Synthetic MRI, dynamic contrast-enhanced MRI combined with diffusion-weighted imaging for identifying molecular subtypes of breast cancer using machine learning models

Synthetic MRI, dynamic contrast-enhanced MRI combined with diffusion-weighted imaging for identifying molecular subtypes of breast cancer using machine learning models | 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 Synthetic MRI, dynamic contrast-enhanced MRI combined with diffusion-weighted imaging for identifying molecular subtypes of breast cancer using machine learning models Mengying Xu, Yali Gao, Pan Zhang, Chunhua Li, Jian Li, Zihan Hong, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5608203/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 : To determine whether quantitative parameters from synthetic magnetic resonance imaging (SyMRI), dynamic contrast-enhanced MRI (DCE-MRI), and diffusion-weighted imaging (DWI) can effectively differentiate between molecular subtypes of breast cancer using various machine learning models. Materials and Methods : This retrospective study included 401 patients with suspicious breast lesions who underwent breast MRI examinations, including SyMRI, DCE-MRI, and DWI, from September 2020 to September 2024. Quantitative parameters obtained from SyMRI included T1-Pre, T2-Pre, and proton density (PD-Pre) values of breast lesions before contrast injection, as well as T1-Gd, T2-Gd, and PD-Gd values after contrast injection. Additionally, difference values (Delta-T1, Delta-T2, Delta-PD) and enhancement ratios (T1-Ratio, T2-Ratio, PD-Ratio) were calculated. Two radiologists retrospectively evaluated the morphological and kinetic characteristics on DCE-MRI, using apparent diffusion coefficient (ADC) values of the lesions to assess tumors on DWI. Logistic regression and ANOVA were applied to identify significant parameter differences among the four breast cancer subtypes. Based on these selected parameters by logistic regression, five machine learning models were developed: Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Decision Tree (DT). We plotted Receiver Operating Characteristic (ROC) curves and calculated the area under the curve (AUC) as the primary metric to assess the performance of the best model. We utilized the SHAP library in Python to generate feature importance values for our model's predictions. Results : A total of 292 patients (median age, 53 years; age range, 27–80 years) met the inclusion criteria. Among these, 204 patients (median age, 52 years; age range, 27–78 years) were assigned to the training cohort, while 88 patients (median age, 53 years; age range, 27–80 years) were included in the testing cohort. Eleven parameters were identified across the four breast cancer subtypes( p ＜0.05). These parameters included two clinical pathological factors: age and menopause( p ＜0.001); five SyMRI parameters: T1-Gd, T2-Gd, PD-Gd, T1-Ratio, and PD-Ratio( p ＜0.05); three DCE-MRI parameters: burr sign, time–intensity curve (TIC), and Breast Imaging Reporting and Date System(BI-RADS) grading( p ＜0.001); and one DWI parameter: ADC-Tumor( p ＜0.001). The SVM model demonstrated the highest overall performance based on the comprehensive evaluation of multiple metrics in the training set, achieving superior diagnostic performance with AUC, accuracy, specificity, and sensitivity of 0.972, 82.5%, 94.76%, and 82.14%, respectively. This SVM model achieved AUC values of 0.979 for luminal A, 0.925 for luminal B, 0.971 for HER2-enriched, and 0.982 for triple-negative (TN) subtypes in the training set; AUC values of 0.973 for luminal A, 0.873 for luminal B, 0.956 for HER2-enriched, and 0.955 for TN subtypes in the testing set. The Shapley Additive Explanations (SHAP) tool to effectively identify the importance of features contributing to the model, with T2-Gd, PD-Ratio, and burr sign showing the highest contributions, achieving mean absolute SHAP values of 0.418, 0.340, and 0.264, respectively. Conclusion : Quantitative parameters derived from SyMRI mappings, DCE-MRI, and DWI may provide a non-invasive approach for differentiating between the molecular subtypes of breast cancer using various machine learning models. Biological sciences/Cancer Health sciences/Diseases Health sciences/Health care Health sciences/Medical research Synthetic Magnetic Resonance Imaging Dynamic Contrast-Enhanced MRI Diffusion-Weighted Imaging Molecular Subtype Breast Cancer Machine Learning Models Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Breast cancer accounts for over 11.7% of all new cancer cases and more than 680000 cancer-related deaths annually, ranking first in new cancer cases and fourth in cancer-related deaths worldwide. It is a highly heterogeneous disease that comprises various molecular subtypes, which exhibit significant differences in terms of incidence, risk factors, prognosis, and treatment sensitivity. Distinct patterns of disease expression and response to therapy can inform targeted treatments, ultimately impacting patient clinical outcomes and guiding therapeutic interventions[ 1 ]. Prognostic factors, including the status of estrogen receptors (ER), progesterone receptors (PR), and human epidermal growth factor receptor 2 (HER2), as well as the proliferation rate (Ki-67), are routinely applied in clinical practice. By utilizing immunohistochemistry and fluorescence in situ- hybridization, the currently accepted molecular subtypes of breast cancer include luminal A (ER+/PR+/HER2−, Ki-67 < 15%), luminal B (ER+/PR + or −/HER2 positive or negative, Ki-67 ≥ 15%), HER2-enriched (ER−/PR−, HER2 positive), and TN(ER−/PR−, HER2 negative)[ 2 ]. However, the use of prognostic factors has proven insufficient for robustly predicting outcomes and presents a barrier due to the invasive nature of the methods. Therefore, noninvasive imaging markers for assessing tumor heterogeneity may offer significant clinical benefits. DCE-MRI imaging is the most accurate and sensitive diagnostic imaging technique for detecting breast cancer. In the context of breast cancer, the ability to predict tumor molecular subtypes using DCE-MRI can significantly enhance early treatment planning and improve understanding of prognosis[ 3 ]. DWI has been extensively explored for the differentiation of malignant from benign breast lesions and helps to both predict the response to neoadjuvant chemotherapy in patients with breast cancer and correlate with molecular prognostic factors. Most studies relied on the ADC, a robust and easy-to-calculate parameter[ 4 ]. However, there is still a demand for alternative methods that utilize MRI markers as prognostic factors, enabling the differentiation of breast cancer into molecular subtypes. Advances in MRI over the past decade have led to a method for the rapid and simultaneous quantification of tissue T1, T2, and PD, known as SyMRI. SyMRI employs a saturation recovery-based fast spin echo sequence that acquires data at multiple saturation delay and echo times. The relaxation of spins across these delay and echo times results in signal changes that can be used to fit the quantitative spin parameters on a pixel-by-pixel basis. The quantitative values obtained with SyMRI have been reported to identify molecular subtypes in breast cancer[ 5 ]. SyMRI does not require gadolinium-based contrast agents; however, gadolinium rim enhancement has been shown to differentiate between triple-negative breast cancer(TNBC) and non-TNBC lesions when assessed in conjunction with pre-contrast T1 and T2 values[ 6 ]. Additionally, histogram features such as the 10th percentile, median, and 90th percentile values derived from the analysis of these quantitative values can help distinguish breast cancer subtypes, including TNBC from luminal B subtypes[ 7 ]. Nevertheless, the extensive value of SyMRI, particularly with contrast enhancement, in evaluating molecular subtypes of breast carcinoma remains unclear. The purpose of this study was to investigate whether various quantitative parameters from SyMRI, DCE-MRI, and DWI can be utilized to differentiate molecular subtypes in patients with breast cancer using multiple machine-learning models. 2. Materials and methods 2.1.Patient Population This retrospective analysis was approved by the institutional review board. A total of 401 consecutive breast MRI examinations were performed on patients known to have, or suspected of having, breast carcinoma between September 2020 and September 2024. The inclusion criteria were as follows: a) patients who had not undergone surgery or neoadjuvant treatment before the MRI examination; b) signal quality of SyMRI, DCE-MRI, and DWI was sufficient for quantitative measurement; c) breast cancers were diagnosed by pathology following surgery, and associated histopathologic examinations with immunohistochemistry confirmed hormone receptor status, including ER, PR, HER2, and Ki-67 results; d) masses larger than 1 cm on DCE-MRI images (Fig. 1 ). The tumors with the largest sizes were analyzed in this study for patients with multifocal tumors, allowing the analysis to be conducted on a per-patient basis. All methods were performed in accordance with the relevant guidelines and regulations of the institution.This study has been approved by the Medical Research Ethics Committee of Ningxia Medical University General Hospital (No.KYLL-2021-280), and all patients signed informed consent. 2.2.MRI Scans All patients underwent bilateral breast MRI examinations using a 3-T MRI system (Architect, GE Healthcare, USA) with an eight-channel phased-array breast coil. The scan sequences included routine T2-weighted and T1-weighted imaging, DWI with two b-values (0 and 1000 s/mm²), DCE-MRI, and SyMRI (MAGiC). Gadodiamide (GE Pharmaceuticals, USA) was administered intravenously as a bolus at a dose of 0.2 mmol/kg of body weight, with an injection rate of 2.5 mL/s, followed by a 20 mL saline flush at a rate of 3 mL/s. DCE-MRI was performed using a fat-suppressed T1-weighted three-dimensional (3D) fast spoiled gradient echo sequence, acquired 12 times continuously after the injection of the contrast agent. Axial bilateral pre-contrast MAGiC scans were obtained before the administration of the contrast agent, and post-contrast MAGiC scans were conducted immediately after the DCE-MRI. The detailed acquisition parameters were as follows: (1) Magic: TR/TE = 4000/12 ms, section thickness/spacing = 5mm/0.5 mm, FOV = 360 ×360 mm, matrix = 320 ×256, number of slices = 24; acquisition time = 5 minutes and 5 seconds; (2)DCE-MRI: TR/TE = 3.8/minimum ms, section thickness/spacing = 1mm/0.5mm, FOV = 350 ×320 mm, matrix = 320×320; (3)DWI: TR/TE = 5000/minimum ms, section thickness/spacing = 5 mm/0.5 mm, FOV = 320×320 mm, matrix = 320 ×320. 2.3.Data Processing Two radiologists, with 5 and 2 years of experience in breast MR imaging,respectively, were blinded to the results of the molecular subtypes and the pathologic diagnosis. They reviewed the SyMRI, DCE-MRI, and DWI scans using the dedicated Advantage Windows workstation (GE Healthcare, Advantage Windows 4.4). Each radiologist evaluated one breast cancer lesion per patient. The conclusions of the two radiologists were compared, and any discrepancies were resolved by consensus. The breast cancers were identified using DCE-MRI as anatomical guidance. The DCE-MRI imaging findings were assessed for morphological characteristics, including size, number, and margin, as well as intratumoral enhancement, a TIC curve, and BI-RADS (Breast Imaging Reporting and Data System) classification. TIC is based on a region of interest (ROI) plotted on the area of brightest enhancement to avoid bleeding and necrosis. The malignant tumors were visually assessed using high b-value (1000 s/mm²) DWI and MAGiC maps. For the MAGiC maps, a ROI was manually delineated to encompass the largest area of the lesion, while necrotic or cystic areas were excluded using DCE-MRI. All quantitative parameters before and after the injection of the contrast agent in the ROI of the axial MAGiC maps were calculated automatically on the GE 3-T MRI machine(Fig. 2 ). 2.4.Standard of reference Histopathologic results were analyzed by a pathologist with 12 years of experience. All pathological findings were obtained from histopathologic reports of surgical specimens. Receptor status included ER, PR, HER2, and Ki-67, and staining results were evaluated according to current guidelines. Molecular subtypes were defined based on the 2013 St. Gallen consensus as follows: luminal A (positive for ER and/or PR, with Ki-67 < 14%); luminal B (positive for ER and/or PR, with overexpressed HER2 or Ki-67 ≥ 14%); HER2-enriched (negative for ER and PR, with overexpressed HER2); and TN (negative for ER, PR, and HER2 receptors)[ 8 ]. 2.5. Machine Learning Model Development and SHAP The dataset was randomly divided into a training set and a test set(the distribution ratio is 7:3). The training set was utilized to develop and train the models, the test set was employed to evaluate the performance of these models. In this study, based on these selected significant parameters by logistic regression, five machine learning models were developed: Logistic Regression (LR)[ 9 ], Support Vector Machine (SVM)[ 10 ], K-Nearest Neighbors (KNN)[ 11 ], Random Forest (RF)[ 12 ], and Decision Tree (DT)[ 13 ]. The models were implemented using Python (version 3.8) and the Scikit-learn (version 0.22 )machine learning toolkit. The optimal hyperparameter combinations for each model were determined through a grid search with 5-fold cross-validation to build the optimal model[ 12 ]. To quantify the predictive capabilities of each optimal model, we plotted ROC curves and calculated AUC as the primary metric for assessing model performance. Additionally, accuracy, sensitivity, and specificity as derived from the confusion matrix, were used to evaluate model performance from multiple perspectives. SHAP (version 3.8) tool is a game theory technique used to explain the output of machine learning models[ 14 ]. It provides a unified framework for estimating feature importance and generates explanations for the model's behavior. We utilized the SHAP library in Python to generate feature importance values for our model's predictions on the validation and test sets. 2.6.Statistical Analysis All statistical analyses were conducted using IBM SPSS Statistics software (IBM Corp, Armonk, NY, USA, version 19.0) for Windows. Metric data values were represented as mean ± standard deviation, median (interquartile range), or percentage values, as appropriate. Continuous data were expressed as mean and range, while categorical variables were presented as percentages. Independent variables were defined as follows: menopause status (0 vs.1); tumor size (≤ 2 cm vs. > 2 cm); tumor histopathologic grade (G1 and G2 vs. G3); burr sign (0 vs. 1); molecular markers including ER, PR, HER2, and Ki-67 (positive vs. negative); and cancer subtype (luminal A, luminal B, HER2-enriched, and TN). Logistic regression and ANOVA were applied to identify significant differences in parameters among the four subtypes of breast cancer. Interobserver consistency for all quantitative parameters between the two radiologists was evaluated using intraclass correlation coefficient (ICC) analysis. The ICC was categorized as strong (r ≥ 0.75), moderate (r = 0.4–0.75), or weak (r < 0.4). 3. Results 3.1. Clinical Characteristics A total of 401 patients were recruited in the study, with 109 patients excluded. Ultimately, 292 patients (median age, 53 years; age range, 27–80 years) met the inclusion criteria (Fig. 1 ). Of these 292 patients, 204 (median age, 52 years [range, 27–78 years]) were in the training cohort, 88 (median age, 53 years [range, 27–80 years]) were in the testing cohort. The clinical and pathological characteristics of patients across different subtypes are presented in Table 1 . Among the 292 breast cancer cases, 60 (20.6%) were classified as luminal A, 149 (51.0%) as luminal B, 59 (20.2%) as HER2-enriched, and 24 (8.2%) as TN group. In our study, invasive ductal carcinoma was the predominant pathological type, accounting for 262 cases (89.7%), while there were only 30 cases (10.3%) of invasive lobular carcinoma. The differences in age and menopausal status among the four molecular types were statistically significant (p < 0.05). Table 1 Clinicopathological features stratified by molecular subtypes. characteristics Luminal A (n = 60) Luminal B (n = 149) HER2 Positive (n = 59) TN (n = 24) p value Age(year) xˉ ± s 52.65 ± 10.48 53.42 ± 11.76 50.10 ± 10.93 55.83 ± 10.85 p <0.001 Histological grade 0.992 Grade 1 29(48.33) 56(37.58) 24(40.67) 8(33.33) Grade 2 19(31.66) 60(40.26) 20(33.89) 7(29.17) Grade 3 12(20) 33(22.14) 15(25.42) 9(37.50) Menopause 30(43.22) 68(45.67) 31(48.21) 10(45.43) p <0.001 Pathological type Invasiveductal carcinoma 44(40.43) 120(46.58) 42(48.33) 14(44.22) 0.077 Invasive lobular carcinoma 16(59.57) 29(53.42) 17(51.67) 10(45,66) Lymph node metastasis,n(%) 0.119 Positive 20(33.33) 75(50.33) 20(33.89) 10(41.66) Native 40(66.67) 74(49.67) 39(66.11) 14(58.34) Note.Unless otherwise noted, other data are presented as numbers of participants, with percentages in parentheses.The p value listed in italics indicates a signiffcant difference. 3.2. Interobserver Agreement on Quantitative Parameters In this study, the ICC for SyMRI, DWI, DCE-MRI parameters ranged from 0.88 to 0.94. 3.3.Imaging findings and quantitative parameters of the four molecular subtypes Quantitative parameters of SyMRI, DCE-MRI, and DWI, stratified by molecular subtypes are presented in Table 2 . Nine parameters were identified across the four breast cancer subtypes(p<0.05). These parameters included five SyMRI parameters: T1-Gd, T2-Gd, PD-Gd, T1-Ratio, and PD-Ratio(p<0.05); three DCE-MRI parameters: burr sign, TIC, and BI-RADS grading(p<0.001); and one DWI parameter: ADC-Tumor(p<0.001). Table 2 Logistic analysis for characteristic parameters of DWI, DCE-MRI and SyMRI. characteristics Luminal A n=(60) Luminal B n=(149) HER2 n=(59) TN n=(24) P value ADC-Tumor 0.94 ± 0.18 1.00 ± 0.19 1.05 ± 0.19 0.95 ± 0.13 p <0.001 T1-Gd 583.83 ± 99.97 553.25 ± 119.83 519.01 ± 93.41 553.40 ± 86.48 0.004 T2-Gd 56.40 ± 10.87 62.08 ± 7.98 64.80 ± 7.42 64.71 ± 9.55 p <0.001 PD-Gd 73.34 ± 8.85 76.56 ± 13.53 79.40 ± 14.48 73.66 ± 10.95 p <0.001 T1-Ratio 0.42 ± 0.13 0.53 ± 0.14 0.52 ± 0.15 0.51 ± 0.15 p <0.001 PD-Ratio 0.42 ± 0.30 0.38 ± 0.28 0.43 ± 0.28 0.36 ± 0.23 p <0.001 Burr sign(+) 44(73.33) 89(59.73) 22 (37.29) 10(41.67) p <0.001 TIC(1,2,3) 4(6.67); 36(60.00); 20(33.33) 4 (2.68); 115(77.18); 30(20.13) 2(3.39); 47(79.66); 10(16.95) 22(91.67); 2 (8.33) p <0.001 BIRADS(2,3,4,5,6,7) 3:5(8.33), 4:5(8.33); 5:24(40.00); 6:26(43.33) 2:1(0.67); 3:1(0.67); 4:5(3.36); 5:63(42.28) 6:73(48.99; 7:6 (4.03) 3:1(1.69); 4:2(3.39); 5:27(45.76); 6:28(47.46); 7:1(1.69) 3:1(4.17); 5:9(37.50); 6:12(50.00); 7:2(8.33) p <0.001 Note.ADC-Tumor, T2-Gd, PD-Gd, T1-Ratio,T1-Ratio date are presented as mean, with standard deviation in parentheses; Burr sign, TIC, BIRADS data are presented as numbers of participants, with percentages in parentheses. The p value listed in italics indicates a signiffcant difference. 3.4. Five machine learning models Based on the characteristic parameters identified through logistic regression analysis, five machine learning models were developed. The SVM model demonstrated the highest overall performance based on the comprehensive evaluation of multiple metrics in the training set, achieving superior diagnostic performance with AUC, accuracy, specificity, and sensitivity of 0.972, 82.5%, 94.76%, and 82.14%, respectively(Fig. 3 , Table 3 ). This SVM model achieved AUC values of 0.979 for luminal A, 0.925 for luminal B, 0.971 for HER2-enriched, and 0.982 for TN subtypes in the training set; AUC values of 0.973 for luminal A, 0.873 for luminal B, 0.956 for HER2-enriched, and 0.955 for TN subtypes in the testing set. (Fig. 4 , Table 4 ). Table 3 Performance comparison of five machine learning models in the training set and testing set, respectively. Model Training set Testing set AUC ACC SPE SEN AUC ACC SPE SEN LR 0.864 68.75 89.37 68.42 0.823 59.58 87.37 60.16 SVM 0.973 83.39 93.32 83.45 0.962 82.50 94.76 82.14 KNN 0.978 81.96 94.43 81.93 0.939 77.91 94.24 77.51 RF 1 1 1 1 0.849 84.17 98.34 84.72 DT 1 1 1 1 0.866 80.41 94.24 79.88 Note.SVM:Support Vecor Machine, KNN: K-NearestNeighbor, AUC:area under the curve, ACC:accuracy, SPE:specificity, SEN: sensibility. Table 4 Performance comparison of SVM model in four molecular subtypes Subtype Training set Testing set AUC ACC SPE SEN AUC ACC SPE SEN Luminal A 0.979 93.39% 94.31% 89.32% 0.973 93.33% 95.54% 89.47% Luminal B 0.925 88.75% 95.01% 65.55% 0.873 87.92% 92.46% 65.85% HER2-enriched 0.971 92.14% 93.95% 85.09% 0.956 90.42% 92.78% 80.43% TN 0.982 93.75% 96.68% 81.48% 0.955 92.92% 92.92% 75.00% Note.TN:triple-negative,AUC:area under the curve, ACC:accuracy, SPE:specificity, SEN: sensibility. 3.5.SHAP values In this research, we utilized SHAP tool to reveal the individual contributions of each feature to the model predictions which is useful in model interpretation. Figure 5 A presents the feature importance plot for the SVM model, in order of importance based on the average absolute SHAP values. The beeswarm plot depicted in Fig. 5 B describes the specific impact of each feature on the prediction of molecular subtypes of breast cancer. Each dot represents the SHAP value of each feature for all individual patients, with the colors ranging from blue (low feature value) to red (high feature value). These points are distributed relative to a vertical line at zero, where all the feature values on the left side of zero exert a negative effect on molecular subtypes, while the feature values on the right side exert a positive effect on molecular subtypes. The features on the right indicated by red dots are positively correlated with molecular subtypes, while the features indicated by blue dots are negatively correlated with molecular subtypes. SHAP tool identifies the features that contribute the most to model prediction, among which T2-Gd, PD-Ratio, and burr sign contribute the highest values, achieving mean absolute SHAP values of 0.418, 0.340, and 0.264, respectively (Table 5 , Fig. 5 A, 5 B). Table 5 Mean absolute SHAP value based on SVM model characteristics Mean absolute SHAP value T2-Gd 0.418 PD-Ratio 0.340 Burr Sign 0.264 T1-Gd 0.206 T1-Ratio 0.189 ADC-Tumor 0.172 BI-RADS 0.161 Menopausal status 0.144 TIC 0.137 Age 0.093 PD-Gd 0.083 4. Discussion Knowing the molecular subtypes of breast cancer is essential for developing an accurate, patient-centered treatment plan[15]. In the context of breast imaging, MRI is a widely utilized modality for assessing tumor heterogeneity. In this study, we propose a diagnostic approach in which quantitative parameters derived from SyMRI, DCE-MRI, and DWI can be employed to create a clinically valuable predictive model that differentiates between various molecular subtypes of breast cancer using multiple machine learning algorithms. The study focused on developing machine learning models for differentiating between molecular subtypes in breast cancer patients. Five different machine-learning methods were employed and evaluated to forecast the survival of the patient. The study used the AUC value, accuracy, sensitivity, and specificity to evaluate the discriminative performance of the model. The AUC value was used as the main indicator for overall classification performance, regardless of how the classification threshold was set. The accuracy was used to evaluate the classification accuracy of the model in the overall sample. However, on our imbalanced data, the model may tend to predict the majority of categories, resulting in high accuracy and poor actual performance. The Sensitivity measures the model’s ability to recognize positive examples, while specificity indicators measure the model’s ability to recognize negative examples. Analysis based on the AUC values indicates that multiple machine-learning methods exhibit strong predictive performance in forecasting the molecular subtypes of breast cancer within the test dataset. Notably, SVM model outperformed the other four algorithms, achieving an AUC of 0.973 and 0.962 on the training and testing sets, respectively. This SVM model achieved AUC values of 0.979 for luminal A, 0.925 for luminal B, 0.971 for HER2-enriched, and 0.982 for TN subtypes in the training set; AUC values of 0.973 for luminal A, 0.873 for luminal B, 0.956 for HER2-enriched, and 0.955 for TN subtypes in the testing set. While its performance may vary across other metrics, including accuracy, sensitivity, and specificity, it demonstrated superior overall efficacy, effectively balancing positive and negative samples. This highlights the SVM model’s robust discriminatory ability among molecular subtypes of breast cancer. SHAP provided a more accurate explanation of the SVM model’s predictions by assigning feature importance scores to individual features. We observed that SHAP was particularly effective in identifying significant features derived from DCEI-MRI, DWI, and SyMRI. We created visualizations of the SHAP values to illustrate the MRI features that contributed most to the model’s predictions. These visualizations offer valuable insights into the model’s decision-making process and can assist medical professionals in interpreting the model’s predictions. SHAP test indicated that T2-Gd and the PD ratio were the parameters that contributed most significantly to the output magnitude of the SVM model. While the roles of DWI and DCE-MRI in identifying molecular subtypes have been confirmed through extensive analysis, SyMRI should not be overlooked, as it represents a promising tool for cancer diagnosis and treatment assessment[16]. SyMRI offers the advantages of quantitative T1, T2, and PD imaging. Its ability to simultaneously acquire this information in a single sequence provides significant benefits over the separate acquisition of these three parameter maps using independent mapping techniques[17]. The multiple SyMRI parameter maps are free from misregistration, and the shorter overall acquisition time of the SyMRI sequence reduces the risk of motion artifacts that could compromise any single image set, which would otherwise affect a combined analysis of the maps. Our study collected MAGIC images before and after the injection of the contrast agent; such data would facilitate the derivation of additional parameters related to perfusion, potentially allowing for a more comprehensive characterization of the tissue. Previous studies have reported similar segmental findings. Matsuda et al.reported that T2-Gd was significantly higher in the Ki-67 high-proliferation group than in the low-proliferation group, which is consistent with our results[18]. The findings of our study further demonstrated that, in addition to Ki-67 status, the molecular subtypes of breast cancer were correlated with synthetic parameters. Gao et al.found that T2 and T2-Gd were significantly different among the molecular subtypes (P = 0.000 and P = 0.006, respectively) and could further differentiate luminal A/B breast cancers from non-luminal subtypes (P = 0.005 and P = 0.015). T2 and T2-Gd values were lower for luminal A/B tumors (79.5 ms and 68.00 ms, respectively) and higher for non-luminal tumors (84.00 ms and 75.00 ms, respectively)[19]. However, it did not include the difference values and enhancement ratios of T1, T2, and PD before and after enhancement, which is why we obtained more valuable results with the PD ratio. This result needs to be confirmed by further studies in the future. Du et al.reported the largest AUC for T2 in differentiating between luminal A and other subtypes[20], while Li et al. found that when discriminating molecular subtypes of IDC, the T2 mean derived from histogram analysis achieved the highest performance. The T2 mean values for TN, luminal B, and luminal A types are arranged in descending order (p < 0.0001)[21]. Although these studies also suggested an important reference role for T2 parameter values of SyMRI, we investigated the diagnostic performance of synthetic MRI after the injection of the contrast agent for the differentiation of breast cancer subtypes. In our study, T2-Gd contributed the highest values to the output magnitude of the SVM model.The pathological basis for the differences in synthetic parameter values among various molecular subtypes is attributed to the cancer cells' fluid content in tissues, such as increased water or blood influx from both intracellular and extracellular spaces[22]. Microstructural differences at the cellular level result in variations in the intrinsic and fundamental tissue properties of T1, T2, and proton density (PD). Estrogen receptor (ER)-positive breast cancers typically exhibit reduced levels of vascular endothelial growth factor (VEGF) and lower neovascularity, which may correspond to lower T1, T2, and T2-Gd values. In contrast, HER2-enriched and TN tumors may present with high cellularity; however, it is likely that neovascularity in these tumors has a more significant impact on T1, T2, and T2-Gd values than cellularity[23]. This observation may explain the higher T1, T2, and T2-Gd values noted in HER2-enriched and TN molecular subtypes. As discussed by Du et al, more malignant breast cancers (ER/PR-negative subtypes or non-luminal breast cancers) thrive due to increased angiogenesis, which is associated with incomplete vascular endothelium leading to heightened vascular wall permeability[24]. Therefore, despite the notable overlap in quantitative measurements among different molecular subtypes, the differentiation capability demonstrated in our study indicates the potential of the synthetic technique[25]. We are confident that future advancements in synthetic technology, with higher spatial resolution and improved imaging quality, will further enhance its clinical applicability in the preoperative classification of breast tumor subtypes. SHAP test also indicated that the burr sign was the third most significant parameter contributing to the output magnitude of the SVM model. The differences in the burr sign may be attributed to the varying intensities of tissue hyperplasia responses[26]. High-grade and rapidly growing masses typically exhibit well-defined margins, while low-grade and slowly growing masses tend to have poorly defined margins and may appear spiculated. This phenomenon can be attributed to the desmoplastic reaction occurring in adjacent breast tissues[27, 28]. Consequently, this is the primary reason for the detection of different morphological characteristics across various subtypes. Numerous breast cancer subtypes possess specific imaging features. According to the relevant literature, TNBC often presents as benign-like masses with relatively circumscribed margins, frequently demonstrating rim enhancement and high signal intensity on T2-weighted images. In contrast, luminal-type breast cancers typically exhibit more irregularly shaped masses on MRI[29, 30]. Despite yielding promising results, it is important to acknowledge certain limitations. First, it was conducted at a single center with a relatively unevenly distributed sample size across the four molecular subtypes, which may limit the generalizability of our findings. Future research should include multi-center studies with larger and more balanced sample distributions to validate our results and enhance their applicability. Second, while necrosis could be identified on the parameter maps, the abundance of adipose tissue in the breast may obscure lesion borders, presenting challenges in accurately contouring breast tumors[31]. Additionally, the SyMRI method does not suppress fat signals. The multiple image sets produced by using SyMRI may facilitate the development of an automated segmentation algorithm in future studies[32]. Third, we did not consider the location of heterogeneity. Further studies incorporating texture analysis on SyMRI, DCE-MRI, and DWI quantitative imaging may more accurately represent intratumor heterogeneity. These directions will contribute to a comprehensive understanding of the utility of breast imaging and optimize its clinical application, ultimately leading to more personalized and effective treatment strategies for breast cancer patients. Conclusion Quantitative parameters derived from SyMRI mappings, DCE-MRI, and DWI may offer a non-invasive method for differentiating between the molecular subtypes of breast cancer using multiple machine learning models. Declarations Fund This study was supported by funding from Ningxia hui autonomous region key research and development project: research on image enhancement technology based on deep learning algorithms to improve quantitative magnetic resonance imaging techniques for rapid breast scanning (grant numbers: 2023BEG03022). Disclosures of Conf l icts of Interest All authors disclosed no relevant relationships. Date availability statement All date generated or analysed during this study are included in this published article and its supplementary information files. Author Contribution M.Y.X. and Y.L.G. wrote the main manuscript text , P.Z. , C.H.L. , J.L. , and Z.H.H. prepared figures 1-5,All authors reviewed the manuscript. References Goldhirsch, A., et al., Strategies for subtypes--dealing with the diversity of breast cancer: highlights of the St. Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2011. Ann Oncol, 2011. 22(8): p. 1736-47. Elkahwagy, D.M.A.S., C.J. Kiriacos and M. Mansour, Correction: Logistic regression and other statistical tools in diagnostic biomarker studies. Clinical & translational oncology, 2024. 26(9): p. 2393-2393. Jiang, H. and W.K. Ching, Correlation kernels for support vector machines classification with applications in cancer data. Comput Math Methods Med, 2012. 2012: p. 205025. Assegie, T.A., et al., Early Prediction of Gestational Diabetes with Parameter-Tuned K-Nearest Neighbor Classifier. Journal of Robotics and Control (JRC), 2023. 4(4): p. 452-457. Song, X., et al., Prognostic prediction of breast cancer patients using machine learning models: a retrospective analysis. Gland Surg, 2024. 13(9): p. 1575-1587. Wu, J., et al., Identifying relations between imaging phenotypes and molecular subtypes of breast cancer: Model discovery and external validation. J Magn Reson Imaging, 2017. 46(4): p. 1017-1027. Aldughayfiq, B., et al., Explainable AI for Retinoblastoma Diagnosis: Interpreting Deep Learning Models with LIME and SHAP. Diagnostics (Basel), 2023. 13(11): p. 1932. Kazama, T., T. Takahara and J. Hashimoto, Breast Cancer Subtypes and Quantitative Magnetic Resonance Imaging: A Systemic Review. Life (Basel), 2022. 12(4). Matsuda, M., et al., Utility of synthetic MRI in predicting the Ki-67 status of oestrogen receptor-positive breast cancer: a feasibility study. Clinical radiology, 2020. 75(5): p. 398.e1-398.e8. Li, H., et al., Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ Breast Cancer, 2016. 2: p. 16012-. Yuen, S., et al., The association between MRI findings and breast cancer subtypes: focused on the combination patterns on diffusion-weighted and T2-weighted images. Breast Cancer, 2020. 27(5): p. 1029-1037. Wang, W., et al., Prediction of Prognostic Factors and Genotypes in Patients With Breast Cancer Using Multiple Mathematical Models of MR Diffusion Imaging. Front Oncol, 2022. 12: p. 825264. Wu, S., et al., Predictive value of breast cancer molecular subtypes in Chinese patients with four or more positive nodes after postmastectomy radiotherapy. Breast (Edinburgh), 2012. 21(5): p. 657-661. Xu, A., et al., Prediction Breast Molecular Typing of Invasive Ductal Carcinoma Based on Dynamic Contrast Enhancement Magnetic Resonance Imaging Radiomics Characteristics: A Feasibility Study. Front Oncol, 2022. 12: p. 799232. Li, W., et al., A quantitative heterogeneity analysis approach to molecular subtype recognition of breast cancer in dynamic contrast-enhanced magnetic imaging images from radiomics data. Quant Imaging Med Surg, 2023. 13(7): p. 4429-4446. Chen, H., et al., Correlation of dynamic contrast-enhanced MRI and diffusion-weighted MR imaging with prognostic factors and subtypes of breast cancers. Front Oncol, 2022. 12: p. 942943. Hwang, K.P., et al., A Radiomics Model Based on Synthetic MRI Acquisition for Predicting Neoadjuvant Systemic Treatment Response in Triple-Negative Breast Cancer. Radiol Imaging Cancer, 2023. 5(4): p. e230009. Kato, F., et al., Differences in morphological features and minimum apparent diffusion coefficient values among breast cancer subtypes using 3-tesla MRI. Eur J Radiol, 2016. 85(1): p. 96-102. Wang, X., et al., Time-Dependent Diffusion MRI Helps Predict Molecular Subtypes and Treatment Response to Neoadjuvant Chemotherapy in Breast Cancer. Radiology, 2024. 313(1): p. e240288. Huang, G., et al., Molecular subtypes of breast cancer identified by dynamically enhanced MRI radiomics: the delayed phase cannot be ignored. Insights Imaging, 2024. 15(1): p. 127. Horvat, J.V., et al., Diffusion-weighted imaging (DWI) with apparent diffusion coefficient (ADC) mapping as a quantitative imaging biomarker for prediction of immunohistochemical receptor status, proliferation rate, and molecular subtypes of breast cancer. J Magn Reson Imaging, 2019. 50(3): p. 836-846. Lima, M., et al., Intravoxel Incoherent Motion and Quantitative Non-Gaussian Diffusion MR Imaging: Evaluation of the Diagnostic and Prognostic Value of Several Markers of Malignant and Benign Breast Lesions1. Radiology, 2018. 0(000). Yue, W.Y., et al., Predicting Breast Cancer Subtypes Using Magnetic Resonance Imaging Based Radiomics With Automatic Segmentation. J Comput Assist Tomogr, 2023. 47(5): p. 729-737. Zhan, T., C. Yi and Y. Lang, Predicting efficacy of neoadjuvant chemotherapy in breast cancer patients with synthetic magnetic resonance imaging method MAGiC: An observational cohort study. Eur J Radiol, 2024. 179: p. 111666. Leithner, D., et al., Radiomic Signatures Derived from Diffusion-Weighted Imaging for the Assessment of Breast Cancer Receptor Status and Molecular Subtypes. Molecular imaging and biology, 2020. 22(2): p. 453-461. Additional Declarations No competing interests reported. Supplementary Files Supplementarytable.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5608203","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":392404948,"identity":"91e215df-1348-40b4-847e-ebd7a5e61323","order_by":0,"name":"Mengying Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzklEQVRIie3RsQrCMBCA4UggXaJdU8SIDyAEAmJB+iwJgs/QUSjURXAWfIiCUHE7ydoHqKOLUwd9AFFHnTw3wfz7x91xhPh8PxijAYBQExlSCjjSCZYWxulMRwtmcERy0JBWzhYrrpCLCWPgmFNbUH6tG5LI4fwjsXBY50zvaHsbb8hUjwAxxUU57+2zdtnlBGyJIMrdctEqHD8jCQcFolKDJ2FIEiwNiNToKGM63ijELf0scBeh7jIM3alu0kR+JG8J7GteybfC5/P5/qIH6ulCRSU7Co4AAAAASUVORK5CYII=","orcid":"","institution":"Ningxia Medical University General Hospital","correspondingAuthor":true,"prefix":"","firstName":"Mengying","middleName":"","lastName":"Xu","suffix":""},{"id":392404950,"identity":"62d95710-1135-488a-a1d1-d793a96d81b4","order_by":1,"name":"Yali Gao","email":"","orcid":"","institution":"Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yali","middleName":"","lastName":"Gao","suffix":""},{"id":392404951,"identity":"16a4fd96-e0c5-483f-84b9-451bbd85f4d1","order_by":2,"name":"Pan Zhang","email":"","orcid":"","institution":"Ningxia Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Pan","middleName":"","lastName":"Zhang","suffix":""},{"id":392404952,"identity":"efb8e3eb-87ab-44b3-9c04-db2283f20084","order_by":3,"name":"Chunhua Li","email":"","orcid":"","institution":"Ningxia Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Chunhua","middleName":"","lastName":"Li","suffix":""},{"id":392404953,"identity":"486da405-819f-4342-aad0-7fe6793402dc","order_by":4,"name":"Jian Li","email":"","orcid":"","institution":"Ningxia Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jian","middleName":"","lastName":"Li","suffix":""},{"id":392404954,"identity":"dc98e75e-d69e-4ddf-a726-9be3f0d570ec","order_by":5,"name":"Zihan Hong","email":"","orcid":"","institution":"Ningxia Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Zihan","middleName":"","lastName":"Hong","suffix":""},{"id":392404955,"identity":"0a8affaf-2049-4ca8-a15c-7a46d4c979ed","order_by":6,"name":"Bing Chen","email":"","orcid":"","institution":"Ningxia Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Bing","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2024-12-09 10:38:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5608203/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5608203/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":72373059,"identity":"732cf04b-c7c5-4e4a-aa3c-b72440e2442b","added_by":"auto","created_at":"2024-12-26 08:06:06","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":26901,"visible":true,"origin":"","legend":"\u003cp\u003eThe flowchart of patient enrollment.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/6f43003f881ac3f9819f7182.png"},{"id":72373065,"identity":"11ecabe9-12af-4da8-98ad-34064efcf458","added_by":"auto","created_at":"2024-12-26 08:06:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":537504,"visible":true,"origin":"","legend":"\u003cp\u003eA 60-year-old woman with invasive ductal carcinoma of the right breast (ER-positive, PR-negative, HER2-negative, and Ki-67 high proliferating, Luminal B [HER2-]).An example of the synthetic images obtained before and after contrast agent injection.(A) T2 weight (synthetic); (B) T1 map; (C)T2map ; (D) Proton density (PD) map; (E) T2-Gd weight (synthetic);(F)T1map; (G)T2map ; (H) PD-Gd map;(I)DCE-MRI(phase 5);(J) DWI. T1-Pre =1004.00ms, T2-Pre = 71.66 ms, PD-Pre = 62.60 pu, T1-Gd =536.67 ms, T2-Gd = 60.30ms, PD-Gd = 83.93pu.ADCtumor=1.17 mm\u003csup\u003e2\u003c/sup\u003e /s、ADCpertumor=1.49 mm\u003csup\u003e2 \u003c/sup\u003e/s.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/22c65206a3e7f18e8cc0973a.png"},{"id":72374482,"identity":"e260a0ea-4b9b-4ee5-95f3-9d237400fe86","added_by":"auto","created_at":"2024-12-26 08:14:06","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":70694,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance comparison of five machine learning models in the training set and testing set\u003c/p\u003e\n\u003cp\u003eNote. LR: Logistic Regression, SVM: Support Vecor Machine, KNN: K-NearestNeighbor, RF: Random Forest, DT: Decision Tree.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/5e9741771e05fa56901be6ae.png"},{"id":72373066,"identity":"9cf26e1e-6051-407e-9e4d-8d840442c774","added_by":"auto","created_at":"2024-12-26 08:06:06","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":66065,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance comparison of SVM model in differentiating four molecular subtypes\u003c/p\u003e\n\u003cp\u003eNote. The normal control group was represented as Control Group.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/fb245b1805f9ede2ca9f5b5b.png"},{"id":72373073,"identity":"fc13e9ee-569c-4372-8ae8-eac2c9ba2c3d","added_by":"auto","created_at":"2024-12-26 08:06:06","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":42115,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP value based on SVM model\u003c/p\u003e\n\u003cp\u003eSummary SHAP plot. (A) Global feature importance in SVM model output. (B) Relationship between features and molecular subtypes in SVM model. T1-Gd, T2-Gd, and PD-Gd are the after enhanced T1, \u0026nbsp;T2, and PD values; T1-Ratio, PD-Ratio are the T1,PD ratios before and after the enhancement; ADC-Tumor is the mean intra-tumoral ADC value; TIC, time-intensity curve; BI-RADS, Breast Imaging Reporting and Date System.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/f5435010314a2b5546fec7a1.png"},{"id":79401525,"identity":"996a3ab3-020c-4967-bc63-4297190c915b","added_by":"auto","created_at":"2025-03-28 02:46:41","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1691599,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/cbd07b06-a0fc-4a3f-8d79-dcde99828ab6.pdf"},{"id":72373060,"identity":"a81afff9-91d8-4fda-a9fd-dd7672053741","added_by":"auto","created_at":"2024-12-26 08:06:06","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":45916,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarytable.docx","url":"https://assets-eu.researchsquare.com/files/rs-5608203/v1/4dcac64fca40de8092f90093.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Synthetic MRI, dynamic contrast-enhanced MRI combined with diffusion-weighted imaging for identifying molecular subtypes of breast cancer using machine learning models","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eBreast cancer accounts for over 11.7% of all new cancer cases and more than 680000 cancer-related deaths annually, ranking first in new cancer cases and fourth in cancer-related deaths worldwide. It is a highly heterogeneous disease that comprises various molecular subtypes, which exhibit significant differences in terms of incidence, risk factors, prognosis, and treatment sensitivity. Distinct patterns of disease expression and response to therapy can inform targeted treatments, ultimately impacting patient clinical outcomes and guiding therapeutic interventions[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Prognostic factors, including the status of estrogen receptors (ER), progesterone receptors (PR), and human epidermal growth factor receptor 2 (HER2), as well as the proliferation rate (Ki-67), are routinely applied in clinical practice. By utilizing immunohistochemistry and fluorescence in situ- hybridization, the currently accepted molecular subtypes of breast cancer include luminal A (ER+/PR+/HER2\u0026minus;, Ki-67\u0026thinsp;\u0026lt;\u0026thinsp;15%), luminal B (ER+/PR\u0026thinsp;+\u0026thinsp;or \u0026minus;/HER2 positive or negative, Ki-67\u0026thinsp;\u0026ge;\u0026thinsp;15%), HER2-enriched (ER\u0026minus;/PR\u0026minus;, HER2 positive), and TN(ER\u0026minus;/PR\u0026minus;, HER2 negative)[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, the use of prognostic factors has proven insufficient for robustly predicting outcomes and presents a barrier due to the invasive nature of the methods. Therefore, noninvasive imaging markers for assessing tumor heterogeneity may offer significant clinical benefits.\u003c/p\u003e \u003cp\u003eDCE-MRI imaging is the most accurate and sensitive diagnostic imaging technique for detecting breast cancer. In the context of breast cancer, the ability to predict tumor molecular subtypes using DCE-MRI can significantly enhance early treatment planning and improve understanding of prognosis[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. DWI has been extensively explored for the differentiation of malignant from benign breast lesions and helps to both predict the response to neoadjuvant chemotherapy in patients with breast cancer and correlate with molecular prognostic factors. Most studies relied on the ADC, a robust and easy-to-calculate parameter[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. However, there is still a demand for alternative methods that utilize MRI markers as prognostic factors, enabling the differentiation of breast cancer into molecular subtypes.\u003c/p\u003e \u003cp\u003eAdvances in MRI over the past decade have led to a method for the rapid and simultaneous quantification of tissue T1, T2, and PD, known as SyMRI. SyMRI employs a saturation recovery-based fast spin echo sequence that acquires data at multiple saturation delay and echo times. The relaxation of spins across these delay and echo times results in signal changes that can be used to fit the quantitative spin parameters on a pixel-by-pixel basis. The quantitative values obtained with SyMRI have been reported to identify molecular subtypes in breast cancer[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. SyMRI does not require gadolinium-based contrast agents; however, gadolinium rim enhancement has been shown to differentiate between triple-negative breast cancer(TNBC) and non-TNBC lesions when assessed in conjunction with pre-contrast T1 and T2 values[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Additionally, histogram features such as the 10th percentile, median, and 90th percentile values derived from the analysis of these quantitative values can help distinguish breast cancer subtypes, including TNBC from luminal B subtypes[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Nevertheless, the extensive value of SyMRI, particularly with contrast enhancement, in evaluating molecular subtypes of breast carcinoma remains unclear.\u003c/p\u003e \u003cp\u003eThe purpose of this study was to investigate whether various quantitative parameters from SyMRI, DCE-MRI, and DWI can be utilized to differentiate molecular subtypes in patients with breast cancer using multiple machine-learning models.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1.Patient Population\u003c/h2\u003e \u003cp\u003e This retrospective analysis was approved by the institutional review board. A total of 401 consecutive breast MRI examinations were performed on patients known to have, or suspected of having, breast carcinoma between September 2020 and September 2024. The inclusion criteria were as follows: a) patients who had not undergone surgery or neoadjuvant treatment before the MRI examination; b) signal quality of SyMRI, DCE-MRI, and DWI was sufficient for quantitative measurement; c) breast cancers were diagnosed by pathology following surgery, and associated histopathologic examinations with immunohistochemistry confirmed hormone receptor status, including ER, PR, HER2, and Ki-67 results; d) masses larger than 1 cm on DCE-MRI images (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The tumors with the largest sizes were analyzed in this study for patients with multifocal tumors, allowing the analysis to be conducted on a per-patient basis. All methods were performed in accordance with the relevant guidelines and regulations of the institution.This study has been approved by the Medical Research Ethics Committee of Ningxia Medical University General Hospital (No.KYLL-2021-280), and all patients signed informed consent.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2.MRI Scans\u003c/h2\u003e \u003cp\u003eAll patients underwent bilateral breast MRI examinations using a 3-T MRI system (Architect, GE Healthcare, USA) with an eight-channel phased-array breast coil. The scan sequences included routine T2-weighted and T1-weighted imaging, DWI with two b-values (0 and 1000 s/mm\u0026sup2;), DCE-MRI, and SyMRI (MAGiC). Gadodiamide (GE Pharmaceuticals, USA) was administered intravenously as a bolus at a dose of 0.2 mmol/kg of body weight, with an injection rate of 2.5 mL/s, followed by a 20 mL saline flush at a rate of 3 mL/s. DCE-MRI was performed using a fat-suppressed T1-weighted three-dimensional (3D) fast spoiled gradient echo sequence, acquired 12 times continuously after the injection of the contrast agent. Axial bilateral pre-contrast MAGiC scans were obtained before the administration of the contrast agent, and post-contrast MAGiC scans were conducted immediately after the DCE-MRI. The detailed acquisition parameters were as follows: (1) Magic: TR/TE\u0026thinsp;=\u0026thinsp;4000/12 ms, section thickness/spacing\u0026thinsp;=\u0026thinsp;5mm/0.5 mm, FOV\u0026thinsp;=\u0026thinsp;360 \u0026times;360 mm, matrix\u0026thinsp;=\u0026thinsp;320 \u0026times;256, number of slices\u0026thinsp;=\u0026thinsp;24; acquisition time\u0026thinsp;=\u0026thinsp;5 minutes and 5 seconds; (2)DCE-MRI: TR/TE\u0026thinsp;=\u0026thinsp;3.8/minimum ms, section thickness/spacing\u0026thinsp;=\u0026thinsp;1mm/0.5mm, FOV\u0026thinsp;=\u0026thinsp;350 \u0026times;320 mm, matrix\u0026thinsp;=\u0026thinsp;320\u0026times;320; (3)DWI: TR/TE\u0026thinsp;=\u0026thinsp;5000/minimum ms, section thickness/spacing\u0026thinsp;=\u0026thinsp;5 mm/0.5 mm, FOV\u0026thinsp;=\u0026thinsp;320\u0026times;320 mm, matrix\u0026thinsp;=\u0026thinsp;320 \u0026times;320.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3.Data Processing\u003c/h2\u003e \u003cp\u003eTwo radiologists, with 5 and 2 years of experience in breast MR imaging,respectively, were blinded to the results of the molecular subtypes and the pathologic diagnosis. They reviewed the SyMRI, DCE-MRI, and DWI scans using the dedicated Advantage Windows workstation (GE Healthcare, Advantage Windows 4.4). Each radiologist evaluated one breast cancer lesion per patient. The conclusions of the two radiologists were compared, and any discrepancies were resolved by consensus. The breast cancers were identified using DCE-MRI as anatomical guidance. The DCE-MRI imaging findings were assessed for morphological characteristics, including size, number, and margin, as well as intratumoral enhancement, a TIC curve, and BI-RADS (Breast Imaging Reporting and Data System) classification. TIC is based on a region of interest (ROI) plotted on the area of brightest enhancement to avoid bleeding and necrosis. The malignant tumors were visually assessed using high b-value (1000 s/mm\u0026sup2;) DWI and MAGiC maps. For the MAGiC maps, a ROI was manually delineated to encompass the largest area of the lesion, while necrotic or cystic areas were excluded using DCE-MRI. All quantitative parameters before and after the injection of the contrast agent in the ROI of the axial MAGiC maps were calculated automatically on the GE 3-T MRI machine(Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4.Standard of reference\u003c/h2\u003e \u003cp\u003eHistopathologic results were analyzed by a pathologist with 12 years of experience. All pathological findings were obtained from histopathologic reports of surgical specimens. Receptor status included ER, PR, HER2, and Ki-67, and staining results were evaluated according to current guidelines. Molecular subtypes were defined based on the 2013 St. Gallen consensus as follows: luminal A (positive for ER and/or PR, with Ki-67\u0026thinsp;\u0026lt;\u0026thinsp;14%); luminal B (positive for ER and/or PR, with overexpressed HER2 or Ki-67\u0026thinsp;\u0026ge;\u0026thinsp;14%); HER2-enriched (negative for ER and PR, with overexpressed HER2); and TN (negative for ER, PR, and HER2 receptors)[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Machine Learning Model Development and SHAP\u003c/h2\u003e \u003cp\u003eThe dataset was randomly divided into a training set and a test set(the distribution ratio is 7:3). The training set was utilized to develop and train the models, the test set was employed to evaluate the performance of these models. In this study, based on these selected significant parameters by logistic regression, five machine learning models were developed: Logistic Regression (LR)[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], Support Vector Machine (SVM)[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], K-Nearest Neighbors (KNN)[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], Random Forest (RF)[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], and Decision Tree (DT)[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The models were implemented using Python (version 3.8) and the Scikit-learn (version 0.22 )machine learning toolkit. The optimal hyperparameter combinations for each model were determined through a grid search with 5-fold cross-validation to build the optimal model[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. To quantify the predictive capabilities of each optimal model, we plotted ROC curves and calculated AUC as the primary metric for assessing model performance. Additionally, accuracy, sensitivity, and specificity as derived from the confusion matrix, were used to evaluate model performance from multiple perspectives.\u003c/p\u003e \u003cp\u003eSHAP (version 3.8) tool is a game theory technique used to explain the output of machine learning models[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. It provides a unified framework for estimating feature importance and generates explanations for the model's behavior. We utilized the SHAP library in Python to generate feature importance values for our model's predictions on the validation and test sets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6.Statistical Analysis\u003c/h2\u003e \u003cp\u003eAll statistical analyses were conducted using IBM SPSS Statistics software (IBM Corp, Armonk, NY, USA, version 19.0) for Windows. Metric data values were represented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation, median (interquartile range), or percentage values, as appropriate. Continuous data were expressed as mean and range, while categorical variables were presented as percentages. Independent variables were defined as follows: menopause status (0 vs.1); tumor size (\u0026le;\u0026thinsp;2 cm vs. \u0026gt; 2 cm); tumor histopathologic grade (G1 and G2 vs. G3); burr sign (0 vs. 1); molecular markers including ER, PR, HER2, and Ki-67 (positive vs. negative); and cancer subtype (luminal A, luminal B, HER2-enriched, and TN). Logistic regression and ANOVA were applied to identify significant differences in parameters among the four subtypes of breast cancer. Interobserver consistency for all quantitative parameters between the two radiologists was evaluated using intraclass correlation coefficient (ICC) analysis. The ICC was categorized as strong (r\u0026thinsp;\u0026ge;\u0026thinsp;0.75), moderate (r\u0026thinsp;=\u0026thinsp;0.4\u0026ndash;0.75), or weak (r\u0026thinsp;\u0026lt;\u0026thinsp;0.4).\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Clinical Characteristics\u003c/h2\u003e \u003cp\u003eA total of 401 patients were recruited in the study, with 109 patients excluded. Ultimately, 292 patients (median age, 53 years; age range, 27\u0026ndash;80 years) met the inclusion criteria (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Of these 292 patients, 204 (median age, 52 years [range, 27\u0026ndash;78 years]) were in the training cohort, 88 (median age, 53 years [range, 27\u0026ndash;80 years]) were in the testing cohort. The clinical and pathological characteristics of patients across different subtypes are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Among the 292 breast cancer cases, 60 (20.6%) were classified as luminal A, 149 (51.0%) as luminal B, 59 (20.2%) as HER2-enriched, and 24 (8.2%) as TN group. In our study, invasive ductal carcinoma was the predominant pathological type, accounting for 262 cases (89.7%), while there were only 30 cases (10.3%) of invasive lobular carcinoma. The differences in age and menopausal status among the four molecular types were statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClinicopathological features stratified by molecular subtypes.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003echaracteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLuminal\u003c/p\u003e \u003cp\u003eA\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;60)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLuminal\u003c/p\u003e \u003cp\u003eB\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;149)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHER2\u003c/p\u003e \u003cp\u003ePositive\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;59)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTN\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;24)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003ep value\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge(year) xˉ \u0026plusmn; s\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52.65\u0026thinsp;\u0026plusmn;\u0026thinsp;10.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53.42\u0026thinsp;\u0026plusmn;\u0026thinsp;11.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e50.10\u0026thinsp;\u0026plusmn;\u0026thinsp;10.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e55.83\u0026thinsp;\u0026plusmn;\u0026thinsp;10.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHistological grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29(48.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e56(37.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24(40.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8(33.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e19(31.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60(40.26)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20(33.89)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7(29.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12(20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e33(22.14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15(25.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9(37.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMenopause\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30(43.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68(45.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e31(48.21)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10(45.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePathological type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInvasiveductal carcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44(40.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e120(46.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e42(48.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14(44.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.077\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInvasive lobular carcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16(59.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29(53.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17(51.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10(45,66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLymph node metastasis,n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.119\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20(33.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75(50.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20(33.89)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10(41.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40(66.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e74(49.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e39(66.11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14(58.34)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eNote.Unless otherwise noted, other data are presented as numbers of participants, with percentages in parentheses.The \u003cem\u003ep\u003c/em\u003e value listed in italics indicates a signiffcant difference.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Interobserver Agreement on Quantitative Parameters\u003c/h2\u003e \u003cp\u003eIn this study, the ICC for SyMRI, DWI, DCE-MRI parameters ranged from 0.88 to 0.94.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3.Imaging findings and quantitative parameters of the four molecular subtypes\u003c/h2\u003e \u003cp\u003eQuantitative parameters of SyMRI, DCE-MRI, and DWI, stratified by molecular subtypes are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Nine parameters were identified across the four breast cancer subtypes(p\u0026lt;0.05). These parameters included five SyMRI parameters: T1-Gd, T2-Gd, PD-Gd, T1-Ratio, and PD-Ratio(p\u0026lt;0.05); three DCE-MRI parameters: burr sign, TIC, and BI-RADS grading(p\u0026lt;0.001); and one DWI parameter: ADC-Tumor(p\u0026lt;0.001).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLogistic analysis for characteristic parameters of DWI, DCE-MRI and SyMRI.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003echaracteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLuminal A\u003c/p\u003e \u003cp\u003en=(60)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLuminal B\u003c/p\u003e \u003cp\u003en=(149)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHER2\u003c/p\u003e \u003cp\u003en=(59)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTN\u003c/p\u003e \u003cp\u003en=(24)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003eP value\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eADC-Tumor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u0026thinsp;\u0026plusmn;\u0026thinsp;0.19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.05\u0026thinsp;\u0026plusmn;\u0026thinsp;0.19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.95\u0026thinsp;\u0026plusmn;\u0026thinsp;0.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1-Gd\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e583.83\u0026thinsp;\u0026plusmn;\u0026thinsp;99.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e553.25\u0026thinsp;\u0026plusmn;\u0026thinsp;119.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e519.01\u0026thinsp;\u0026plusmn;\u0026thinsp;93.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e553.40\u0026thinsp;\u0026plusmn;\u0026thinsp;86.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.004\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2-Gd\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56.40\u0026thinsp;\u0026plusmn;\u0026thinsp;10.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e62.08\u0026thinsp;\u0026plusmn;\u0026thinsp;7.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e64.80\u0026thinsp;\u0026plusmn;\u0026thinsp;7.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.71\u0026thinsp;\u0026plusmn;\u0026thinsp;9.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePD-Gd\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e73.34\u0026thinsp;\u0026plusmn;\u0026thinsp;8.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e76.56\u0026thinsp;\u0026plusmn;\u0026thinsp;13.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e79.40\u0026thinsp;\u0026plusmn;\u0026thinsp;14.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e73.66\u0026thinsp;\u0026plusmn;\u0026thinsp;10.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1-Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.42\u0026thinsp;\u0026plusmn;\u0026thinsp;0.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.53\u0026thinsp;\u0026plusmn;\u0026thinsp;0.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.52\u0026thinsp;\u0026plusmn;\u0026thinsp;0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.51\u0026thinsp;\u0026plusmn;\u0026thinsp;0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePD-Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.42\u0026thinsp;\u0026plusmn;\u0026thinsp;0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.38\u0026thinsp;\u0026plusmn;\u0026thinsp;0.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.43\u0026thinsp;\u0026plusmn;\u0026thinsp;0.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.36\u0026thinsp;\u0026plusmn;\u0026thinsp;0.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBurr sign(+)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44(73.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e89(59.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (37.29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10(41.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTIC(1,2,3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4(6.67);\u003c/p\u003e \u003cp\u003e36(60.00);\u003c/p\u003e \u003cp\u003e20(33.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (2.68);\u003c/p\u003e \u003cp\u003e115(77.18);\u003c/p\u003e \u003cp\u003e30(20.13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2(3.39);\u003c/p\u003e \u003cp\u003e47(79.66); 10(16.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22(91.67);\u003c/p\u003e \u003cp\u003e2 (8.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBIRADS(2,3,4,5,6,7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3:5(8.33), 4:5(8.33); 5:24(40.00);\u003c/p\u003e \u003cp\u003e6:26(43.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2:1(0.67); 3:1(0.67); 4:5(3.36); 5:63(42.28)\u003c/p\u003e \u003cp\u003e6:73(48.99; 7:6 (4.03)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3:1(1.69); 4:2(3.39); 5:27(45.76);\u003c/p\u003e \u003cp\u003e6:28(47.46);\u003c/p\u003e \u003cp\u003e7:1(1.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3:1(4.17); 5:9(37.50);\u003c/p\u003e \u003cp\u003e6:12(50.00); 7:2(8.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003ep\u003c/b\u003e\u003cb\u003e\u0026lt;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eNote.ADC-Tumor, T2-Gd, PD-Gd, T1-Ratio,T1-Ratio date are presented as mean, with standard deviation in parentheses; Burr sign, TIC, BIRADS data are presented as numbers of participants, with percentages in parentheses. The \u003cem\u003ep\u003c/em\u003e value listed in italics indicates a signiffcant difference.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Five machine learning models\u003c/h2\u003e \u003cp\u003eBased on the characteristic parameters identified through logistic regression analysis, five machine learning models were developed. The SVM model demonstrated the highest overall performance based on the comprehensive evaluation of multiple metrics in the training set, achieving superior diagnostic performance with AUC, accuracy, specificity, and sensitivity of 0.972, 82.5%, 94.76%, and 82.14%, respectively(Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). This SVM model achieved AUC values of 0.979 for luminal A, 0.925 for luminal B, 0.971 for HER2-enriched, and 0.982 for TN subtypes in the training set; AUC values of 0.973 for luminal A, 0.873 for luminal B, 0.956 for HER2-enriched, and 0.955 for TN subtypes in the testing set. (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance comparison of five machine learning models in the training set and testing set, respectively.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c9\" namest=\"c6\"\u003e \u003cp\u003eTesting set\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSPE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSEN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSPE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSEN\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.864\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e89.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e68.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.823\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e59.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e87.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e60.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e83.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e93.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e83.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.962\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e82.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e94.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e82.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e81.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e94.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e81.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.939\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e77.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e94.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e77.51\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.849\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e84.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e98.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e84.72\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.866\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e80.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e94.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e79.88\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003eNote.SVM:Support Vecor Machine, KNN: K-NearestNeighbor, AUC:area under the curve, ACC:accuracy, SPE:specificity, SEN: sensibility.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance comparison of SVM model in four molecular subtypes\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSubtype\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c9\" namest=\"c6\"\u003e \u003cp\u003eTesting set\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSPE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSEN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSPE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSEN\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLuminal A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.979\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e93.39%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e94.31%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e89.32%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e93.33%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e95.54%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e89.47%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLuminal B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.925\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e88.75%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e95.01%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e65.55%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.873\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e87.92%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e92.46%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e65.85%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHER2-enriched\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.971\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e92.14%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e93.95%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e85.09%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e90.42%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e92.78%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e80.43%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e93.75%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e96.68%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e81.48%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.955\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e92.92%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e92.92%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e75.00%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003eNote.TN:triple-negative,AUC:area under the curve, ACC:accuracy, SPE:specificity, SEN: sensibility.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.5.SHAP values\u003c/h2\u003e \u003cp\u003eIn this research, we utilized SHAP tool to reveal the individual contributions of each feature to the model predictions which is useful in model interpretation. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003eA presents the feature importance plot for the SVM model, in order of importance based on the average absolute SHAP values. The beeswarm plot depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003eB describes the specific impact of each feature on the prediction of molecular subtypes of breast cancer. Each dot represents the SHAP value of each feature for all individual patients, with the colors ranging from blue (low feature value) to red (high feature value). These points are distributed relative to a vertical line at zero, where all the feature values on the left side of zero exert a negative effect on molecular subtypes, while the feature values on the right side exert a positive effect on molecular subtypes. The features on the right indicated by red dots are positively correlated with molecular subtypes, while the features indicated by blue dots are negatively correlated with molecular subtypes. SHAP tool identifies the features that contribute the most to model prediction, among which T2-Gd, PD-Ratio, and burr sign contribute the highest values, achieving mean absolute SHAP values of 0.418, 0.340, and 0.264, respectively (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMean absolute SHAP value based on SVM model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003echaracteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean absolute SHAP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2-Gd\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.418\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePD-Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.340\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBurr Sign\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.264\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1-Gd\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.206\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1-Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.189\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eADC-Tumor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.172\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBI-RADS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.161\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMenopausal status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.144\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTIC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.137\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.093\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePD-Gd\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.083\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eKnowing the molecular subtypes of breast cancer is essential for developing an accurate, patient-centered treatment plan[15]. In the context of breast imaging, MRI is a widely utilized modality for assessing tumor heterogeneity. In this study, we propose a diagnostic approach in which quantitative parameters derived from SyMRI, DCE-MRI, and DWI can be employed to create a clinically valuable predictive model that differentiates between various molecular subtypes of breast cancer using multiple machine learning algorithms.\u003c/p\u003e\n\u003cp\u003eThe study focused on developing machine learning models for differentiating between molecular subtypes in breast cancer patients. Five different machine-learning methods were employed and evaluated to forecast the survival of the patient. The study used the AUC value, accuracy, sensitivity, and specificity to evaluate the discriminative performance of the model. The AUC value was used as the main indicator for overall classification performance, regardless of how the classification threshold was set. The accuracy was used to evaluate the classification accuracy of the model in the overall sample. However, on our imbalanced data, the model may tend to predict the majority of categories, resulting in high accuracy and poor actual performance. The Sensitivity measures the model’s ability to recognize positive examples, while specificity indicators measure the model’s ability to recognize negative examples. Analysis based on the AUC values indicates that multiple machine-learning methods exhibit strong predictive performance in forecasting the molecular subtypes of breast cancer within the test dataset. Notably, SVM model outperformed the other four algorithms, achieving an AUC of 0.973 and 0.962 on the training and testing sets, respectively. This SVM model achieved AUC values of 0.979 for luminal A, 0.925 for luminal B, 0.971 for HER2-enriched, and 0.982 for TN subtypes in the training set; AUC values of 0.973 for luminal A, 0.873 for luminal B, 0.956 for HER2-enriched, and 0.955 for TN subtypes in the testing set. While its performance may vary across other metrics, including accuracy, sensitivity, and specificity, it demonstrated superior overall efficacy, effectively balancing positive and negative samples. This highlights the SVM model’s robust discriminatory ability among molecular subtypes of breast cancer.\u003c/p\u003e\n\u003cp\u003eSHAP provided a more accurate explanation of the SVM model’s predictions by assigning feature importance scores to individual features. We observed that SHAP was particularly effective in identifying significant features derived from DCEI-MRI, DWI, and SyMRI. We created visualizations of the SHAP values to illustrate the MRI features that contributed most to the model’s predictions. These visualizations offer valuable insights into the model’s decision-making process and can assist medical professionals in interpreting the model’s predictions.\u003c/p\u003e\n\u003cp\u003eSHAP test indicated that T2-Gd and the PD ratio were the parameters that contributed most significantly to the output magnitude of the SVM model. While the roles of DWI and DCE-MRI in identifying molecular subtypes have been confirmed through extensive analysis, SyMRI should not be overlooked, as it represents a promising tool for cancer diagnosis and treatment assessment[16]. SyMRI offers the advantages of quantitative T1, T2, and PD imaging. Its ability to simultaneously acquire this information in a single sequence provides significant benefits over the separate acquisition of these three parameter maps using independent mapping techniques[17]. The multiple SyMRI parameter maps are free from misregistration, and the shorter overall acquisition time of the SyMRI sequence reduces the risk of motion artifacts that could compromise any single image set, which would otherwise affect a combined analysis of the maps. Our study collected MAGIC images before and after the injection of the contrast agent; such data would facilitate the derivation of additional parameters related to perfusion, potentially allowing for a more comprehensive characterization of the tissue. Previous studies have reported similar segmental findings. Matsuda et al.reported that T2-Gd was significantly higher in the Ki-67 high-proliferation group than in the low-proliferation group, which is consistent with our results[18]. The findings of our study further demonstrated that, in addition to Ki-67 status, the molecular subtypes of breast cancer were correlated with synthetic parameters. Gao et al.found that T2 and T2-Gd were significantly different among the molecular subtypes (P = 0.000 and P = 0.006, respectively) and could further differentiate luminal A/B breast cancers from non-luminal subtypes (P = 0.005 and P = 0.015). T2 and T2-Gd values were lower for luminal A/B tumors (79.5 ms and 68.00 ms, respectively) and higher for non-luminal tumors (84.00 ms and 75.00 ms, respectively)[19]. However, it did not include the difference values and enhancement ratios of T1, T2, and PD before and after enhancement, which is why we obtained more valuable results with the PD ratio. This result needs to be confirmed by further studies in the future. Du et al.reported the largest AUC for T2 in differentiating between luminal A and other subtypes[20], while Li et al. found that when discriminating molecular subtypes of IDC, the T2 mean derived from histogram analysis achieved the highest performance. The T2 mean values for TN, luminal B, and luminal A types are arranged in descending order (p \u0026lt; 0.0001)[21]. Although these studies also suggested an important reference role for T2 parameter values of SyMRI, we investigated the diagnostic performance of synthetic MRI after the injection of the contrast agent for the differentiation of breast cancer subtypes. In our study, T2-Gd contributed the highest values to the output magnitude of the SVM model.The pathological basis for the differences in synthetic parameter values among various molecular subtypes is attributed to the cancer cells' fluid content in tissues, such as increased water or blood influx from both intracellular and extracellular spaces[22]. Microstructural differences at the cellular level result in variations in the intrinsic and fundamental tissue properties of T1, T2, and proton density (PD). Estrogen receptor (ER)-positive breast cancers typically exhibit reduced levels of vascular endothelial growth factor (VEGF) and lower neovascularity, which may correspond to lower T1, T2, and T2-Gd values. In contrast, HER2-enriched and TN tumors may present with high cellularity; however, it is likely that neovascularity in these tumors has a more significant impact on T1, T2, and T2-Gd values than cellularity[23]. This observation may explain the higher T1, T2, and T2-Gd values noted in HER2-enriched and TN molecular subtypes. As discussed by Du et al, more malignant breast cancers (ER/PR-negative subtypes or non-luminal breast cancers) thrive due to increased angiogenesis, which is associated with incomplete vascular endothelium leading to heightened vascular wall permeability[24]. Therefore, despite the notable overlap in quantitative measurements among different molecular subtypes, the differentiation capability demonstrated in our study indicates the potential of the synthetic technique[25]. We are confident that future advancements in synthetic technology, with higher spatial resolution and improved imaging quality, will further enhance its clinical applicability in the preoperative classification of breast tumor subtypes.\u003c/p\u003e\n\u003cp\u003eSHAP test also indicated that the burr sign was the third most significant parameter contributing to the output magnitude of the SVM model. The differences in the burr sign may be attributed to the varying intensities of tissue hyperplasia responses[26]. High-grade and rapidly growing masses typically exhibit well-defined margins, while low-grade and slowly growing masses tend to have poorly defined margins and may appear spiculated. This phenomenon can be attributed to the desmoplastic reaction occurring in adjacent breast tissues[27, 28]. Consequently, this is the primary reason for the detection of different morphological characteristics across various subtypes. Numerous breast cancer subtypes possess specific imaging features. According to the relevant literature, TNBC often presents as benign-like masses with relatively circumscribed margins, frequently demonstrating rim enhancement and high signal intensity on T2-weighted images. In contrast, luminal-type breast cancers typically exhibit more irregularly shaped masses on MRI[29, 30].\u003c/p\u003e\n\u003cp\u003eDespite yielding promising results, it is important to acknowledge certain limitations. First, it was conducted at a single center with a relatively unevenly distributed sample size across the four molecular subtypes, which may limit the generalizability of our findings. Future research should include multi-center studies with larger and more balanced sample distributions to validate our results and enhance their applicability. Second, while necrosis could be identified on the parameter maps, the abundance of adipose tissue in the breast may obscure lesion borders, presenting challenges in accurately contouring breast tumors[31]. Additionally, the SyMRI method does not suppress fat signals. The multiple image sets produced by using SyMRI may facilitate the development of an automated segmentation algorithm in future studies[32]. Third, we did not consider the location of heterogeneity. Further studies incorporating texture analysis on SyMRI, DCE-MRI, and DWI quantitative imaging may more accurately represent intratumor heterogeneity. These directions will contribute to a comprehensive understanding of the utility of breast imaging and optimize its clinical application, ultimately leading to more personalized and effective treatment strategies for breast cancer patients.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eQuantitative parameters derived from SyMRI mappings, DCE-MRI, and DWI may offer a non-invasive method for differentiating between the molecular subtypes of breast cancer using multiple machine learning models.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFund\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by funding from Ningxia hui autonomous region key research and development project: research on image enhancement technology based on deep learning algorithms to improve quantitative magnetic resonance imaging techniques for rapid breast scanning \u0026nbsp; (grant numbers: 2023BEG03022).\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDisclosures of Conf\u003c/strong\u003e\u003cstrong\u003el\u003c/strong\u003e\u003cstrong\u003eicts of Interest\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll authors disclosed no relevant relationships.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDate availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll date generated or analysed during this study are included in this published article and its supplementary information files.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eM.Y.X. and Y.L.G. wrote the main manuscript text , P.Z. , C.H.L. , J.L. , and Z.H.H. prepared figures 1-5,All authors reviewed the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eGoldhirsch, A., et al., Strategies for subtypes--dealing with the diversity of breast cancer: highlights of the St. Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2011. Ann Oncol, 2011. 22(8): p. 1736-47.\u003c/li\u003e\n\u003cli\u003eElkahwagy, D.M.A.S., C.J. Kiriacos and M. Mansour, Correction: Logistic regression and other statistical tools in diagnostic biomarker studies. Clinical \u0026amp; translational oncology, 2024. 26(9): p. 2393-2393.\u003c/li\u003e\n\u003cli\u003eJiang, H. and W.K. Ching, Correlation kernels for support vector machines classification with applications in cancer data. Comput Math Methods Med, 2012. 2012: p. 205025.\u003c/li\u003e\n\u003cli\u003eAssegie, T.A., et al., Early Prediction of Gestational Diabetes with Parameter-Tuned K-Nearest Neighbor Classifier. Journal of Robotics and Control (JRC), 2023. 4(4): p. 452-457.\u003c/li\u003e\n\u003cli\u003eSong, X., et al., Prognostic prediction of breast cancer patients using machine learning models: a retrospective analysis. Gland Surg, 2024. 13(9): p. 1575-1587.\u003c/li\u003e\n\u003cli\u003eWu, J., et al., Identifying relations between imaging phenotypes and molecular subtypes of breast cancer: Model discovery and external validation. J Magn Reson Imaging, 2017. 46(4): p. 1017-1027.\u003c/li\u003e\n\u003cli\u003eAldughayfiq, B., et al., Explainable AI for Retinoblastoma Diagnosis: Interpreting Deep Learning Models with LIME and SHAP. Diagnostics (Basel), 2023. 13(11): p. 1932.\u003c/li\u003e\n\u003cli\u003eKazama, T., T. Takahara and J. Hashimoto, Breast Cancer Subtypes and Quantitative Magnetic Resonance Imaging: A Systemic Review. Life (Basel), 2022. 12(4).\u003c/li\u003e\n\u003cli\u003eMatsuda, M., et al., Utility of synthetic MRI in predicting the Ki-67 status of oestrogen receptor-positive breast cancer: a feasibility study. Clinical radiology, 2020. 75(5): p. 398.e1-398.e8.\u003c/li\u003e\n\u003cli\u003eLi, H., et al., Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ Breast Cancer, 2016. 2: p. 16012-.\u003c/li\u003e\n\u003cli\u003eYuen, S., et al., The association between MRI findings and breast cancer subtypes: focused on the combination patterns on diffusion-weighted and T2-weighted images. Breast Cancer, 2020. 27(5): p. 1029-1037.\u003c/li\u003e\n\u003cli\u003eWang, W., et al., Prediction of Prognostic Factors and Genotypes in Patients With Breast Cancer Using Multiple Mathematical Models of MR Diffusion Imaging. Front Oncol, 2022. 12: p. 825264.\u003c/li\u003e\n\u003cli\u003eWu, S., et al., Predictive value of breast cancer molecular subtypes in Chinese patients with four or more positive nodes after postmastectomy radiotherapy. Breast (Edinburgh), 2012. 21(5): p. 657-661.\u003c/li\u003e\n\u003cli\u003eXu, A., et al., Prediction Breast Molecular Typing of Invasive Ductal Carcinoma Based on Dynamic Contrast Enhancement Magnetic Resonance Imaging Radiomics Characteristics: A Feasibility Study. Front Oncol, 2022. 12: p. 799232.\u003c/li\u003e\n\u003cli\u003eLi, W., et al., A quantitative heterogeneity analysis approach to molecular subtype recognition of breast cancer in dynamic contrast-enhanced magnetic imaging images from radiomics data. Quant Imaging Med Surg, 2023. 13(7): p. 4429-4446.\u003c/li\u003e\n\u003cli\u003eChen, H., et al., Correlation of dynamic contrast-enhanced MRI and diffusion-weighted MR imaging with prognostic factors and subtypes of breast cancers. Front Oncol, 2022. 12: p. 942943.\u003c/li\u003e\n\u003cli\u003eHwang, K.P., et al., A Radiomics Model Based on Synthetic MRI Acquisition for Predicting Neoadjuvant Systemic Treatment Response in Triple-Negative Breast Cancer. Radiol Imaging Cancer, 2023. 5(4): p. e230009.\u003c/li\u003e\n\u003cli\u003eKato, F., et al., Differences in morphological features and minimum apparent diffusion coefficient values among breast cancer subtypes using 3-tesla MRI. Eur J Radiol, 2016. 85(1): p. 96-102.\u003c/li\u003e\n\u003cli\u003eWang, X., et al., Time-Dependent Diffusion MRI Helps Predict Molecular Subtypes and Treatment Response to Neoadjuvant Chemotherapy in Breast Cancer. Radiology, 2024. 313(1): p. e240288.\u003c/li\u003e\n\u003cli\u003eHuang, G., et al., Molecular subtypes of breast cancer identified by dynamically enhanced MRI radiomics: the delayed phase cannot be ignored. Insights Imaging, 2024. 15(1): p. 127.\u003c/li\u003e\n\u003cli\u003eHorvat, J.V., et al., Diffusion-weighted imaging (DWI) with apparent diffusion coefficient (ADC) mapping as a quantitative imaging biomarker for prediction of immunohistochemical receptor status, proliferation rate, and molecular subtypes of breast cancer. J Magn Reson Imaging, 2019. 50(3): p. 836-846.\u003c/li\u003e\n\u003cli\u003eLima, M., et al., Intravoxel Incoherent Motion and Quantitative Non-Gaussian Diffusion MR Imaging: Evaluation of the Diagnostic and Prognostic Value of Several Markers of Malignant and Benign Breast Lesions1. Radiology, 2018. 0(000).\u003c/li\u003e\n\u003cli\u003eYue, W.Y., et al., Predicting Breast Cancer Subtypes Using Magnetic Resonance Imaging Based Radiomics With Automatic Segmentation. J Comput Assist Tomogr, 2023. 47(5): p. 729-737.\u003c/li\u003e\n\u003cli\u003eZhan, T., C. Yi and Y. Lang, Predicting efficacy of neoadjuvant chemotherapy in breast cancer patients with synthetic magnetic resonance imaging method MAGiC: An observational cohort study. Eur J Radiol, 2024. 179: p. 111666.\u003c/li\u003e\n\u003cli\u003eLeithner, D., et al., Radiomic Signatures Derived from Diffusion-Weighted Imaging for the Assessment of Breast Cancer Receptor Status and Molecular Subtypes. Molecular imaging and biology, 2020. 22(2): p. 453-461.\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":"Synthetic Magnetic Resonance Imaging, Dynamic Contrast-Enhanced MRI, Diffusion-Weighted Imaging, Molecular Subtype, Breast Cancer, Machine Learning Models","lastPublishedDoi":"10.21203/rs.3.rs-5608203/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5608203/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eObjective\u003c/strong\u003e: To determine whether quantitative parameters from synthetic magnetic resonance imaging (SyMRI), dynamic contrast-enhanced MRI (DCE-MRI), and diffusion-weighted imaging (DWI) can effectively differentiate between molecular subtypes of breast cancer using various machine learning models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMaterials and Methods\u003c/strong\u003e: This retrospective study included 401 patients with suspicious breast lesions who underwent breast MRI examinations, including SyMRI, DCE-MRI, and DWI, from September 2020 to September 2024. Quantitative parameters obtained from SyMRI included T1-Pre, T2-Pre, and proton density (PD-Pre) values of breast lesions before contrast injection, as well as T1-Gd, T2-Gd, and PD-Gd values after contrast injection. Additionally, difference values (Delta-T1, Delta-T2, Delta-PD) and enhancement ratios (T1-Ratio, T2-Ratio, PD-Ratio) were calculated. Two radiologists retrospectively evaluated the morphological and kinetic characteristics on DCE-MRI, using apparent diffusion coefficient (ADC) values of the lesions to assess tumors on DWI. Logistic regression and ANOVA were applied to identify significant parameter differences among the four breast cancer subtypes. Based on these selected parameters by logistic regression, five machine learning models were developed: Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Decision Tree (DT). We plotted Receiver Operating Characteristic (ROC) curves and calculated the area under the curve (AUC) as the primary metric to assess the performance of the best model. We utilized the SHAP library in Python to generate feature importance values for our model's predictions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: A total of 292 patients (median age, 53 years; age range, 27–80 years) met the inclusion criteria. Among these, 204 patients (median age, 52 years; age range, 27–78 years) were assigned to the training cohort, while 88 patients (median age, 53 years; age range, 27–80 years) were included in the testing cohort. Eleven parameters were identified across the four breast cancer subtypes(\u003cem\u003ep\u003c/em\u003e＜0.05). These parameters included two clinical pathological factors: age and menopause(\u003cem\u003ep\u003c/em\u003e＜0.001); five SyMRI parameters: T1-Gd, T2-Gd, PD-Gd, T1-Ratio, and PD-Ratio(\u003cem\u003ep\u003c/em\u003e＜0.05); three DCE-MRI parameters: burr sign, time–intensity curve (TIC), and Breast Imaging Reporting and Date System(BI-RADS) grading(\u003cem\u003ep\u003c/em\u003e＜0.001); and one DWI parameter: ADC-Tumor(\u003cem\u003ep\u003c/em\u003e＜0.001). The SVM model demonstrated the highest overall performance based on the comprehensive evaluation of multiple metrics in the training set, achieving superior diagnostic performance with AUC, accuracy, specificity, and sensitivity of 0.972, 82.5%, 94.76%, and 82.14%, respectively. This SVM model achieved AUC values of 0.979 for luminal A, 0.925 for luminal B, 0.971 for HER2-enriched, and 0.982 for triple-negative (TN) subtypes in the training set; AUC values of 0.973 for luminal A, 0.873 for luminal B, 0.956 for HER2-enriched, and 0.955 for TN subtypes in the testing set. The Shapley Additive Explanations (SHAP) tool to effectively identify the importance of features contributing to the model, with T2-Gd, PD-Ratio, and burr sign showing the highest contributions, achieving mean absolute SHAP values of 0.418, 0.340, and 0.264, respectively.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e: \u0026nbsp;Quantitative parameters derived from SyMRI mappings, DCE-MRI, and DWI may provide a non-invasive approach for differentiating between the molecular subtypes of breast cancer using various machine learning models.\u003c/p\u003e","manuscriptTitle":"Synthetic MRI, dynamic contrast-enhanced MRI combined with diffusion-weighted imaging for identifying molecular subtypes of breast cancer using machine learning models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-26 08:06:01","doi":"10.21203/rs.3.rs-5608203/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":"f2040d83-3d61-41d5-bf43-214b1e82984f","owner":[],"postedDate":"December 26th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":41829529,"name":"Biological sciences/Cancer"},{"id":41829530,"name":"Health sciences/Diseases"},{"id":41829531,"name":"Health sciences/Health care"},{"id":41829532,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-03-28T02:38:34+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-26 08:06:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5608203","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5608203","identity":"rs-5608203","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}