756\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e0.5936 - 0.9192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7895%;\"\u003e\n \u003cp\u003e0.615\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e0.792\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.615\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.42105%;\"\u003e\n \u003cp\u003e0.792\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003etest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e3.3 Evaluation of the efficacy of the combined model of clinical and imaging signs and imaging features\u003c/p\u003e\n\u003cp\u003eEleven features integrating clinical characteristics and radiomic signatures comprised the Joint Model. Baseline features included plaque surface morphology and maximum plaque thickness. Radiomic features involved:\u003c/p\u003e\n\u003cp\u003ewavelet_HHH_firstorder_Kurtosis\u003c/p\u003e\n\u003cp\u003ewavelet_LLL_ngtdm_Strength\u003c/p\u003e\n\u003cp\u003ewavelet_HLL_glszm_GrayLevelNonUniformity\u003c/p\u003e\n\u003cp\u003elbp_3D_k_gldm_LargeDependenceHighGrayLevelEmphasis\u003c/p\u003e\n\u003cp\u003elbp_3D_m2_ngtdm_Busyness\u003c/p\u003e\n\u003cp\u003ewavelet_LHH_ngtdm_Strength\u003c/p\u003e\n\u003cp\u003esquare_glszm_ZonePercentage\u003c/p\u003e\n\u003cp\u003elbp_3D_k_ngtdm_Busyness\u003c/p\u003e\n\u003cp\u003elbp_3D_k_ngtdm_Coarseness\u003c/p\u003e\n\u003cp\u003ePlaque surface morphology demonstrated the highest feature weight (Fig 6), particularly ulcerative changes, indicating significant diagnostic value for IPH detection. In addition, multiple wavelet transform-based texture features occupy an important position in the model, reflecting the contribution of the internal heterogeneity of the plaque to the diagnosis of IPH. The joint model (LR-CombM) demonstrated optimal diagnostic performance with a test set AUC of 0.824 (95% CI: 0.682-0.965), significantly outperforming other models (Fig.7). Calibration analysis revealed minimal deviations between predicted and observed IPH probabilities (Fig.8a), while decision curve analysis confirmed substantial net clinical benefit across probability thresholds, indicating robust clinical utility (Fig 8b).\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study developed and validated a CT-based radiomics model incorporating clinical and CTA features for non-invasive detection of carotid intraplaque hemorrhage (IPH). Multiple machine learning algorithms demonstrated robust diagnostic performance, with logistic regression (LR) achieving optimal results in both radiomics (AUC\u0026thinsp;=\u0026thinsp;0.804) and integrated models (AUC\u0026thinsp;=\u0026thinsp;0.824). Plaque surface morphology emerged as a significant predictor in clinical baseline features. The findings establish CT radiomics with clinical integration as an effective approach for IPH identification, enabling enhanced risk stratification and clinical decision-making for carotid plaque patients.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Clinical significance of carotid plaque identification of IPH\u003c/h2\u003e\u003cp\u003eIntraplaque hemorrhage (IPH) is a recognized biomarker of carotid plaque vulnerability, strongly associated with ischemic stroke incidence and recurrence. Mechanistically, IPH increases cerebrovascular event risk by promoting intraplaque neovascular rupture and inflammation, accelerating plaque instability(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Multiple studies confirm IPH as a key predictor of stroke outcomes(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e), with Che et al. reporting a striking hazard ratio (HR\u0026thinsp;=\u0026thinsp;6.64; 95%CI 2.84\u0026ndash;15.54; P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) for ipsilateral stroke prediction(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e).Therefore, identifying IPH not only helps to predict stroke risk, but also provides clinicians with an important basis for treatment decisions. In addition, treatment needs to be altered for patients with IPH, and some studies have suggested that vitamin K antagonists (VKA) and antiplatelet agents need to be used with caution when IPH is present (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). In conclusion, the diagnosis of IPH not only helps to identify high-risk patients, but also provides an important basis for individualized treatment to avoid unnecessary drug use and potential risks.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Clinical imaging features associated with carotid plaque IPH\u003c/h2\u003e\u003cp\u003eMaximum plaque thickness, volume, stenosis degree, and ulceration were identified as significant IPH risk factors, aligning with Larson et al.'s findings of increased stenosis severity in IPH\u0026thinsp;+\u0026thinsp;versus IPH- plaques (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). This correlation likely reflects hemodynamically mediated alterations in intraplaque neovascularization. This is also similar to our results, although the degree of plaque stenosis was ultimately removed during LASS screening, it can be found that the percentage of plaques with severe luminal stenosis was also higher in the group without intra-plaque hemorrhage than in the group with intra-plaque hemorrhage (15.85% vs. 36.59%). Plaque volume was significantly elevated in IPH+ (978.06\u0026thinsp;\u0026plusmn;\u0026thinsp;806.32 mm\u0026sup3;) versus IPH- plaques (600.41\u0026thinsp;\u0026plusmn;\u0026thinsp;684.75 mm\u0026sup3;). This likely reflects greater overall plaque burden - encompassing lipid cores, fibrous caps, calcified components, and hemorrhagic areas - as measured by plaque volume and thickness (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). Plaque surface morphology (especially ulcerative changes) had the highest feature weighting in the combined model, indicating its importance in the diagnosis of IPH. Ulcerated plaques with broken endothelial layer on the surface and exposed plaque cores rich in lipids and inflammatory cells are prone to rupture and hemorrhage(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e), which is consistent with the findings of Frink RJ et al (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Ability of CTA Imaging Histology Machine Learning Models in Diagnosing IPH\u003c/h2\u003e\u003cp\u003eCTA is the predominant imaging modality for carotid plaque assessment.CT-based IPH diagnosis yields inconsistent results across studies. Notably, some investigations demonstrate a 25 Hounsfield Unit (HU) threshold provides high sensitivity and specificity for discriminating IPH from lipid necrotic cores (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e); others have taken a different view, arguing that the diagnostic efficacy of CT in distinguishing IPH cannot be demonstrated due to the overlap in the distribution of CT values(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e).MRI remains the gold standard for IPH diagnosis(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e), demonstrating high sensitivity and specificity.However, MRI is not as widely used as CTA due to its long scanning time, high price and many contraindications. Radiomics has garnered substantial attention in biomedical research as an emerging analytical approach.With the advantages of non-invasiveness, high precision, and multimodal integration, it plays an important role in disease diagnosis, treatment prediction, disease monitoring and prognosis assessment, etc. Moreover, it can be quantitatively analyzed, which can more accurately reflect the subtle changes of lesions, and help to realize personalized medicine. For example, the wavelet_HHH_firstorder_Kurtosis feature reflects the higher-order statistical properties of intra-plaque signals and can identify signal inhomogeneity caused by intra-plaque hemorrhage; wavelet_LLL_ngtdm_Strength and wavelet_HLL_glszm_GrayLevelNonUniformity, and other wavelet transform-based features, on the other hand, are able to capture texture differences in different regions within the plaque, which may be related to local signal variations caused by IPH. Therefore, this study diagnosed IPH based on CT imaging histology and obtained good results (rad model AUC reached 0.670\u0026ndash;0.804 in the test set).\u003c/p\u003e\u003cp\u003eAmong the multiple machine learning models selected for this study, the LR model has the best diagnostic efficacy, and the DAC and fitting curves also show that it has good stability. This may be due to the fact that it uses a logistic function to map the results of linear regression to the range of 0 to 1, converting the output of linear regression to probability (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e), which can directly reflect the likelihood of the sample belonging to a certain category, and has a high degree of interpretability. In contrast, the outputs of models such as KNN and SVM are category labels or distance measures, which are less interpretable. There have been studies(\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e) to diagnose vulnerable plaques by constructing models using LR, and good diagnostic efficacy has been obtained. Zhang et al. (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e) developed an MRI-based radiomics LR model (AUC\u0026thinsp;=\u0026thinsp;0.85) identifying high-risk carotid plaques, while Yan et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e) achieved comparable accuracy (AUC\u0026thinsp;=\u0026thinsp;0.82) using multimodal ultrasound and clinical risk factors. These studies consistently demonstrate LR's efficacy in vulnerable plaque diagnosis.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e4.4 Limitations of this study\u003c/h2\u003e\u003cp\u003eThis study has limitations: First, the modest cohort size may introduce bias despite tenfold cross-validation mitigating overfitting, necessitating prospective validation in larger populations for model stability. Second, while MRI served as the IPH diagnostic reference (current clinical standard (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e)), pathological confirmation remains the gold standard, with verification available for only select cases. Third, the single-center design without external validation constrains generalizability, warranting multicenter trials with independent cohorts. Future investigations should explore CT/US multimodal integration to leverage complementary imaging advantages for enhanced diagnostic precision.\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003e This study developed a CT-based radiomics LR model for IPH detection, demonstrating clinically applicable diagnostic efficacy to support decision-making and refine risk stratification in carotid plaque management. In the future, the performance of the model can be further optimized through the application of joint deep learning technology and biomarker combination to promote its wide application in clinical practice.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThis study was supported by the Ordos Science and Technology Plan Project (No. 2019501).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eMeilan Zhang and Jing Li: contributed equally to this study, being responsible for the study design, data collection and analysis, as well as manuscript drafting and revisions. Yan Wang and Yani Zhao and Wanting Xie: assisted with data processing and experimental analysis. Hai Du: provided technical support and suggestions on the research direction and reviewed the manuscript. Hai Du and Fang Zhang: as the corresponding authors, supervised the overall study, provided critical guidance, and finalized the manuscript revisions. Meilan Zhang and Jing Li are co-first authors. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets generated and analysed during the current study are not publicly available due patient privacy but are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSong P, Fang Z, Wang H, et al. Global and regional prevalence, burden, and risk factors for carotid atherosclerosis: A systematic review, meta-analysis, and modelling study. \u003cem\u003eLancet Glob Health\u003c/em\u003e. 2020;8(5):e721-e729. doi:10.1016/S2214-109X(20)30117-0\u003c/li\u003e\n\u003cli\u003eAboyans V, Ricco JB. The \u0026ldquo;ten commandments\u0026rdquo; of 2017 ESC guidelines on the diagnosis and treatment of peripheral arterial diseases. \u003cem\u003eEur Heart J\u003c/em\u003e. 2018;39(9):722. doi:10.1093/eurheartj/ehy045\u003c/li\u003e\n\u003cli\u003eSaba L, Saam T, J\u0026auml;ger HR, et al. Imaging biomarkers of vulnerable carotid plaques for stroke risk prediction and their potential clinical implications. \u003cem\u003eLancet Neurol\u003c/em\u003e. 2019;18(6):559-572. doi:10.1016/S1474-4422(19)30035-3\u003c/li\u003e\n\u003cli\u003eSaam T, Hetterich H, Hoffmann V, et al. Meta-analysis and systematic review of the predictive value of carotid plaque hemorrhage on cerebrovascular events by magnetic resonance imaging. \u003cem\u003eJ Am Coll Cardiol\u003c/em\u003e. 2013;62(12):1081-1091. doi:10.1016/j.jacc.2013.06.015\u003c/li\u003e\n\u003cli\u003eKramer CM, Treiman GS. Vulnerable plaque in carotid arteries without \u0026ldquo;significant\u0026rdquo; stenosis: Unmasking the hidden links to stroke. \u003cem\u003eJ Am Coll Cardiol\u003c/em\u003e. 2020;76(19):2223-2225. doi:10.1016/j.jacc.2020.09.531\u003c/li\u003e\n\u003cli\u003eLarson AS, Nasr DM, Rizvi A, et al. Embolic stroke of undetermined source: The association with carotid intraplaque hemorrhage. \u003cem\u003eJACC Cardiovasc Imaging\u003c/em\u003e. 2021;14(2):506-508. doi:10.1016/j.jcmg.2020.08.007\u003c/li\u003e\n\u003cli\u003eQi H, Sun J, Qiao H, et al. Carotid intraplaque hemorrhage imaging with quantitative vessel wall T1 mapping: Technical development and initial experience. \u003cem\u003eRadiology\u003c/em\u003e. 2018;287(1):276-284. doi:10.1148/radiol.2017170526\u003c/li\u003e\n\u003cli\u003eChe F, Mi D, Wang A, et al. Extracranial carotid plaque hemorrhage predicts ipsilateral stroke recurrence in patients with carotid atherosclerosis - a study based on high-resolution vessel wall imaging MRI. \u003cem\u003eBMC Neurol\u003c/em\u003e. 2022;22(1):237. doi:10.1186/s12883-022-02758-3\u003c/li\u003e\n\u003cli\u003eSaba L, Yuan C, Hatsukami TS, et al. Carotid artery wall imaging: perspective and guidelines from the ASNR vessel wall imaging study group and expert consensus recommendations of the american society of neuroradiology. \u003cem\u003eAm J Neuroradiol\u003c/em\u003e. 2018;39(2):E9-E31. doi:10.3174/ajnr.A5488\u003c/li\u003e\n\u003cli\u003eRajiah P. Updates in vascular computed tomography. \u003cem\u003eRadiol Clin North Am\u003c/em\u003e. 2020;58(4):671-691. doi:10.1016/j.rcl.2020.02.011\u003c/li\u003e\n\u003cli\u003eZhou T, Jia S, Wang X, et al. Diagnostic performance of MRI for detecting intraplaque hemorrhage in the carotid arteries: A meta-analysis. \u003cem\u003eEur Radiol\u003c/em\u003e. 2019;29(10):5129-5138. doi:10.1007/s00330-019-06053-7\u003c/li\u003e\n\u003cli\u003eZhang L, Wang Y, Peng Z, et al. The progress of multimodal imaging combination and subregion based radiomics research of cancers. \u003cem\u003eInt J Biol Sci\u003c/em\u003e. 2022;18(8):3458-3469. doi:10.7150/ijbs.71046\u003c/li\u003e\n\u003cli\u003eZhang R, Zhang Q, Ji A, et al. Identification of high-risk carotid plaque with MRI-based radiomics and machine learning. \u003cem\u003eEur Radiol\u003c/em\u003e. 2021;31(5):3116-3126. doi:10.1007/s00330-020-07361-z\u003c/li\u003e\n\u003cli\u003eZhao X, Hippe DS, Li R, et al. Prevalence and characteristics of carotid artery high-risk atherosclerotic plaques in chinese patients with cerebrovascular symptoms: a chinese atherosclerosis risk evaluation II study. \u003cem\u003eJ Am Heart Assoc\u003c/em\u003e. 2017;6(8):e005831. doi:10.1161/JAHA.117.005831\u003c/li\u003e\n\u003cli\u003eGupta A, Gialdini G, Lerario MP, et al. Magnetic resonance angiography detection of abnormal carotid artery plaque in patients with cryptogenic stroke. \u003cem\u003eJ Am Heart Assoc\u003c/em\u003e. 2015;4(6):e002012. doi:10.1161/JAHA.115.002012\u003c/li\u003e\n\u003cli\u003eSaba L, Caddeo G, Sanfilippo R, Montisci R, Mallarini G. CT and Ultrasound in the Study of Ulcerated Carotid Plaque Compared with Surgical Results: Potentialities and Advantages of Multidetector Row CT Angiography. \u003cem\u003eAm J Neuroradiol\u003c/em\u003e. 2007;28:1061-1066. doi:10.3174/ajnr.A0486\u003c/li\u003e\n\u003cli\u003eNorth american symptomatic carotid endarterectomy trial. Methods, patient characteristics, and progress. \u003cem\u003eStroke\u003c/em\u003e. 1991;22(6):711-720. doi:10.1161/01.STR.22.6.711\u003c/li\u003e\n\u003cli\u003eDam-Nolen DV van, Truijman M, Kolk AG van der, et al. Carotid Plaque Characteristics Predict Recurrent Ischemic Stroke and TIA: The PARISK (Plaque At RISK) Study. \u003cem\u003eJACC Cardiovasc Imaging\u003c/em\u003e. 2022;15 10:1715-1726. doi:10.1016/j.jcmg.2022.04.003\u003c/li\u003e\n\u003cli\u003eLin R, Chen S, Liu G, Xue Y, Zhao X. Association between carotid atherosclerotic plaque calcification and intraplaque hemorrhage: A magnetic resonance imaging study. \u003cem\u003eArterioscler Thromb Vasc Biol\u003c/em\u003e. 2017;37(6):1228-1233. doi:10.1161/ATVBAHA.116.308360\u003c/li\u003e\n\u003cli\u003eTran KA, Kondrashova O, Bradley A, Williams ED, Pearson JV, Waddell N. Deep learning in cancer diagnosis, prognosis and treatment selection. \u003cem\u003eGenome Med\u003c/em\u003e. 2021;13(1):152. doi:10.1186/s13073-021-00968-x\u003c/li\u003e\n\u003cli\u003eMujaj B, Bos D, Muka T, et al. Antithrombotic treatment is associated with intraplaque haemorrhage in the atherosclerotic carotid artery: A cross-sectional analysis of the rotterdam study. \u003cem\u003eEur Heart J\u003c/em\u003e. 2018;39(36):3369-3376. doi:10.1093/eurheartj/ehy433\u003c/li\u003e\n\u003cli\u003eLarson AS, Brinjikji W, Savastano LE, Huston Iii J, Benson JC. Carotid intraplaque hemorrhage is associated with cardiovascular risk factors. \u003cem\u003eCerebrovasc Dis Basel Switz\u003c/em\u003e. 2020;49(4):355-360. doi:10.1159/000508733\u003c/li\u003e\n\u003cli\u003eZhu C, Tian X, Degnan AJ, et al. Clinical significance of intraplaque hemorrhage in low- and high-grade basilar artery stenosis on high-resolution MRI. \u003cem\u003eAm J Neuroradiol\u003c/em\u003e. 2018;39(7):1286-1292. doi:10.3174/ajnr.A5676\u003c/li\u003e\n\u003cli\u003eChen X, Meschia JF, Huang J, et al. Intraplaque hemorrhage and plaque ulceration are more likely in patients with symptomatic mild-to-moderate carotid artery stenosis than in symptomatic and asymptomatic high-grade stenosis: a retrospective cohort study. \u003cem\u003eAnn Vasc Surg\u003c/em\u003e. 2025;112:82-92. doi:10.1016/j.avsg.2024.12.004\u003c/li\u003e\n\u003cli\u003eFrink RJ. Inflammation, chronic ulcerated plaques, and unstable coronary syndromes. \u003cem\u003eCardiol Rev\u003c/em\u003e. 1998;6(5):302-311. doi:10.1097/00045415-199809000-00012\u003c/li\u003e\n\u003cli\u003eSaba L, Francone M, Bassareo PP, et al. CT Attenuation Analysis of Carotid Intraplaque Hemorrhage. \u003cem\u003eAm J Neuroradiol\u003c/em\u003e. 2018;39:131-137. doi:10.3174/ajnr.A5461\u003c/li\u003e\n\u003cli\u003eU-King-Im JM, Fox AJ, Aviv RI, et al. Characterization of carotid plaque hemorrhage: A CT angiography and MR intraplaque hemorrhage study. \u003cem\u003eStroke\u003c/em\u003e. 2010;41(8):1623-1629. doi:10.1161/STROKEAHA.110.579474\u003c/li\u003e\n\u003cli\u003eZhang S, Gao L, Kang B, Yu X, Zhang R, Wang X. Radiomics assessment of carotid intraplaque hemorrhage: detecting the vulnerable patients. \u003cem\u003eInsights Imaging\u003c/em\u003e. 2022;13(1):200. doi:10.1186/s13244-022-01324-2\u003c/li\u003e\n\u003cli\u003eMark IT, Nasr DM, Huston J, et al. Embolic stroke of undetermined source and carotid intraplaque hemorrhage on MRI : A systemic review and meta-analysis. \u003cem\u003eClin Neuroradiol\u003c/em\u003e. 2021;31(2):307-313. doi:10.1007/s00062-020-00921-2\u003c/li\u003e\n\u003cli\u003eDevi S, Gaikwad SR, R H. Prediction and detection of cervical malignancy using machine learning models. \u003cem\u003eAsian Pac J Cancer Prev APJCP\u003c/em\u003e. 2023;24(4):1419-1433. doi:10.31557/APJCP.2023.24.4.1419\u003c/li\u003e\n\u003cli\u003eCurcio N, Rosato A, Mazzaccaro D, Nano G, Conti M, Matrone G. 3D patient-specific modeling and structural finite element analysis of atherosclerotic carotid artery based on computed tomography angiography. \u003cem\u003eSci Rep\u003c/em\u003e. 2023;13(1):19911. doi:10.1038/s41598-023-46949-5\u003c/li\u003e\n\u003cli\u003eYan L, Ye X, Fu L, Hou W, Lin S, Su H. Construction of vulnerable plaque prediction model based on multimodal vascular ultrasound parameters and clinical risk factors. \u003cem\u003eSci Rep\u003c/em\u003e. 2024;14(1):24255. doi:10.1038/s41598-024-75375-4\u003c/li\u003e\n\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":"Carotid plaque, Radiomics༛Machine learning༛Intraplaque hemorrhage","lastPublishedDoi":"10.21203/rs.3.rs-7592769/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7592769/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eObjective\u003c/b\u003e Develop and validate a machine learning model integrating CT-based radiomics with clinical and imaging data to diagnose carotid intraplaque hemorrhage (IPH), aiding clinical decision-making.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethod\u003c/b\u003e This retrospective study analyzed 127 carotid plaques from 66 patients undergoing concurrent head/neck CTA and MRI at Ordos Central Hospital (April 2023\u0026ndash;September 2024). Based on MRI results, plaques were categorized as IPH-positive (n\u0026thinsp;=\u0026thinsp;41) or IPH-negative (n\u0026thinsp;=\u0026thinsp;86). Radiomics features derived from CTA images were combined with clinical baseline characteristics. Three machine learning models were developed: 1) clinical baseline, 2) radiomics, and 3) integrated. Model performance was evaluated using AUC-ROC, calibration curves, and DCA.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e The logistic regression (LR) algorithm demonstrated superior diagnostic performance in both radiomics and integrated models, achieving test set AUCs of 0.804 (95% CI: 0.651\u0026ndash;0.958) and 0.824 (95% CI: 0.682\u0026ndash;0.965), respectively. The integrated model exhibited enhanced calibration and clinical utility via decision curve analysis. Plaque surface morphology was identified as a significant independent predictor of IPH.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion\u003c/b\u003e The machine learning model combining CT-based radiomics features with clinical and imaging characteristics effectively diagnoses carotid artery IPH. This integrated approach provides valuable support for risk stratification and clinical decision-making in carotid plaque management, demonstrating significant potential for clinical application.\u003c/p\u003e","manuscriptTitle":"Machine Learning Prediction of Carotid Intraplaque Hemorrhage: Fusing CT Radiomics and Clinical Data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-03 01:05:05","doi":"10.21203/rs.3.rs-7592769/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":"c5a3ccee-02d4-4570-9a29-7ab9ebca705d","owner":[],"postedDate":"October 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":55522390,"name":"Health sciences/Biomarkers"},{"id":55522391,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":55522392,"name":"Health sciences/Diseases"},{"id":55522393,"name":"Health sciences/Medical research"},{"id":55522394,"name":"Health sciences/Risk factors"}],"tags":[],"updatedAt":"2026-04-06T07:26:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-03 01:05:05","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7592769","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7592769","identity":"rs-7592769","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}