Multiparametric MRI-based Habitat Analysis Integrating Deep Learning and Radiomics for Predicting Preoperative Ki-67 Expression Level in Breast Cancer

Multiparametric MRI-based Habitat Analysis Integrating Deep Learning and Radiomics for Predicting Preoperative Ki-67 Expression Level in Breast Cancer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Multiparametric MRI-based Habitat Analysis Integrating Deep Learning and Radiomics for Predicting Preoperative Ki-67 Expression Level in Breast Cancer Yuqian Wang, Yue Zhang, Zaiyi Liu, Yiming Xiong, Mifang Li, Lingyan Zhang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7709859/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 16 Jan, 2026 Read the published version in BMC Medical Imaging → Version 1 posted 12 You are reading this latest preprint version Abstract Background Breast cancer (BC) is the most common malignant tumor in women globally. Ki-67, vital for prognosis, is currently detected invasively. Non-invasive MRI prediction faces challenges due to intratumoral heterogeneity. Materials and Methods This retrospective study included 254 breast cancer patients from two centers, divided into training (142 patients), internal validation set (60 patients), and external validation set (52 patients). T2WI and DCE-MRI sequences were analyzed. Traditional radiomics features were extracted from intratumoral, peritumoral (5 mm, 10 mm), and habitat regions. A pre-trained ResNet-50 model extracted 2.5D deep learning features. Feature selection used ICC, Z-score normalization, t-tests, Pearson correlations, and LASSO. Models were built using SVM, RF, and ET algorithms, evaluated via AUC, accuracy, sensitivity, specificity, and F1 score. SHAP analysis enhanced interpretability. Results The best-performing traditional radiomics model achieved an AUC of 0.825 (95% CI: 0.708–0.942) in the internal validation set. The optimal deep learning model obtained an AUC of 0.804 (95% CI: 0.641–0.966) in the internal validation set. The combined model, utilizing both best traditional radiomics and deep learning features, demonstrated superior performance with an AUC of 0.885 (95% CI: 0.787–0.984) in the internal validation set and 0.839 (95% CI: 0.727–0.951) in the external validation set. Conclusion The integrated model combining traditional radiomics and deep learning from MRI significantly predicts Ki-67 expression in breast cancer, enhancing preoperative prediction accuracy and interpretability for personalized treatment. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Breast cancer is the most common and highly malignant tumor in women worldwide( 1 ). Accurate risk-stratification and timely treatment are essential for improving prognosis. The routinely used IHC biomarkers include Progesterone Receptor (PR), Estrogen Receptor (ER), Human Epidermal Growth Factor Receptor 2 (HER-2), and Ki-67. Among them, the proliferation index Ki-67 has been demonstrated to correlate strongly with tumor invasiveness( 2 ), early metastasis( 3 ), recurrence rate ( 4 )and overall survival( 5 , 6 ). However, Ki-67 assessment currently relies on core-needle or surgical biopsy followed by immunohistochemistry (IHC), an invasive procedure that is painful, may delay therapy, and occasionally provokes adverse reactions( 7 ). Consequently, there is an MRI has become the preferred imaging modality for breast-cancer evaluation. T2-weighted imaging (T2WI) provides high-resolution anatomical information and reflects the tumor microenvironment, whereas dynamic contrast-enhanced MRI (DCE-MRI) reveals detailed hemodynamic characteristics( 8 , 9 ). Nevertheless, conventional qualitative interpretation fails to exploit the rich quantitative information latent in these images, particularly the quantitative features related to the Ki67 proliferation index( 10 ). Radiomics can extract high-throughput imaging features to build predictive models, yet traditional whole-tumor radiomics analyses ignore intratumoral heterogeneity (ITH)( 11 – 13 ). Habitat analysis partitions tumors into biologically distinct subregions (“habitats”), thereby explicitly quantifying ITH and improving model Although fusion of radiomics and deep learning is promising, the relative merits of image-level versus feature-level fusion for Ki-67 prediction remain unexplored( 14 ). Furthermore, model interpretability remains a major barrier to clinical translation; the SHAP algorithm has demonstrated utility in identifying key features and explaining single-sample predictions across multiple medical scenarios( 15 – 17 ). Therefore, the aim of this study was twofold: (i) to develop habitat-based radiomics and convolutional neural network (CNN) models, and (ii) to systematically compare their efficacy—alongside image- and feature-level fusion strategies—for the non-invasive prediction of Ki-67 expression in breast cancer. Materials and Methods Study Sample We collected clinical data and MRI images of 320 patients with histologically confirmed breast cancer from two centers. Detailed MRI acquisition parameters and sample size estimation based on EPV are provided in Appendix 1. After applying inclusion and exclusion criteria, patients from Center A were randomly assigned to the training set (n = 142) and internal validation set (n = 60) in a 7:3 ratio, while patients from Center B constituted the external test set (n = 52). The patient recruitment process is shown in Fig. 1 . Ethics approval was waived by the IRBs of both centers for this minimal-risk, retrospective analysis of de-identified data. Clinical and MRI imaging characteristics This study included clinical characteristics crucial for prognosis. Ki-67, defined by positive nuclear staining percentage, uses 20% as a threshold to distinguish prognosis groups. Two blinded radiologists assessed imaging features per BI-RADS guidelines; disagreements were resolved by a senior radiologist. Image acquisition and mask segmentation All patients underwent MRI scans with uniform settings. The scans covered the entire breast, using FS-T2WI and DCE-MRI sequences. Images were preprocessed by resampling to 1 × 1 × 1 mm resolution, with N4 bias field correction and normalization. Two junior radiologists manually outlined tumor boundaries on the sequences with the most obvious T2WI and DCE-MRI using ITK-SNAP 3.6.0 software to generate ROIs. ICC assessed intra - and inter-observer agreement; features were stable and reliable when ROI ICC values were above 0.80. A senior radiologist reviewed each ROI. Radiologists were blinded to clinical information during segmentation and review. The research process is shown in Fig. 2 . Construction of traditional radiomics models We expanded the ROI peritumorally by 5 mm and 10 mm using the Onekey AI platform's Mask Filling Toolkit and merged intratumoral and peritumoral regions to form a new ROI for image fusion analysis. We used an unsupervised K-means clustering method (K = 3–10) to generate subregions in the tumor endosomes, determining the optimal number of clusters ( 3 ) via the Calinski-Harabasz (CH) value. Pyradiomics (v3.0.1) extracted 1197 image histologic features from each ROI of T2WI and DCE-MRI sequences. We designed two fusion strategies (image fusion and feature fusion) and performed multi-stage feature screening (ICC ≥ 0.80, Z-score standardization, t-test, Pearson correlation coefficient, LASSO regression) to reduce overfitting. Finally, we used grid search with 5-fold cross-validation to construct 48 prediction models based on SVM, RF, and ET algorithms. Construction of 2.5D Deep Learning Model We extracted 2D ROIs from the largest tumor cross-section and adjacent sections in T2WI and DCE sequences. After cropping and Z-score normalization, tumor regions from three sections were combined into a multichannel image. Using a pre-trained ResNet-50 CNN, we extracted deep features, reducing them from 2048 to 8 via PCA to minimize overfitting. We designed two fusion strategies and built 12 predictive models using three machine learning algorithms. Grad-CAM was used to generate heat maps for model interpretation. Construction of the combined model First, we screened the model features with the best performance of AUC values from the traditional radiomics model and 2.5D deep learning model for feature splicing. The fused features were processed through the same standardized processing and feature selection as traditional radiomics models and deep learning models. Three combined prediction models were constructed using three machine learning algorithms. This staged feature fusion and multi-algorithm validation set framework retains the interpretability of traditional radiomics features, integrates deep image features extracted by deep learning, and ensures model stability through cross-validation. External Verification and Interpretability We compared the predictive performance of the best traditional radiomics model, the best deep learning model, and the best fusion model on the internal validation set and identified the optimal predictive model. Subsequently, we evaluated the performance of this optimal model on an external validation set. To enhance the interpretability of the best model, we applied the SHAP algorithm, which decomposes the contribution of each radiomic feature to the model’s prediction, allowing us to clearly understand the individual impact of each feature. Statistical analysis Statistical analysis used Python 3.9.13. Normally distributed continuous variables were described as mean ± SD and compared via t-test; non-normally distributed variables as median (quartiles) and compared via Man-Whitney U test. Categorical variables were expressed as counts/percentages and compared using chi-square or Fisher's test. Variables with P < 0.05 in univariate analyses were included in multivariate analyses, with P < 0.05 indicating significance. Model performance was assessed using AUC, 95% CI, sensitivity, specificity, F1 score, calibration curves, and decision curve analysis (DCA) for clinical applicability. Results Patient Characteristics Center A identified 202 cases and was randomized in a 7:3 ratio into a training s (N=142) and an internal validation set (N=60). The sample from Center B was used as the external test set (N=52) and clinical baseline data were compared between the training, internal validation and external test sets. The Clinical and MRI Imaging characteristics of all patients are listed in Table 1. In the training and internal validation cohort, PR, ER, and HER-2 status showed significant differences between the high Ki-67 and low Ki-67 groups (P < 0.05). In the external cohort, significant differences were observed in PR and HER-2 status (P < 0.05). However, no significant differences were found in age, tumor diameter, menopausal status, T stage, TIC, BPE, BI-RADS, FGT, tumor internal enhancement, mrALN, pathological LNM, and LVI between the high Ki-67 and low Ki-67 groups in any of the three cohorts (P > 0.05). Table 1: Comparison of Clinicopathologic Characteristics between the Training and External Validation Data Sets Training cohort Validation cohort Test cohort Characteristics High Ki-67 ( n =95) Low Ki-67 ( n =47) p Value High Ki-67 ( n =47) Low Ki-67 ( n =13) p Value High Ki-67 ( n =27) Low Ki-67 ( n =24) p Value Age (mean±SD) 45.60±9.69 47.53±9.18 0.31 46.13±9.06 48.23±9.37 0.465 44.29±8.33 49.26±10.31 0.062 Maximum_tumor_diameter 34.26±20.14 37.36±24.15 0.45 47.13±22.21 40.62±23.29 0.29 28.00±11.20 23.61±6.83 0.56 Menopausal_status 69 (72.63) 35 (74.47) 0.98 19 (40.43) 8 (61.54) 0.30 24 (85.71) 16 (69.57) 0.292 26 (27.37) 12 (25.53) 28 (59.57) 5 (38.46) 4 (14.29) 7 (30.43) Tumor_location Left 48 (50.53) 20 (42.55) 0.47 25 (53.19) 9 (69.23) 0.47 16 (57.14) 13 (56.52) 1.00 Right 47 (49.47) 27 (57.45) 22 (46.81) 4 (30.77) 12 (42.86) 10 (43.48) cT_stage Ⅰ 10 (10.53) 4 (8.51) 0.41 6 (12.77) 1 (7.69) 0.806 8 (28.57) 4 (17.39) 0.35 Ⅱ 40 (42.11) 19 (40.43) 19 (40.43) 6 (46.15) 19 (67.86) 16 (69.57) Ⅲ 37 (38.95) 23 (48.94) 15 (31.91) 3 (23.08) 1 (3.57) 3 (13.04) Ⅳ 8 (8.42) 1 (2.13) 7 (14.89) 3 (23.08) NA NA TIC Ⅰ NA 1 (2.13) 0.361 NA NA 0.21 1 (3.57) NA 0.654 Ⅱ 19 (20.00) 9 (19.15) 26 (55.32) 4 (30.77) 9 (32.14) 8 (34.78) Ⅲ 76 (80.00) 37 (78.72) 21 (44.68) 9 (69.23) 18 (64.29) 15 (65.22) BPE 1 47 (49.47) 30 (63.83) 0.18 24 (51.06) 9 (69.23) 0.188 1 (3.57) null 0.29 2 35 (36.84) 10 (21.28) 21 (44.68) 2 (15.38) 14 (50.00) 17 (73.91) 3 11 (11.58) 7 (14.89) 1 (2.13) 1 (7.69) 12 (42.86) 5 (21.74) 4 2 (2.11) NA 1 (2.13) 1 (7.69) 1 (3.57) 1 (4.35) BI_RADS 3 NA 1 (2.13) 0.24 1 (2.13) 1 (7.69) 0.72 NA NA 0.92 4 14 (14.74) 8 (17.02) 6 (12.77) 1 (7.69) 17 (60.71) 15 (65.22) 5 42 (44.21) 25 (53.19) 19 (40.43) 6 (46.15) 10 (35.71) 7 (30.43) 6 39 (41.05) 13 (27.66) 21 (44.68) 5 (38.46) 1 (3.57) 1 (4.35) FGT 1 1 (1.05) 1 (2.13) 0.16 NA NA 0.19 NA NA 1.00 2 2 (2.11) 3 (6.38) 4 (8.51) NA NA NA 3 86 (90.53) 36 (76.60) 37 (78.72) 13 (100.00) 28 (100.00) 23 (100.00) 4 6 (6.32) 7 (14.89) 6 (12.77) NA NA NA Tumor_internal_enhancement Negative 47 (49.47) 25 (53.19) 0.81 18 (38.30) 7 (53.85) 0.49 17 (60.71) 18 (78.26) 0.30 Positive 48 (50.53) 22 (46.81) 29 (61.70) 6 (46.15) 11 (39.29) 5 (21.74) mrALN Negative 25 (26.32) 20 (42.55) 0.08 12 (25.53) 5 (38.46) 0.57 15 (53.57) 16 (69.57) 0.38 Positive 70 (73.68) 27 (57.45) 35 (74.47) 8 (61.54) 13 (46.43) 7 (30.43) LNM Negative 36 (37.89) 21 (44.68) 0.55 16 (34.04) 8 (61.54) 0.14 10 (35.71) 14 (60.87) 0.13 Positive 59 (62.11) 26 (55.32) 31 (65.96) 5 (38.46) 18 (64.29) 9 (39.13) LVI Negative 46 (48.42) 25 (53.19) 0.72 35 (74.47) 7 (53.85) 0.27 24 (85.71) 21 (91.30) 0.857 Positive 49 (51.58) 22 (46.81) 12 (25.53) 6 (46.15) 4 (14.29) 2 (8.70) PR status Negative 45 (47.37) 10 (21.28) 0.01 8 (17.02) 7 (53.85) 0.02 6 (21.43) 13 (56.52) 0.02 Positive 50 (52.63) 37 (78.72) 39 (82.98) 6 (46.15) 22 (78.57) 10 (43.48) ER status Negative 35 (36.84) 8 (17.02) 0.03 9 (19.15) 7 (53.85) 0.03 7 (25.00) 4 (17.39) 0.753 Positive 60 (63.16) 39 (82.98) 38 (80.85) 6 (46.15) 21 (75.00) 19 (82.61) HER2 status Negative 47 (49.47) 33 (70.21) 0.03 19 (40.43) 9 (69.23) 0.13 11 (39.29) 17 (73.91) 0.03 Positive 48 (50.53) 14 (29.79) 28 (59.57) 4 (30.77) 17 (60.71) 6 (26.09) Performance of traditional radiomics models We compared 48 radiomics models built from different sequences, regions, and fusion strategies in Figure 3. Habitat analysis identified three optimal intratumoral habitats (k = 3) and Grad-CAM mapped ROIs to corresponding heatmaps in Figure 4. Feature fusion strategies outperformed image fusion strategies. In DCE-MRI, the intratumoral habitat feature fusion model using ET achieved an AUC of 0.791 (95% CI: 0.667 - 0.914) on internal validation, surpassing the intratumoral ROI model's AUC of 0.772 (95% CI: 0.624 - 0.919). In T2W sequences, the feature fusion model combining intratumoral and 5-mm peritumoral regions achieved an AUC of 0.811 (95% CI: 0.680 - 0.942), higher than the 0.708 (95% CI: 0.564 - 0.852) from image-level fusion. DCE-MRI models consistently outperformed T2WI models. For example, the intratumoral model alone achieved an AUC of 0.772 (95% CI: 0.624 - 0.919) for DCE-MRI versus 0.762 (95% CI: 0.616 - 0.908) for T2WI. In peritumoral analysis, the 5-mm peritumoral zone outperformed the 10-mm extension. The T2WI-based habitat model using ET and integrating 5-mm peritumoral features achieved an AUC of 0.814 (95% CI: 0.708 - 0.921), surpassing the 0.804 (95% CI: 0.679 - 0.930) with 10-mm extension. The ET-based DCE-MRI fusion radiomics model integrating tumor habitat and 5 mm peritumoral features showed excellent performance: AUC of 0.878 (95% CI: 0.819 - 0.937), accuracy 0.810, sensitivity 0.789, specificity 0.851 in training; AUC 0.825 (95% CI: 0.708 - 0.942), accuracy 0.817, sensitivity 0.851, specificity 0.692 in internal validation. These results confirm the superiority of feature fusion and identify 5 mm as the optimal peritumoral extension distance. Figure 3 ： Heatmap of AUC values for models with different feature types, sequences, regions, and fusion methods. AUC stands for Area Under the Curve, SVM for Support Vector Machine, RF for Random Forest, ET for Extreme Trees, Rad for traditional radiomics models, DL for 2.5D deep learning models, T2WI for T2-weighted imaging sequences, and DCE-MRI for Dynamic Contrast-Enhanced MRI sequences, I for intra-tumoral，P5 for peritumoral5mm, P10 for peritumoral10mm,H for habitat，MC for multi-channel, MS for multi-slice,”*” for image fusion. Figure 4: (A) Calinski–Harabasz index curve: the score decreases monotonically with increasing cluster numbers, peaking at k = 3, indicating the optimal cluster number is three. (B) Habitat visualization: the Habitat ROIs from each sequence's habitat are mapped to the corresponding three Grad-CAM heatmaps above. Performance of deep learning models We compared 12 deep learning models built using different sequences and fusion strategies in Figure 3. The multi-slice feature fusion strategy showed significant advantages. For example, under the DCE sequence of the ET, multi-slice feature fusion achieved an AUC of 0.768 (95% CI: 0.616 - 0.919) in the internal test set, higher than the 0.718 (95% CI: 0.557 - 0.879) from multi-channel fusion. In sequence comparison, the T2WI sequence models showed better discriminative performance. Specifically, using a multi-channel image fusion strategy based on SVM, the T2WI sequence achieved an AUC of 0.763 (95% CI: 0.610 - 0.915) in the internal test set, while the DCE-MRI sequence only reached 0.699 (95% CI: 0.548 - 0.850). Notably, the radiomics model based on the T2WI sequence, integrating the feature fusion algorithm of the maximum tumor multi-slice and its two adjacent slices, performed the best overall. In the training cohort, this model achieved an AUC of 0.871 (95% CI: 0.813 - 0.929), and in the internal validation set, an AUC of 0.804 (95% CI: 0.641 - 0.966). The model also showed accuracy of 0.789 and 0.850, sensitivities of 0.747 and 0.915, and specificities of 0.872 and 0.615 in the training and test sets respectively. These results confirm the superiority of the multi-slice feature fusion strategy over multi-channel image fusion and show that the T2WI sequence outperforms the DCE-MRI sequence in radiomics models based on deep learning. Performance of the fusion model We first selected the best feature extraction models based on the AUC performance of the internal validation set. For traditional radiomics, we used ET to process a feature fusion model of intratumoral and 5mm peritumoral regions under DCE-MRI, obtaining 4788 features. For deep learning, we applied ET to a multi-slice feature fusion model under T2WI, extracting 24 features. These features underwent concatenation and fusion. We filtered stable features (ICC ≥ 0.80), normalized them using Z-score, and selected 192 features via t-test (p 0.9), we retained 48 features. LASSO regression further selected 10 key traditional radiomics features and 2 key deep learning features. We constructed three fusion prediction models using different machine learning algorithms. The ET-based model showed the best performance: training AUC 0.946 (95% CI: 0.911 - 0.981), internal validation AUC 0.885 (95% CI: 0.787 - 0.984), with accuracies of 0.887 and 0.783, sensitivities of 0.895 and 0.766, and specificities of 0.872 and 0.846. Complete performance metrics for all compared models are listed in Appendix 1. Interpretation of the Optimal Model Table 2 and Figure 5 compare the predictive performance of the best traditional radiomics, deep-learning, and fusion models using ROC, calibration curves, and DCA. The ET-based fusion model, combining habitat and peritumoral 5mm features from DCE-MRI and deep learning features from T2 sequences, achieved the highest AUC of 0.885 (95% CI: 0.787–0.984) in the internal validation set. This model also showed good generalization with an AUC of 0.839 (95% CI: 0.727–0.951), accuracy of 0.769, sensitivity of 0.786, and specificity of 0.750 in the external test set. SHAP analysis identified 12 key features, with 2 from T2WI deep learning and 7 from DCE-MRI habitat and peritumoral features, highlighting the value of multimodal fusion. Figure 5: Figure 5 compares the performance of the three best ET-based predictive models in both the training and internal validation cohorts. In the training set, (A) ROC curves showed that the ensemble model achieved the highest AUC of 0.946; (B) DCA curves indicated that the ensemble model delivered the greatest net benefit across a threshold probability range of 20–95%. In the internal validation set, (C) ROC curves again confirmed the ensemble model’s superiority with an AUC of 0.885; and (D) DCA curves demonstrated that the ensemble model provided the highest net benefit over a threshold probability range of 30–95%. Table 2: Comparison of the best-performing models from three distinct model types for Ki-67 status prediction Model_name ML AUC (95% CI) ACC SEN SPE PPV NPV F-1 score DCE-MRI (habitat+peri5mm) SVM Training 0.821(0.746 - 0.895) 0.77 0.75 0.81 0.89 0.61 0.81 Validation 0.817(0.711 - 0.923) 0.70 0.62 1.00 1.00 0.42 0.76 RF Training 0.856(0.784 - 0.928) 0.79 0.77 0.83 0.90 0.64 0.83 Validation 0.818(0.707 - 0.928) 0.73 0.70 0.85 0.94 0.44 0.81 ET Training 0.878(0.819 - 0.937) 0.81 0.79 0.85 0.92 0.67 0.85 Validation 0.825(0.708 - 0.942) 0.82 0.85 0.69 0.91 0.56 0.88 T2WI (2.5D_features_fusion) SVM Training 0.814(0.740 - 0.889) 0.75 0.78 0.70 0.84 0.61 0.81 Validation 0.776(0.624 - 0.928) 0.77 0.77 0.77 0.92 0.48 0.84 RF Training 0.772(0.688 - 0.856) 0.78 0.93 0.49 0.79 0.77 0.85 Validation 0.762(0.624 - 0.900) 0.70 0.66 0.85 0.94 0.41 0.78 ET Training 0.871(0.813 - 0.929) 0.79 0.75 0.87 0.92 0.63 0.83 Validation 0.804(0.641 - 0.966) 0.85 0.92 0.62 0.90 0.67 0.91 Rad_DL_features_fusion SVM Training 0.796(0.715 - 0.876) 0.75 0.76 0.72 0.85 0.60 0.80 Validation 0.781(0.640 - 0.922) 0.78 0.81 0.69 0.91 0.50 0.85 RF Training 0.869(0.803 - 0.935) 0.83 0.83 0.83 0.91 0.71 0.87 Validation 0.799(0.670 - 0.928) 0.70 0.64 0.92 0.97 0.41 0.77 ET Training 0.946(0.911 - 0.981) 0.89 0.90 0.87 0.93 0.80 0.91 Validation 0.885(0.787 - 0.984) 0.78 0.77 0.85 0.95 0.50 0.85 Note. —SVM, Support Vector Machine; RF, Random Forest; ET, Extreme Trees; AUC, the area under curve; ACC, accuracy; SEN, sensitivity; SPE, specificity; NPV, negative predictive value; PPV, positive predictive value; ML, machine learning. CI: confidence interval. Discussion In this study, traditional radiomics models outperformed deep learning models in predicting Ki-67 expression within the internal validation set. For T2WI, the best traditional radiomics model achieved an AUC of 0.814 (95% CI: 0.708–0.921), compared to 0.804 (95% CI: 0.641–0.966) for deep learning. For DCE-MRI, the AUCs were 0.825 (95% CI: 0.708–0.942) and 0.768 (95% CI: 0.616–0.919), respectively ( 18 , 19 ). This may be due to the direct pathophysiological relevance of hand-designed imaging features, which capture spatial heterogeneity of the tumor microenvironment linked to Ki-67 activity. Deep learning models might lose key spatial information during feature extraction, especially with limited data. Previous studies highlight the importance of peritumoral features in tumor behavior and treatment response. This study employed three machine learning algorithms to integrate radiomics features from varying tumor region sizes, revealing spatial heterogeneity characteristics. Models combining intratumoral and 5 mm peritumoral regions significantly outperformed those with 10 mm extensions. For instance, in T2WI, the best model integrating intratumoral and 5 mm peritumoral features achieved an AUC of 0.811 (95% CI: 0.680–0.942) versus 0.786 (95% CI: 0.652–0.921) for 10 mm peritumoral models. Wang et al. ( 20 ) reported that a 5 mm peritumoral model achieved an AUC of 0.820 (95% CI: 0.760–0.880), compared to an AUC of 0.798 (95% CI: 0.730–0.866) for a 10 mm peritumoral model. Additionally, Zhang et al.( 21 )found that 5 mm peritumoral models also outperformed 10 mm models in predicting HER2 status. These findings suggest that 5 mm peritumoral regions better capture the heterogeneous distribution of the tumor microenvironment, enhancing prediction accuracy for various tumor-related biological characteristics. In this study, a K-means clustering method optimized by C-H values was used for habitat analysis, identifying three subregions reflecting differences in tumor metabolism, blood flow, and cell density. The model based solely on intra-tumor habitat features achieved AUC values of 0.770 (95% CI: 0.621–0.919) for T2WI and 0.791 (95% CI: 0.667–0.914) for DCE-MRI on the internal validation set, outperforming the entire tumor internal ROI (AUC: 0.762 and 0.772). Ye et al. ( 22 ) found the habitat region model for predicting pCR in NSCLC had an AUC of 0.781 compared to 0.723 for the whole-tumor model. Similarly, Wang et al. ( 23 ) showed the habitat region model for HGSOC achieved an AUC of 0.808 versus 0.749 for the whole-tumor model. We systematically compared feature fusion and image fusion strategies for predicting Ki-67 expression in invasive breast cancer. Feature fusion models achieved a mean validation AUC of 0.791, a 10.8% improvement over image fusion’s 0.714. The training-validation gap decreased from 0.077 to 0.017, showcasing enhanced generalization. Specifically, for T2WI "intra-tumoral + 10 mm peritumoral," ET demonstrated superior performance: feature fusion achieved a validation AUC of 0.786 (95% CI: 0.652–0.921) and training AUC of 0.839 (95% CI: 0.771–0.906), while image fusion had a validation AUC of 0.650 (95% CI: 0.474–0.826) and training AUC of 0.831 (95% CI: 0.761–0.902), marking a 20.9% improvement. Overfitting reduced from 0.181 to 0.053. In DCE-MRI, ET achieved a validation AUC of 0.816 (95% CI: 0.653–0.978) with feature fusion, compared to 0.746 (95% CI: 0.584–0.909) with image fusion—a 9.4% improvement ( 24 , 25 ). Deep learning features, capturing tumor microstructure and high-dimensional nonlinear patterns, comprised two of the top nine key features, enhancing prediction robustness and accuracy. This aligns with Bhuiyan et al. ( 26 )'s glioma research, where T2-FLAIR deep learning features achieved 0.91 prediction accuracy for Ki-67. Traditional DCE-MRI radiomics features, such as volume ratios and texture, are crucial for assessing tumor invasiveness and treatment response, as shown by Xv et al. ( 27 ) in clear cell renal cell carcinoma studies. The integrated model achieved AUC values of 0.858 and 0.849 on internal and external validation sets, demonstrating clinical potential. This study had some limitations. Despite using a multicenter dataset, bias and patient heterogeneity persist, impacting model accuracy and generalization. Manual tumor segmentation, though precise, is time-consuming. Future work should focus on developing semi-automated or fully automated ROI delineation algorithms and integrating deep learning-based adaptive noise reduction to enhance accuracy. Deep learning models require extensive training data; thus, incorporating more patient imaging data is crucial for better generalization. The retrospective design limits causal inference: prospective studies are needed for stronger evidence. Standardizing image quality across multi-center MRI data with varied scanner settings remains challenging, potentially affecting radiomics feature consistency. In conclusion, we developed and validated an explainable machine learning model integrating traditional radiomics and deep learning features. This model accurately predicts Ki-67 expression in breast cancer patients, enhancing clinicians' understanding of its decision-making process and aiding in personalized treatment planning. Declarations Ethical Approval and Consent to participate This study was conducted in accordance with the Declaration of Helsinki and has been approved by the Ethics Committee of Longgang Central Hospital of Shenzhen (Shenzhen No.9 People's Hospital) (No.: 2025ECYJ087). A waiver of informed consent was granted due to the retrospective nature of the study and the use of de-identified data. Consent for publication Not applicable Funding This project is supported by the Shenzhen Science and Technology Program (JCYJ20240813114628038) and Guangdong Basic and Applied Basic Research Foundation (2024A1515220001) and Talent Development Program of Longgang Central Hospital of Shenzhen, the National Natural Science Foundation of China (82472062), the Natural Science Foundation of Guangdong Province of China (No.2024A1515011672). Medical Artificial Intelligence Clinical Application Research Project of Hospital Management Institute, National Health Commission of China (YLXX24AIA010) Author Contribution Y.W. and Z.S. contributed equally to this work. Y.W. and Z.S. led the project conception and design, developed and validated the methods, conducted experiments, collected data, performed data analysis and interpretation, created visualizations and charts, wrote and edited the manuscript, and contributed to reviewing and providing feedback. Z.L. provided resources and support, managed and coordinated the project, and contributed to reviewing and providing feedback. Y.X. conducted experiments, collected data, and contributed to data analysis and interpretation. M.L. conducted experiments, collected data, and contributed to reviewing and providing feedback. Y.Z. conducted experiments, collected data, and contributed to reviewing and providing feedback. L.Z. provided resources and support and contributed to reviewing and providing feedback. All authors contributed to reviewing and editing the manuscript. Acknowledgements Not applicable Data Availability The data generated or analyzed from cohorts A, B during the study are available from the corresponding author by request. References Arnold M, Morgan E, Rumgay H, et al. Current and future burden of breast cancer: Global statistics for 2020 and 2040. Breast. 2022;66:15–23. 10.1016/j.breast.2022.08.010 . Savva K-V, MacKenzie A, Coombes RC, Zhifang NM, Hanna BG, Peters CJ. An original study assessing biomarker success rate in breast cancer recurrence biomarker research. BMC Med. 2024;22(1):307. 10.1186/s12916-024-03460-6 . Inwald EC, Klinkhammer-Schalke M, Hofstädter F, et al. Ki-67 is a prognostic parameter in breast cancer patients: results of a large population-based cohort of a cancer registry. Breast Cancer Res Treat. 2013;139(2):539–52. 10.1007/s10549-013-2560-8 . Đokić S, Gazić B, Grčar Kuzmanov B, et al. Clinical and Analytical Validation of Two Methods for Ki-67 Scoring in Formalin Fixed and Paraffin Embedded Tissue Sections of Early Breast Cancer. Cancers (Basel). 2024;16(7):1405. 10.3390/cancers16071405 . Wang C, Chen J, Lv X, et al. Ki-67—Playing a key role in breast cancer but difficult to apply precisely in the real world. BMC Cancer. 2025;25(1):962. 10.1186/s12885-025-14374-8 . Lee J, Lee Y-J, Bae SJ, et al. Ki-67, 21-Gene Recurrence Score, Endocrine Resistance, and Survival in Patients With Breast Cancer. JAMA Netw Open. 2023;6(8):e2330961. 10.1001/jamanetworkopen.2023.30961 . Lin J-Y, Ye J-Y, Chen J-G, Lin S-T, Lin S, Cai S-Q. Prediction of Receptor Status in Radiomics: Recent Advances in Breast Cancer Research. Acad Radiol. 2024;31(7):3004–14. 10.1016/j.acra.2023.12.012 . Otikovs M, Nissan N, Furman-Haran E, et al. Relaxation-Diffusion T2-ADC Correlations in Breast Cancer Patients: A Spatiotemporally Encoded 3T MRI Assessment. Diagnostics (Basel). 2023;13(23):3516. 10.3390/diagnostics13233516 . Nie T, Feng M, Yang K, et al. Correlation between dynamic contrast-enhanced MRI characteristics and apparent diffusion coefficient with Ki-67-positive expression in non-mass enhancement of breast cancer. Sci Rep Nat Publishing Group. 2023;13(1):21451. 10.1038/s41598-023-48445-2 . Zhang Y-P, Zhang X-Y, Cheng Y-T, et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Military Med Res. 2023;10(1):22. 10.1186/s40779-023-00458-8 . Alabi RO, Elmusrati M, Leivo I, Almangush A, Mäkitie AA. Artificial Intelligence-Driven Radiomics in Head and Neck Cancer: Current Status and Future Prospects. Int J Med Inf. 2024;188:105464. 10.1016/j.ijmedinf.2024.105464 . Taghavi M, Staal F, Gomez Munoz F, et al. CT-Based Radiomics Analysis Before Thermal Ablation to Predict Local Tumor Progression for Colorectal Liver Metastases. Cardiovasc Intervent Radiol. 2021;44(6):913–20. 10.1007/s00270-020-02735-8 . Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6. 10.1016/j.ejca.2011.11.036 . Tan R, Sui C, Wang C, Zhu T. MRI-based intratumoral and peritumoral radiomics for preoperative prediction of glioma grade: a multicenter study. Front Oncol Front. 2024;14. 10.3389/fonc.2024.1401977 . Wang L, Wang C, Deng X, et al. Multimodal Ultrasound Radiomic Technology for Diagnosing Benign and Malignant Thyroid Nodules of Ti-Rads 4–5: A Multicenter Study. Sens (Basel). 2024;24(19):6203. 10.3390/s24196203 . Zuo D, Yang L, Jin Y, Qi H, Liu Y, Ren L. Machine learning-based models for the prediction of breast cancer recurrence risk. BMC Med Inf Decis Mak. 2023;23(1):276. 10.1186/s12911-023-02377-z . Liu J, He H, Wang Y et al. Predictive models for secondary epilepsy in patients with acute ischemic stroke within one year. eLife 13:RP98759. 10.7554/eLife.98759 Pei L, Han X, Ni C, Ke J. Prediction of prognosis in acute ischemic stroke after mechanical thrombectomy based on multimodal MRI radiomics and deep learning. Front Neurol. 2025;16:1587347. 10.3389/fneur.2025.1587347 . Du L, Yuan J, Gan M, et al. A comparative study between deep learning and radiomics models in grading liver tumors using hepatobiliary phase contrast-enhanced MR images. BMC Med Imaging. 2022;22(1):218. 10.1186/s12880-022-00946-8 . Wang F, Cheng M, Du B, et al. Predicting microvascular invasion in small (≤ 5 cm) hepatocellular carcinomas using radiomics-based peritumoral analysis. Insights Imaging. 2024;15(1):90. 10.1186/s13244-024-01649-0 . Zhang H, Miao Q, Fu Y, et al. Intratumoral and peritumoral radiomics based on automated breast volume scanner for predicting human epidermal growth factor receptor 2 status. Front Oncol. 2025;15:1556317. 10.3389/fonc.2025.1556317 . Ye G, Wu G, Zhang C, et al. CT-based quantification of intratumoral heterogeneity for predicting pathologic complete response to neoadjuvant immunochemotherapy in non-small cell lung cancer. Front Immunol. 2024;15:1414954. 10.3389/fimmu.2024.1414954 . Wang X, Xu C, Grzegorzek M, Sun H. Habitat radiomics analysis of pet/ct imaging in high-grade serous ovarian cancer: Application to Ki-67 status and progression-free survival. Front Physiol. 2022;13:948767. 10.3389/fphys.2022.948767 . R C, S Q, Y W, et al. Deep radiomic model based on the sphere-shell partition for predicting treatment response to chemotherapy in lung cancer. Translational Oncol Transl Oncol; 2023;35. 10.1016/j.tranon.2023.101719 Abdullakutty F, Akbari Y, Al-Maadeed S, Bouridane A, Talaat IM, Hamoudi R. Histopathology in focus: a review on explainable multi-modal approaches for breast cancer diagnosis. Front Med Front. 2024;11. 10.3389/fmed.2024.1450103 . Bhuiyan EH, Khan MM, Hossain SA, et al. Classification of glioma grade and Ki-67 level prediction in MRI data: A SHAP-driven interpretation. Comput Med Imaging Graph. 2025;124:102578. 10.1016/j.compmedimag.2025.102578 . Xv Y, Xiao B, Wei Z, et al. Interpretable CT Radiomics-based Machine Learning Model for Preoperative Prediction of Ki-67 Expression in Clear Cell Renal Cell Carcinoma. Acad Radiol. 2025;32(5):2739–50. 10.1016/j.acra.2024.11.072 . Additional Declarations No competing interests reported. Supplementary Files Appendix1.docx Cite Share Download PDF Status: Published Journal Publication published 16 Jan, 2026 Read the published version in BMC Medical Imaging → Version 1 posted Editorial decision: Revision requested 09 Dec, 2025 Reviews received at journal 08 Dec, 2025 Reviewers agreed at journal 27 Nov, 2025 Reviews received at journal 27 Nov, 2025 Reviewers agreed at journal 26 Nov, 2025 Reviewers agreed at journal 21 Oct, 2025 Reviewers agreed at journal 19 Oct, 2025 Reviewers invited by journal 10 Oct, 2025 Editor invited by journal 01 Oct, 2025 Editor assigned by journal 30 Sep, 2025 Submission checks completed at journal 30 Sep, 2025 First submitted to journal 25 Sep, 2025 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-7709859","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":532867803,"identity":"77b1c0f5-a8e0-41e5-9323-83db477677f7","order_by":0,"name":"Yuqian Wang","email":"","orcid":"","institution":"Guangzhou University of Chinese Medicine, Shenzhen Clinical Medical College","correspondingAuthor":false,"prefix":"","firstName":"Yuqian","middleName":"","lastName":"Wang","suffix":""},{"id":532867805,"identity":"2f7c4924-742b-4c1f-9c29-ea2092680cf2","order_by":1,"name":"Yue Zhang","email":"","orcid":"","institution":"Guangzhou University of Chinese Medicine, Shenzhen Clinical Medical College","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Zhang","suffix":""},{"id":532867806,"identity":"4e67cf5b-997a-47e4-b81c-e80119de1871","order_by":2,"name":"Zaiyi Liu","email":"","orcid":"","institution":"Department of Radiology, Guangdong Provincial People’s Hospital (Guangdong Academy of Medical Sciences), Southern Medical University, Guangzhou 510080, China","correspondingAuthor":false,"prefix":"","firstName":"Zaiyi","middleName":"","lastName":"Liu","suffix":""},{"id":532867807,"identity":"5e0aba36-c9d6-4a01-a7cb-ea2ab157bf48","order_by":3,"name":"Yiming Xiong","email":"","orcid":"","institution":"Longgang District Maternity \u0026 Child Healthcare Hospital of Shenzhen City, Longgang Maternity and Child Institute of Shantou University Medical College","correspondingAuthor":false,"prefix":"","firstName":"Yiming","middleName":"","lastName":"Xiong","suffix":""},{"id":532867808,"identity":"94bad3af-193a-4eb6-a38f-d2df97d76143","order_by":4,"name":"Mifang Li","email":"","orcid":"","institution":"Longgang Central Hospital of Shenzhen (Longgang Clinical Institute of Shantou University Medical College)","correspondingAuthor":false,"prefix":"","firstName":"Mifang","middleName":"","lastName":"Li","suffix":""},{"id":532867809,"identity":"7b451d14-2b37-4d36-98db-9a300ec3b5fb","order_by":5,"name":"Lingyan Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2klEQVRIie2RsQrCMBCGA4G4RLOeIPUVAoXi4yQITiqOGUSESjtYdwUfwtGxpZAp7h11d6ibo9FVMXVzyDffx3//HUIezx9CWnGe1woCgtnlLNTcrXSolsXWDELWyjA/G+1WAhiHZTtRcpcZ0r2scIPFqBYlNRDyaqKVXBLE0rVwdin2tguvpqNKHnsIzOngTMmvrxQRVdIQxGHiUGDM83YC8mCVmUxwM6V4KrZ+hJopdjF7ZLBHToYgjKbOLv00LutaLewrcXG7q3nA0s135Q3627jH4/F4PvIAXD9RXKCjX5UAAAAASUVORK5CYII=","orcid":"","institution":"Guangzhou University of Chinese Medicine, Shenzhen Clinical Medical College","correspondingAuthor":true,"prefix":"","firstName":"Lingyan","middleName":"","lastName":"Zhang","suffix":""},{"id":532867810,"identity":"84999319-78f9-441a-bf8d-3c4498725ff1","order_by":6,"name":"Zhenwei Shi","email":"","orcid":"","institution":"Department of Radiology, Guangdong Provincial People’s Hospital (Guangdong Academy of Medical Sciences), Southern Medical University, Guangzhou 510080, China","correspondingAuthor":false,"prefix":"","firstName":"Zhenwei","middleName":"","lastName":"Shi","suffix":""}],"badges":[],"createdAt":"2025-09-25 07:23:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7709859/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7709859/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12880-026-02151-3","type":"published","date":"2026-01-16T16:31:10+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":94410015,"identity":"dd8cae7b-4d25-4bf2-b58c-b325678825a5","added_by":"auto","created_at":"2025-10-27 14:04:24","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1492965,"visible":true,"origin":"","legend":"","description":"","filename":"MultiparametricMRIbasedHabitatAnalysisIntegratingDeepLearningandRadiomicsforPredictingPreoperativeKi67ExpressionLevelinBreastCancer.docx","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/4e2be6ac695dd90f7529a317.docx"},{"id":94408253,"identity":"c2ca47ab-48ca-4654-88b0-4a6a5c4aac7f","added_by":"auto","created_at":"2025-10-27 14:03:31","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8878,"visible":true,"origin":"","legend":"","description":"","filename":"421f1696e3544d458bf3890e25fcc6e1.json","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/29ffac9b345a2a76d6d95205.json"},{"id":94408907,"identity":"7ef37e43-8d5d-4a87-bbe5-7b538e862245","added_by":"auto","created_at":"2025-10-27 14:03:55","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":55268,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/ac39327d700612b2c8046380.docx"},{"id":94408236,"identity":"a9a1e6d2-dfa7-491c-adcf-fd2b443393ee","added_by":"auto","created_at":"2025-10-27 14:03:30","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139798,"visible":true,"origin":"","legend":"","description":"","filename":"421f1696e3544d458bf3890e25fcc6e11enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/19e51901f5c34068364774ae.xml"},{"id":94410022,"identity":"14bcef80-ba41-4655-8235-25e7e15e4cef","added_by":"auto","created_at":"2025-10-27 14:04:24","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94591,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/9912997c62139bfa12685f46.png"},{"id":94410133,"identity":"3a6c3676-66b3-4f36-b257-102adaff0f58","added_by":"auto","created_at":"2025-10-27 14:04:32","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96110,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/ceaddd3ec8a4f1be63435937.png"},{"id":94408828,"identity":"120f0569-ee3c-4e90-a9d9-b540dcc55aa3","added_by":"auto","created_at":"2025-10-27 14:03:51","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":52731,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/63f35865031c9e254217eaef.png"},{"id":94409817,"identity":"d3f10e0a-b3c0-4343-a6b2-2d86206fa095","added_by":"auto","created_at":"2025-10-27 14:04:18","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":74760,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/2bfd900e32084821e9cb8710.png"},{"id":94410083,"identity":"0db1b118-79a3-434e-ada0-32ceecc3222d","added_by":"auto","created_at":"2025-10-27 14:04:28","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57078,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/ea3f68831a8c002c55b04ecc.png"},{"id":94410265,"identity":"fa23d386-ef32-4446-a0e4-381283c8f404","added_by":"auto","created_at":"2025-10-27 14:04:37","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26137,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/3da79ec2fb1eefe9b0237979.png"},{"id":94408561,"identity":"ebd1a0f7-d8e7-49b3-97f8-bcc424638ac6","added_by":"auto","created_at":"2025-10-27 14:03:39","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":20794,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/bb883a6b5df4f83ea45897c7.png"},{"id":94408659,"identity":"72b42783-09ba-4bc3-8d1e-0ef22a3348b0","added_by":"auto","created_at":"2025-10-27 14:03:44","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21367,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/bf5e7166d603324d29e3f508.png"},{"id":94409646,"identity":"4caf10d7-8856-4681-98b5-9a690d9856e9","added_by":"auto","created_at":"2025-10-27 14:04:14","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":135667,"visible":true,"origin":"","legend":"","description":"","filename":"421f1696e3544d458bf3890e25fcc6e11structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/73ebcb534d2610d5b9e3eb6a.xml"},{"id":94407789,"identity":"3dc9c370-5c78-4441-a4ab-f1b9aa3b02b7","added_by":"auto","created_at":"2025-10-27 14:03:11","extension":"html","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":147628,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/71b1b3d79ca021778224a481.html"},{"id":94410017,"identity":"57371f6b-f1ad-4fbb-8769-7a99414342fb","added_by":"auto","created_at":"2025-10-27 14:04:24","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":244569,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart shows patient exclusion for dataset A (_) and dataset B（_）\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/12df8ef5cdafbb294980f90e.png"},{"id":94409481,"identity":"7e7c79a7-fac5-4f1d-b9f3-a238a2207239","added_by":"auto","created_at":"2025-10-27 14:04:09","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":492370,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow schematic for constructing the traditional radiomics, 2.5D deep learning, and fusion models proposed in this study. First, after manual tumor delineation on T2WI and DCE-MRI, peritumoral rims, habitat subregions, and 2.5D deep-learning slices are systematically generated to create distinct image-fusion and feature-fusion target volumes. Second, corresponding features are extracted for each model and region. Third, feature selection: Robust filtering retains the most predictive features. Finally, traditional radiomics, 2.5D deep-learning, and fusion models—integrating the best features from both—are built and systematically assessed on independent validation sets.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/85d01a402b13c90a18ea9eea.png"},{"id":94409254,"identity":"8d437a86-f654-4c9f-8644-e9a1c9af4a5d","added_by":"auto","created_at":"2025-10-27 14:04:03","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":323363,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap of AUC values for models with different feature types, sequences, regions, and fusion methods. AUC stands for Area Under the Curve, SVM for Support Vector Machine, RF for Random Forest, ET for Extreme Trees, Rad for traditional radiomics models, DL for 2.5D deep learning models, T2WI for T2-weighted imaging sequences, and DCE-MRI for Dynamic Contrast-Enhanced MRI sequences, I for intra-tumoral，P5 for peritumoral5mm, P10 for peritumoral10mm,H for habitat，MC for multi-channel, MS for multi-slice,”*” for image fusion.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/58e5ab69e0d98a3c53ec9ef9.png"},{"id":94410247,"identity":"4db78118-e9ed-47de-b523-4869d70b9861","added_by":"auto","created_at":"2025-10-27 14:04:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":147496,"visible":true,"origin":"","legend":"\u003cp\u003e(A) Calinski–Harabasz index curve: the score decreases monotonically with increasing cluster numbers, peaking at k = 3, indicating the optimal cluster number is three. (B) Habitat visualization: the Habitat ROIs from each sequence's habitat are mapped to the corresponding three Grad-CAM heatmaps above.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/722d01ae819514f2ae5e5925.png"},{"id":94409161,"identity":"d48c716b-127f-4d42-b1a5-34b088cfbf81","added_by":"auto","created_at":"2025-10-27 14:04:01","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":91941,"visible":true,"origin":"","legend":"\u003cp\u003ecompares the performance of the three best ET-based predictive models in both the training and internal validation cohorts. In the training set, (A) ROC curves showed that the ensemble model achieved the highest AUC of 0.946; (B) DCA curves indicated that the ensemble model delivered the greatest net benefit across a threshold probability range of 20–95%. In the internal validation set, (C) ROC curves again confirmed the ensemble model’s superiority with an AUC of 0.885; and (D) DCA curves demonstrated that the ensemble model provided the highest net benefit over a threshold probability range of 30–95%.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/2426a3230a0cf3bf3ea5b999.jpg"},{"id":100614875,"identity":"ccd02677-ccdf-4f01-b44a-37bc33a6e27e","added_by":"auto","created_at":"2026-01-19 17:27:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2443317,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/ebb687eb-f4e7-40ca-a7de-49bae7bf7c50.pdf"},{"id":94408871,"identity":"fba9b2ad-41fd-406e-9604-b7a2570464cd","added_by":"auto","created_at":"2025-10-27 14:03:54","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":55268,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7709859/v1/b8ba8fd4dce11684b0d486fd.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multiparametric MRI-based Habitat Analysis Integrating Deep Learning and Radiomics for Predicting Preoperative Ki-67 Expression Level in Breast Cancer","fulltext":[{"header":"Introduction","content":"\u003cp\u003eBreast cancer is the most common and highly malignant tumor in women worldwide(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). Accurate risk-stratification and timely treatment are essential for improving prognosis. The routinely used IHC biomarkers include Progesterone Receptor (PR), Estrogen Receptor (ER), Human Epidermal Growth Factor Receptor 2 (HER-2), and Ki-67. Among them, the proliferation index Ki-67 has been demonstrated to correlate strongly with tumor invasiveness(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e), early metastasis(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e), recurrence rate (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e)and overall survival(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). However, Ki-67 assessment currently relies on core-needle or surgical biopsy followed by immunohistochemistry (IHC), an invasive procedure that is painful, may delay therapy, and occasionally provokes adverse reactions(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). Consequently, there is an MRI has become the preferred imaging modality for breast-cancer evaluation. T2-weighted imaging (T2WI) provides high-resolution anatomical information and reflects the tumor microenvironment, whereas dynamic contrast-enhanced MRI (DCE-MRI) reveals detailed hemodynamic characteristics(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Nevertheless, conventional qualitative interpretation fails to exploit the rich quantitative information latent in these images, particularly the quantitative features related to the Ki67 proliferation index(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eRadiomics can extract high-throughput imaging features to build predictive models, yet traditional whole-tumor radiomics analyses ignore intratumoral heterogeneity (ITH)(\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Habitat analysis partitions tumors into biologically distinct subregions (\u0026ldquo;habitats\u0026rdquo;), thereby explicitly quantifying ITH and improving model Although fusion of radiomics and deep learning is promising, the relative merits of image-level versus feature-level fusion for Ki-67 prediction remain unexplored(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). Furthermore, model interpretability remains a major barrier to clinical translation; the SHAP algorithm has demonstrated utility in identifying key features and explaining single-sample predictions across multiple medical scenarios(\u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTherefore, the aim of this study was twofold: (i) to develop habitat-based radiomics and convolutional neural network (CNN) models, and (ii) to systematically compare their efficacy\u0026mdash;alongside image- and feature-level fusion strategies\u0026mdash;for the non-invasive prediction of Ki-67 expression in breast cancer.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy Sample\u003c/h2\u003e\u003cp\u003eWe collected clinical data and MRI images of 320 patients with histologically confirmed breast cancer from two centers. Detailed MRI acquisition parameters and sample size estimation based on EPV are provided in Appendix 1. After applying inclusion and exclusion criteria, patients from Center A were randomly assigned to the training set (n\u0026thinsp;=\u0026thinsp;142) and internal validation set (n\u0026thinsp;=\u0026thinsp;60) in a 7:3 ratio, while patients from Center B constituted the external test set (n\u0026thinsp;=\u0026thinsp;52). The patient recruitment process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Ethics approval was waived by the IRBs of both centers for this minimal-risk, retrospective analysis of de-identified data.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eClinical and MRI imaging characteristics\u003c/h3\u003e\n\u003cp\u003eThis study included clinical characteristics crucial for prognosis. Ki-67, defined by positive nuclear staining percentage, uses 20% as a threshold to distinguish prognosis groups. Two blinded radiologists assessed imaging features per BI-RADS guidelines; disagreements were resolved by a senior radiologist.\u003c/p\u003e\n\u003ch3\u003eImage acquisition and mask segmentation\u003c/h3\u003e\n\u003cp\u003eAll patients underwent MRI scans with uniform settings. The scans covered the entire breast, using FS-T2WI and DCE-MRI sequences. Images were preprocessed by resampling to 1 \u0026times; 1 \u0026times; 1 mm resolution, with N4 bias field correction and normalization. Two junior radiologists manually outlined tumor boundaries on the sequences with the most obvious T2WI and DCE-MRI using ITK-SNAP 3.6.0 software to generate ROIs. ICC assessed intra - and inter-observer agreement; features were stable and reliable when ROI ICC values were above 0.80. A senior radiologist reviewed each ROI. Radiologists were blinded to clinical information during segmentation and review. The research process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eConstruction of traditional radiomics models\u003c/h3\u003e\n\u003cp\u003eWe expanded the ROI peritumorally by 5 mm and 10 mm using the Onekey AI platform's Mask Filling Toolkit and merged intratumoral and peritumoral regions to form a new ROI for image fusion analysis. We used an unsupervised K-means clustering method (K\u0026thinsp;=\u0026thinsp;3\u0026ndash;10) to generate subregions in the tumor endosomes, determining the optimal number of clusters (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) via the Calinski-Harabasz (CH) value. Pyradiomics (v3.0.1) extracted 1197 image histologic features from each ROI of T2WI and DCE-MRI sequences. We designed two fusion strategies (image fusion and feature fusion) and performed multi-stage feature screening (ICC\u0026thinsp;\u0026ge;\u0026thinsp;0.80, Z-score standardization, t-test, Pearson correlation coefficient, LASSO regression) to reduce overfitting. Finally, we used grid search with 5-fold cross-validation to construct 48 prediction models based on SVM, RF, and ET algorithms.\u003c/p\u003e\n\u003ch3\u003eConstruction of 2.5D Deep Learning Model\u003c/h3\u003e\n\u003cp\u003eWe extracted 2D ROIs from the largest tumor cross-section and adjacent sections in T2WI and DCE sequences. After cropping and Z-score normalization, tumor regions from three sections were combined into a multichannel image. Using a pre-trained ResNet-50 CNN, we extracted deep features, reducing them from 2048 to 8 via PCA to minimize overfitting. We designed two fusion strategies and built 12 predictive models using three machine learning algorithms. Grad-CAM was used to generate heat maps for model interpretation.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eConstruction of the combined model\u003c/h2\u003e\u003cp\u003eFirst, we screened the model features with the best performance of AUC values from the traditional radiomics model and 2.5D deep learning model for feature splicing. The fused features were processed through the same standardized processing and feature selection as traditional radiomics models and deep learning models. Three combined prediction models were constructed using three machine learning algorithms. This staged feature fusion and multi-algorithm validation set framework retains the interpretability of traditional radiomics features, integrates deep image features extracted by deep learning, and ensures model stability through cross-validation.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eExternal Verification and Interpretability\u003c/h3\u003e\n\u003cp\u003eWe compared the predictive performance of the best traditional radiomics model, the best deep learning model, and the best fusion model on the internal validation set and identified the optimal predictive model. Subsequently, we evaluated the performance of this optimal model on an external validation set. To enhance the interpretability of the best model, we applied the SHAP algorithm, which decomposes the contribution of each radiomic feature to the model\u0026rsquo;s prediction, allowing us to clearly understand the individual impact of each feature.\u003c/p\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eStatistical analysis used Python 3.9.13. Normally distributed continuous variables were described as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD and compared via t-test; non-normally distributed variables as median (quartiles) and compared via Man-Whitney U test. Categorical variables were expressed as counts/percentages and compared using chi-square or Fisher's test. Variables with P\u0026thinsp;\u0026lt;\u0026thinsp;0.05 in univariate analyses were included in multivariate analyses, with P\u0026thinsp;\u0026lt;\u0026thinsp;0.05 indicating significance. Model performance was assessed using AUC, 95% CI, sensitivity, specificity, F1 score, calibration curves, and decision curve analysis (DCA) for clinical applicability.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003ePatient Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCenter A identified 202 cases and was randomized in a 7:3 ratio into a training s (N=142) and an internal validation set (N=60). The sample from Center B was used as the external test set (N=52) and clinical baseline data were compared between the training, internal validation and external test sets. The Clinical and MRI Imaging characteristics of all patients are listed in Table 1. In the training and internal validation cohort, PR, ER, and HER-2 status showed significant differences between the high Ki-67 and low Ki-67 groups (P \u0026lt; 0.05). In the external cohort, significant differences were observed in PR and HER-2 status (P \u0026lt; 0.05). However, no significant differences were found in age, tumor diameter, menopausal status, T stage, TIC, BPE, BI-RADS, FGT, tumor internal enhancement, mrALN, pathological LNM, and LVI between the high Ki-67 and low Ki-67 groups in any of the three cohorts (P \u0026gt; 0.05).\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"12\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 1: Comparison of Clinicopathologic Characteristics between the Training and External Validation Data Sets\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\"\u003e\n \u003cp\u003eTraining cohort\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\"\u003e\n \u003cp\u003eValidation cohort\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\"\u003e\n \u003cp\u003eTest cohort\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCharacteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh Ki-67 (\u003cem\u003en\u003c/em\u003e=95)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLow Ki-67 (\u003cem\u003en\u003c/em\u003e=47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e Value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh Ki-67 (\u003cem\u003en\u003c/em\u003e=47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLow Ki-67 (\u003cem\u003en\u003c/em\u003e=13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e Value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh Ki-67 (\u003cem\u003en\u003c/em\u003e=27)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLow Ki-67 (\u003cem\u003en\u003c/em\u003e=24)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003ep Value\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAge (mean\u0026plusmn;SD)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e45.60\u0026plusmn;9.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e47.53\u0026plusmn;9.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.31\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e46.13\u0026plusmn;9.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48.23\u0026plusmn;9.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.465\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e44.29\u0026plusmn;8.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e49.26\u0026plusmn;10.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.062\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMaximum_tumor_diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e34.26\u0026plusmn;20.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e37.36\u0026plusmn;24.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.45\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e47.13\u0026plusmn;22.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e40.62\u0026plusmn;23.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.29\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e28.00\u0026plusmn;11.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23.61\u0026plusmn;6.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.56\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMenopausal_status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e69 (72.63)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35 (74.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.98\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (40.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (61.54)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.30\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24 (85.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (69.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.292\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e26 (27.37)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12 (25.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e28 (59.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5 (38.46)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (14.29)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (30.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTumor_location\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLeft\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48 (50.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e20 (42.55)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.47\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25 (53.19)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (69.23)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.47\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (57.14)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13 (56.52)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e47 (49.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e27 (57.45)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22 (46.81)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (30.77)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12 (42.86)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (43.48)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ecT_stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅠ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (10.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (8.51)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.41\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (12.77)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (7.69)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.806\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (28.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (17.39)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.35\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅡ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e40 (42.11)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (40.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (40.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (46.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (67.86)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (69.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅢ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e37 (38.95)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23 (48.94)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15 (31.91)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3 (23.08)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (3.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3 (13.04)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅣ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (8.42)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (14.89)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3 (23.08)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTIC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅠ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.361\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (3.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.654\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅡ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (20.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (19.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e26 (55.32)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (30.77)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (32.14)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (34.78)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eⅢ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e76 (80.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e37 (78.72)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21 (44.68)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (69.23)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18 (64.29)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15 (65.22)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBPE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e47 (49.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e30 (63.83)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.18\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24 (51.06)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (69.23)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.188\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (3.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003enull\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.29\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35 (36.84)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (21.28)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21 (44.68)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2 (15.38)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14 (50.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17 (73.91)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11 (11.58)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (14.89)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (7.69)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12 (42.86)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5 (21.74)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2 (2.11)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (7.69)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (3.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (4.35)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBI_RADS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.24\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (7.69)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.72\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14 (14.74)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (17.02)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (12.77)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (7.69)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17 (60.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15 (65.22)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e42 (44.21)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25 (53.19)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (40.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (46.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (35.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (30.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e39 (41.05)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13 (27.66)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21 (44.68)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5 (38.46)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (3.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (4.35)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eFGT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (1.05)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1 (2.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.16\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.19\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2 (2.11)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3 (6.38)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (8.51)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86 (90.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e36 (76.60)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e37 (78.72)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13 (100.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e28 (100.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23 (100.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (6.32)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (14.89)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (12.77)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTumor_internal_enhancement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e47 (49.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25 (53.19)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18 (38.30)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (53.85)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.49\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17 (60.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18 (78.26)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.30\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48 (50.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22 (46.81)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e29 (61.70)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (46.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11 (39.29)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5 (21.74)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003emrALN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25 (26.32)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e20 (42.55)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.08\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12 (25.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5 (38.46)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15 (53.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (69.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.38\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e70 (73.68)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e27 (57.45)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35 (74.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (61.54)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13 (46.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (30.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLNM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e36 (37.89)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21 (44.68)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.55\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16 (34.04)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (61.54)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.14\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (35.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14 (60.87)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.13\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e59 (62.11)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e26 (55.32)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e31 (65.96)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5 (38.46)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18 (64.29)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (39.13)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLVI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e46 (48.42)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25 (53.19)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.72\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35 (74.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (53.85)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.27\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24 (85.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21 (91.30)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e49 (51.58)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22 (46.81)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12 (25.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (46.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (14.29)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2 (8.70)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePR status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e45 (47.37)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (21.28)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.01\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (17.02)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (53.85)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.02\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (21.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13 (56.52)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.02\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e50 (52.63)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e37 (78.72)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e39 (82.98)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (46.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22 (78.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10 (43.48)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eER status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35 (36.84)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8 (17.02)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (19.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (53.85)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7 (25.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (17.39)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.753\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e60 (63.16)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e39 (82.98)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e38 (80.85)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (46.15)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e21 (75.00)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (82.61)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHER2 status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e47 (49.47)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e33 (70.21)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e19 (40.43)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9 (69.23)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.13\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11 (39.29)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17 (73.91)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48 (50.53)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14 (29.79)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e28 (59.57)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4 (30.77)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17 (60.71)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6 (26.09)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003ePerformance of traditional radiomics models\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe compared 48 radiomics models built from different sequences, regions, and fusion strategies in Figure 3. Habitat analysis identified three optimal intratumoral habitats (k = 3) and Grad-CAM mapped ROIs to corresponding heatmaps in Figure 4. Feature fusion strategies outperformed image fusion strategies. In DCE-MRI, the intratumoral habitat feature fusion model using ET achieved an AUC of 0.791 (95% CI: 0.667 - 0.914) on internal validation, surpassing the intratumoral ROI model\u0026apos;s AUC of 0.772 (95% CI: 0.624 - 0.919). In T2W sequences, the feature fusion model combining intratumoral and 5-mm peritumoral regions achieved an AUC of 0.811 (95% CI: 0.680 - 0.942), higher than the 0.708 (95% CI: 0.564 - 0.852) from image-level fusion. DCE-MRI models consistently outperformed T2WI models. For example, the intratumoral model alone achieved an AUC of 0.772 (95% CI: 0.624 - 0.919) for DCE-MRI versus 0.762 (95% CI: 0.616 - 0.908) for T2WI. In peritumoral analysis, the 5-mm peritumoral zone outperformed the 10-mm extension. The T2WI-based habitat model using ET and integrating 5-mm peritumoral features achieved an AUC of 0.814 (95% CI: 0.708 - 0.921), surpassing the 0.804 (95% CI: 0.679 - 0.930) with 10-mm extension. The ET-based DCE-MRI fusion radiomics model integrating tumor habitat and 5 mm peritumoral features showed excellent performance: AUC of 0.878 (95% CI: 0.819 - 0.937), accuracy 0.810, sensitivity 0.789, specificity 0.851 in training; AUC 0.825 (95% CI: 0.708 - 0.942), accuracy 0.817, sensitivity 0.851, specificity 0.692 in internal validation. These results confirm the superiority of feature fusion and identify 5 mm as the optimal peritumoral extension distance.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 3\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003eHeatmap of AUC values for models with different feature types, sequences, regions, and fusion methods. AUC stands for Area Under the Curve, SVM for Support Vector Machine, RF for Random Forest, ET for Extreme Trees, Rad for traditional radiomics models, DL for 2.5D deep learning models, T2WI for T2-weighted imaging sequences, and DCE-MRI for Dynamic Contrast-Enhanced MRI sequences, I for intra-tumoral，P5 for peritumoral5mm, P10 for peritumoral10mm,H for habitat，MC for multi-channel, MS for multi-slice,\u0026rdquo;*\u0026rdquo; for image fusion.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 4:\u003c/strong\u003e (A) Calinski\u0026ndash;Harabasz index curve: the score decreases monotonically with increasing cluster numbers, peaking at k = 3, indicating the optimal cluster number is three. (B) Habitat visualization: the Habitat ROIs from each sequence\u0026apos;s habitat are mapped to the corresponding three Grad-CAM heatmaps above.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePerformance of deep learning models\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe compared 12 deep learning models built using different sequences and fusion strategies in Figure 3. The multi-slice feature fusion strategy showed significant advantages. For example, under the DCE sequence of the ET, multi-slice feature fusion achieved an AUC of 0.768 (95% CI: 0.616 - 0.919) in the internal test set, higher than the 0.718 (95% CI: 0.557 - 0.879) from multi-channel fusion. In sequence comparison, the T2WI sequence models showed better discriminative performance. Specifically, using a multi-channel image fusion strategy based on SVM, the T2WI sequence achieved an AUC of 0.763 (95% CI: 0.610 - 0.915) in the internal test set, while the DCE-MRI sequence only reached 0.699 (95% CI: 0.548 - 0.850). Notably, the radiomics model based on the T2WI sequence, integrating the feature fusion algorithm of the maximum tumor multi-slice and its two adjacent slices, performed the best overall. In the training cohort, this model achieved an AUC of 0.871 (95% CI: 0.813 - 0.929), and in the internal validation set, an AUC of 0.804 (95% CI: 0.641 - 0.966). The model also showed accuracy of 0.789 and 0.850, sensitivities of 0.747 and 0.915, and specificities of 0.872 and 0.615 in the training and test sets respectively. These results confirm the superiority of the multi-slice feature fusion strategy over multi-channel image fusion and show that the T2WI sequence outperforms the DCE-MRI sequence in radiomics models based on deep learning.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePerformance of the fusion model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe first selected the best feature extraction models based on the AUC performance of the internal validation set. For traditional radiomics, we used ET to process a feature fusion model of intratumoral and 5mm peritumoral regions under DCE-MRI, obtaining 4788 features. For deep learning, we applied ET to a multi-slice feature fusion model under T2WI, extracting 24 features. These features underwent concatenation and fusion. We filtered stable features (ICC \u0026ge; 0.80), normalized them using Z-score, and selected 192 features via t-test (p \u0026lt; 0.05). After removing redundant features (r \u0026gt; 0.9), we retained 48 features. LASSO regression further selected 10 key traditional radiomics features and 2 key deep learning features. We constructed three fusion prediction models using different machine learning algorithms. The ET-based model showed the best performance: training AUC 0.946 (95% CI: 0.911 - 0.981), internal validation AUC 0.885 (95% CI: 0.787 - 0.984), with accuracies of 0.887 and 0.783, sensitivities of 0.895 and 0.766, and specificities of 0.872 and 0.846. Complete performance metrics for all compared models are listed in Appendix 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInterpretation of the Optimal Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable 2 and Figure 5 compare the predictive performance of the best traditional radiomics, deep-learning, and fusion models using ROC, calibration curves, and DCA. The ET-based fusion model, combining habitat and peritumoral 5mm features from DCE-MRI and deep learning features from T2 sequences, achieved the highest AUC of 0.885 (95% CI: 0.787\u0026ndash;0.984) in the internal validation set. This model also showed good generalization with an AUC of 0.839 (95% CI: 0.727\u0026ndash;0.951), accuracy of 0.769, sensitivity of 0.786, and specificity of 0.750 in the external test set. SHAP analysis identified 12 key features, with 2 from T2WI deep learning and 7 from DCE-MRI habitat and peritumoral features, highlighting the value of multimodal fusion.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 5:\u0026nbsp;\u003c/strong\u003eFigure 5 compares the performance of the three best ET-based predictive models in both the training and internal validation cohorts. In the training set, (A) ROC curves showed that the ensemble model achieved the highest AUC of 0.946; (B) DCA curves indicated that the ensemble model delivered the greatest net benefit across a threshold probability range of 20\u0026ndash;95%. In the internal validation set, (C) ROC curves again confirmed the ensemble model\u0026rsquo;s superiority with an AUC of 0.885; and (D) DCA curves demonstrated that the ensemble model provided the highest net benefit over a threshold probability range of 30\u0026ndash;95%.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" class=\"fr-table-selection-hover\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 2: Comparison of the best-performing models from three distinct model types for Ki-67 status prediction\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel_name\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eML\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAUC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eF-1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDCE-MRI (habitat+peri5mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.821(0.746 - 0.895)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.75\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.89\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.61\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.817(0.711 - 0.923)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.70\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.62\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.42\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.76\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.856(0.784 - 0.928)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.79\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.83\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.90\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.64\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.83\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.818(0.707 - 0.928)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.73\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.70\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.94\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.44\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eET\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.878(0.819 - 0.937)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.79\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.67\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.825(0.708 - 0.942)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.82\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.69\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.56\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.88\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eT2WI (2.5D_features_fusion)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.814(0.740 - 0.889)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.75\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.78\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.70\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.84\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.61\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.776(0.624 - 0.928)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.48\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.84\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.772(0.688 - 0.856)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.78\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.93\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.49\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.79\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.762(0.624 - 0.900)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.70\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.66\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.94\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.41\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.78\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eET\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.871(0.813 - 0.929)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.79\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.75\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.87\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.63\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.83\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.804(0.641 - 0.966)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.62\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.90\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.67\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRad_DL_features_fusion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.796(0.715 - 0.876)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.75\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.76\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.72\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.60\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.80\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.781(0.640 - 0.922)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.78\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.69\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.50\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.869(0.803 - 0.935)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.83\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.83\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.83\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.71\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.87\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.799(0.670 - 0.928)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.70\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.64\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.97\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.41\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eET\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.946(0.911 - 0.981)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.89\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.90\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.87\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.93\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.80\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.885(0.787 - 0.984)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.78\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.77\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.95\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.50\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.85\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\" valign=\"top\"\u003e\n \u003cp\u003eNote. \u0026mdash;SVM, Support Vector Machine; RF, Random Forest; ET, Extreme Trees; AUC, the area under curve; ACC, accuracy; SEN, sensitivity; SPE, specificity; NPV, negative predictive value; PPV, positive predictive value; ML, machine learning. CI: confidence interval.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, traditional radiomics models outperformed deep learning models in predicting Ki-67 expression within the internal validation set. For T2WI, the best traditional radiomics model achieved an AUC of 0.814 (95% CI: 0.708\u0026ndash;0.921), compared to 0.804 (95% CI: 0.641\u0026ndash;0.966) for deep learning. For DCE-MRI, the AUCs were 0.825 (95% CI: 0.708\u0026ndash;0.942) and 0.768 (95% CI: 0.616\u0026ndash;0.919), respectively (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). This may be due to the direct pathophysiological relevance of hand-designed imaging features, which capture spatial heterogeneity of the tumor microenvironment linked to Ki-67 activity. Deep learning models might lose key spatial information during feature extraction, especially with limited data.\u003c/p\u003e\u003cp\u003ePrevious studies highlight the importance of peritumoral features in tumor behavior and treatment response. This study employed three machine learning algorithms to integrate radiomics features from varying tumor region sizes, revealing spatial heterogeneity characteristics. Models combining intratumoral and 5 mm peritumoral regions significantly outperformed those with 10 mm extensions. For instance, in T2WI, the best model integrating intratumoral and 5 mm peritumoral features achieved an AUC of 0.811 (95% CI: 0.680\u0026ndash;0.942) versus 0.786 (95% CI: 0.652\u0026ndash;0.921) for 10 mm peritumoral models. Wang et al. (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e) reported that a 5 mm peritumoral model achieved an AUC of 0.820 (95% CI: 0.760\u0026ndash;0.880), compared to an AUC of 0.798 (95% CI: 0.730\u0026ndash;0.866) for a 10 mm peritumoral model. Additionally, Zhang et al.(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e)found that 5 mm peritumoral models also outperformed 10 mm models in predicting HER2 status. These findings suggest that 5 mm peritumoral regions better capture the heterogeneous distribution of the tumor microenvironment, enhancing prediction accuracy for various tumor-related biological characteristics.\u003c/p\u003e\u003cp\u003eIn this study, a K-means clustering method optimized by C-H values was used for habitat analysis, identifying three subregions reflecting differences in tumor metabolism, blood flow, and cell density. The model based solely on intra-tumor habitat features achieved AUC values of 0.770 (95% CI: 0.621\u0026ndash;0.919) for T2WI and 0.791 (95% CI: 0.667\u0026ndash;0.914) for DCE-MRI on the internal validation set, outperforming the entire tumor internal ROI (AUC: 0.762 and 0.772). Ye et al. (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e) found the habitat region model for predicting pCR in NSCLC had an AUC of 0.781 compared to 0.723 for the whole-tumor model. Similarly, Wang et al. (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e) showed the habitat region model for HGSOC achieved an AUC of 0.808 versus 0.749 for the whole-tumor model.\u003c/p\u003e\u003cp\u003eWe systematically compared feature fusion and image fusion strategies for predicting Ki-67 expression in invasive breast cancer. Feature fusion models achieved a mean validation AUC of 0.791, a 10.8% improvement over image fusion\u0026rsquo;s 0.714. The training-validation gap decreased from 0.077 to 0.017, showcasing enhanced generalization. Specifically, for T2WI \"intra-tumoral\u0026thinsp;+\u0026thinsp;10 mm peritumoral,\" ET demonstrated superior performance: feature fusion achieved a validation AUC of 0.786 (95% CI: 0.652\u0026ndash;0.921) and training AUC of 0.839 (95% CI: 0.771\u0026ndash;0.906), while image fusion had a validation AUC of 0.650 (95% CI: 0.474\u0026ndash;0.826) and training AUC of 0.831 (95% CI: 0.761\u0026ndash;0.902), marking a 20.9% improvement. Overfitting reduced from 0.181 to 0.053. In DCE-MRI, ET achieved a validation AUC of 0.816 (95% CI: 0.653\u0026ndash;0.978) with feature fusion, compared to 0.746 (95% CI: 0.584\u0026ndash;0.909) with image fusion\u0026mdash;a 9.4% improvement (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eDeep learning features, capturing tumor microstructure and high-dimensional nonlinear patterns, comprised two of the top nine key features, enhancing prediction robustness and accuracy. This aligns with Bhuiyan et al. (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e)'s glioma research, where T2-FLAIR deep learning features achieved 0.91 prediction accuracy for Ki-67. Traditional DCE-MRI radiomics features, such as volume ratios and texture, are crucial for assessing tumor invasiveness and treatment response, as shown by Xv et al. (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e) in clear cell renal cell carcinoma studies. The integrated model achieved AUC values of 0.858 and 0.849 on internal and external validation sets, demonstrating clinical potential.\u003c/p\u003e\u003cp\u003eThis study had some limitations. Despite using a multicenter dataset, bias and patient heterogeneity persist, impacting model accuracy and generalization. Manual tumor segmentation, though precise, is time-consuming. Future work should focus on developing semi-automated or fully automated ROI delineation algorithms and integrating deep learning-based adaptive noise reduction to enhance accuracy. Deep learning models require extensive training data; thus, incorporating more patient imaging data is crucial for better generalization. The retrospective design limits causal inference: prospective studies are needed for stronger evidence. Standardizing image quality across multi-center MRI data with varied scanner settings remains challenging, potentially affecting radiomics feature consistency.\u003c/p\u003e\u003cp\u003eIn conclusion, we developed and validated an explainable machine learning model integrating traditional radiomics and deep learning features. This model accurately predicts Ki-67 expression in breast cancer patients, enhancing clinicians' understanding of its decision-making process and aiding in personalized treatment planning.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthical Approval and Consent to participate\u003c/strong\u003e\u003cp\u003e This study was conducted in accordance with the Declaration of Helsinki and has been approved by the Ethics Committee of Longgang Central Hospital of Shenzhen (Shenzhen No.9 People's Hospital) (No.: 2025ECYJ087). A waiver of informed consent was granted due to the retrospective nature of the study and the use of de-identified data.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003cp\u003eNot applicable\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis project is supported by the Shenzhen Science and Technology Program (JCYJ20240813114628038) and Guangdong Basic and Applied Basic Research Foundation (2024A1515220001) and Talent Development Program of Longgang Central Hospital of Shenzhen, the National Natural Science Foundation of China (82472062), the Natural Science Foundation of Guangdong Province of China (No.2024A1515011672). Medical Artificial Intelligence Clinical Application Research Project of Hospital Management Institute, National Health Commission of China (YLXX24AIA010)\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eY.W. and Z.S. contributed equally to this work. Y.W. and Z.S. led the project conception and design, developed and validated the methods, conducted experiments, collected data, performed data analysis and interpretation, created visualizations and charts, wrote and edited the manuscript, and contributed to reviewing and providing feedback. Z.L. provided resources and support, managed and coordinated the project, and contributed to reviewing and providing feedback. Y.X. conducted experiments, collected data, and contributed to data analysis and interpretation. M.L. conducted experiments, collected data, and contributed to reviewing and providing feedback. Y.Z. conducted experiments, collected data, and contributed to reviewing and providing feedback. L.Z. provided resources and support and contributed to reviewing and providing feedback. All authors contributed to reviewing and editing the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e\u003cp\u003eNot applicable\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data generated or analyzed from cohorts A, B during the study are available from the corresponding author by request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eArnold M, Morgan E, Rumgay H, et al. Current and future burden of breast cancer: Global statistics for 2020 and 2040. Breast. 2022;66:15\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.breast.2022.08.010\u003c/span\u003e\u003cspan address=\"10.1016/j.breast.2022.08.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSavva K-V, MacKenzie A, Coombes RC, Zhifang NM, Hanna BG, Peters CJ. An original study assessing biomarker success rate in breast cancer recurrence biomarker research. BMC Med. 2024;22(1):307. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12916-024-03460-6\u003c/span\u003e\u003cspan address=\"10.1186/s12916-024-03460-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eInwald EC, Klinkhammer-Schalke M, Hofst\u0026auml;dter F, et al. Ki-67 is a prognostic parameter in breast cancer patients: results of a large population-based cohort of a cancer registry. Breast Cancer Res Treat. 2013;139(2):539\u0026ndash;52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10549-013-2560-8\u003c/span\u003e\u003cspan address=\"10.1007/s10549-013-2560-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eĐokić S, Gazić B, Grčar Kuzmanov B, et al. Clinical and Analytical Validation of Two Methods for Ki-67 Scoring in Formalin Fixed and Paraffin Embedded Tissue Sections of Early Breast Cancer. Cancers (Basel). 2024;16(7):1405. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/cancers16071405\u003c/span\u003e\u003cspan address=\"10.3390/cancers16071405\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang C, Chen J, Lv X, et al. Ki-67\u0026mdash;Playing a key role in breast cancer but difficult to apply precisely in the real world. BMC Cancer. 2025;25(1):962. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12885-025-14374-8\u003c/span\u003e\u003cspan address=\"10.1186/s12885-025-14374-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee J, Lee Y-J, Bae SJ, et al. Ki-67, 21-Gene Recurrence Score, Endocrine Resistance, and Survival in Patients With Breast Cancer. JAMA Netw Open. 2023;6(8):e2330961. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamanetworkopen.2023.30961\u003c/span\u003e\u003cspan address=\"10.1001/jamanetworkopen.2023.30961\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLin J-Y, Ye J-Y, Chen J-G, Lin S-T, Lin S, Cai S-Q. Prediction of Receptor Status in Radiomics: Recent Advances in Breast Cancer Research. Acad Radiol. 2024;31(7):3004\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.acra.2023.12.012\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2023.12.012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOtikovs M, Nissan N, Furman-Haran E, et al. Relaxation-Diffusion T2-ADC Correlations in Breast Cancer Patients: A Spatiotemporally Encoded 3T MRI Assessment. Diagnostics (Basel). 2023;13(23):3516. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics13233516\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13233516\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNie T, Feng M, Yang K, et al. Correlation between dynamic contrast-enhanced MRI characteristics and apparent diffusion coefficient with Ki-67-positive expression in non-mass enhancement of breast cancer. Sci Rep Nat Publishing Group. 2023;13(1):21451. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-023-48445-2\u003c/span\u003e\u003cspan address=\"10.1038/s41598-023-48445-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang Y-P, Zhang X-Y, Cheng Y-T, et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Military Med Res. 2023;10(1):22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40779-023-00458-8\u003c/span\u003e\u003cspan address=\"10.1186/s40779-023-00458-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlabi RO, Elmusrati M, Leivo I, Almangush A, M\u0026auml;kitie AA. Artificial Intelligence-Driven Radiomics in Head and Neck Cancer: Current Status and Future Prospects. Int J Med Inf. 2024;188:105464. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijmedinf.2024.105464\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2024.105464\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTaghavi M, Staal F, Gomez Munoz F, et al. CT-Based Radiomics Analysis Before Thermal Ablation to Predict Local Tumor Progression for Colorectal Liver Metastases. Cardiovasc Intervent Radiol. 2021;44(6):913\u0026ndash;20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00270-020-02735-8\u003c/span\u003e\u003cspan address=\"10.1007/s00270-020-02735-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ejca.2011.11.036\u003c/span\u003e\u003cspan address=\"10.1016/j.ejca.2011.11.036\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan R, Sui C, Wang C, Zhu T. MRI-based intratumoral and peritumoral radiomics for preoperative prediction of glioma grade: a multicenter study. Front Oncol Front. 2024;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2024.1401977\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2024.1401977\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang L, Wang C, Deng X, et al. Multimodal Ultrasound Radiomic Technology for Diagnosing Benign and Malignant Thyroid Nodules of Ti-Rads 4\u0026ndash;5: A Multicenter Study. Sens (Basel). 2024;24(19):6203. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/s24196203\u003c/span\u003e\u003cspan address=\"10.3390/s24196203\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZuo D, Yang L, Jin Y, Qi H, Liu Y, Ren L. Machine learning-based models for the prediction of breast cancer recurrence risk. BMC Med Inf Decis Mak. 2023;23(1):276. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12911-023-02377-z\u003c/span\u003e\u003cspan address=\"10.1186/s12911-023-02377-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu J, He H, Wang Y et al. Predictive models for secondary epilepsy in patients with acute ischemic stroke within one year. eLife 13:RP98759. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7554/eLife.98759\u003c/span\u003e\u003cspan address=\"10.7554/eLife.98759\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePei L, Han X, Ni C, Ke J. Prediction of prognosis in acute ischemic stroke after mechanical thrombectomy based on multimodal MRI radiomics and deep learning. Front Neurol. 2025;16:1587347. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fneur.2025.1587347\u003c/span\u003e\u003cspan address=\"10.3389/fneur.2025.1587347\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDu L, Yuan J, Gan M, et al. A comparative study between deep learning and radiomics models in grading liver tumors using hepatobiliary phase contrast-enhanced MR images. BMC Med Imaging. 2022;22(1):218. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12880-022-00946-8\u003c/span\u003e\u003cspan address=\"10.1186/s12880-022-00946-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang F, Cheng M, Du B, et al. Predicting microvascular invasion in small (\u0026le;\u0026thinsp;5 cm) hepatocellular carcinomas using radiomics-based peritumoral analysis. Insights Imaging. 2024;15(1):90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13244-024-01649-0\u003c/span\u003e\u003cspan address=\"10.1186/s13244-024-01649-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang H, Miao Q, Fu Y, et al. Intratumoral and peritumoral radiomics based on automated breast volume scanner for predicting human epidermal growth factor receptor 2 status. Front Oncol. 2025;15:1556317. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2025.1556317\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2025.1556317\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYe G, Wu G, Zhang C, et al. CT-based quantification of intratumoral heterogeneity for predicting pathologic complete response to neoadjuvant immunochemotherapy in non-small cell lung cancer. Front Immunol. 2024;15:1414954. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fimmu.2024.1414954\u003c/span\u003e\u003cspan address=\"10.3389/fimmu.2024.1414954\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang X, Xu C, Grzegorzek M, Sun H. Habitat radiomics analysis of pet/ct imaging in high-grade serous ovarian cancer: Application to Ki-67 status and progression-free survival. Front Physiol. 2022;13:948767. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fphys.2022.948767\u003c/span\u003e\u003cspan address=\"10.3389/fphys.2022.948767\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR C, S Q, Y W, et al. Deep radiomic model based on the sphere-shell partition for predicting treatment response to chemotherapy in lung cancer. Translational Oncol Transl Oncol; 2023;35. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.tranon.2023.101719\u003c/span\u003e\u003cspan address=\"10.1016/j.tranon.2023.101719\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbdullakutty F, Akbari Y, Al-Maadeed S, Bouridane A, Talaat IM, Hamoudi R. Histopathology in focus: a review on explainable multi-modal approaches for breast cancer diagnosis. Front Med Front. 2024;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmed.2024.1450103\u003c/span\u003e\u003cspan address=\"10.3389/fmed.2024.1450103\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBhuiyan EH, Khan MM, Hossain SA, et al. Classification of glioma grade and Ki-67 level prediction in MRI data: A SHAP-driven interpretation. Comput Med Imaging Graph. 2025;124:102578. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compmedimag.2025.102578\u003c/span\u003e\u003cspan address=\"10.1016/j.compmedimag.2025.102578\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXv Y, Xiao B, Wei Z, et al. Interpretable CT Radiomics-based Machine Learning Model for Preoperative Prediction of Ki-67 Expression in Clear Cell Renal Cell Carcinoma. Acad Radiol. 2025;32(5):2739\u0026ndash;50. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.acra.2024.11.072\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2024.11.072\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7709859/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7709859/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eBreast cancer (BC) is the most common malignant tumor in women globally. Ki-67, vital for prognosis, is currently detected invasively. Non-invasive MRI prediction faces challenges due to intratumoral heterogeneity.\u003c/p\u003e\u003ch2\u003eMaterials and Methods\u003c/h2\u003e\u003cp\u003eThis retrospective study included 254 breast cancer patients from two centers, divided into training (142 patients), internal validation set (60 patients), and external validation set (52 patients). T2WI and DCE-MRI sequences were analyzed. Traditional radiomics features were extracted from intratumoral, peritumoral (5 mm, 10 mm), and habitat regions. A pre-trained ResNet-50 model extracted 2.5D deep learning features. Feature selection used ICC, Z-score normalization, t-tests, Pearson correlations, and LASSO. Models were built using SVM, RF, and ET algorithms, evaluated via AUC, accuracy, sensitivity, specificity, and F1 score. SHAP analysis enhanced interpretability.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThe best-performing traditional radiomics model achieved an AUC of 0.825 (95% CI: 0.708\u0026ndash;0.942) in the internal validation set. The optimal deep learning model obtained an AUC of 0.804 (95% CI: 0.641\u0026ndash;0.966) in the internal validation set. The combined model, utilizing both best traditional radiomics and deep learning features, demonstrated superior performance with an AUC of 0.885 (95% CI: 0.787\u0026ndash;0.984) in the internal validation set and 0.839 (95% CI: 0.727\u0026ndash;0.951) in the external validation set.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eThe integrated model combining traditional radiomics and deep learning from MRI significantly predicts Ki-67 expression in breast cancer, enhancing preoperative prediction accuracy and interpretability for personalized treatment.\u003c/p\u003e","manuscriptTitle":"Multiparametric MRI-based Habitat Analysis Integrating Deep Learning and Radiomics for Predicting Preoperative Ki-67 Expression Level in Breast Cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-26 13:24:50","doi":"10.21203/rs.3.rs-7709859/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-12-10T03:19:25+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-08T06:13:26+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"189129188648269823216653971425007168008","date":"2025-11-28T02:16:56+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-27T12:53:29+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"310093227504096119705645415239794097155","date":"2025-11-26T17:54:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"19431166650369547441002456997693091962","date":"2025-10-21T13:03:34+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"275705689757003067571733119657787096570","date":"2025-10-19T15:15:01+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-10T08:04:02+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-01T15:03:58+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-30T15:45:47+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-30T15:43:37+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-09-25T07:09:49+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"58878136-0f78-4e34-89a8-80d8bb1cfeb5","owner":[],"postedDate":"October 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-01-19T16:50:36+00:00","versionOfRecord":{"articleIdentity":"rs-7709859","link":"https://doi.org/10.1186/s12880-026-02151-3","journal":{"identity":"bmc-medical-imaging","isVorOnly":false,"title":"BMC Medical Imaging"},"publishedOn":"2026-01-16 16:31:10","publishedOnDateReadable":"January 16th, 2026"},"versionCreatedAt":"2025-10-26 13:24:50","video":"","vorDoi":"10.1186/s12880-026-02151-3","vorDoiUrl":"https://doi.org/10.1186/s12880-026-02151-3","workflowStages":[]},"version":"v1","identity":"rs-7709859","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7709859","identity":"rs-7709859","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}