{"paper_id":"08eb3c35-37c0-4ef8-b2c8-a7a8acf21171","body_text":"2D and 2.5D Deep Learning Models for Neoadjuvant Chemotherapy Response Prediction 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 Article 2D and 2.5D Deep Learning Models for Neoadjuvant Chemotherapy Response Prediction in Breast Cancer Yalei Wang, Fuqiang Pan, Mengqing Kang, Baoqi Zhang, Yang Zhang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8229046/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Neoadjuvant chemotherapy (NAC) is central to managing locally advanced breast cancer, and accurate response assessment is essential for guiding clinical decision-making. This study developed and compared 2D and 2.5D deep learning models using multimodal breast Magnetic resonance imaging (MRI) to predict NAC response. A retrospective cohort of 187 patients (mean age 50.24 years) treated at two different hospitals between January 2020 − December 2024 was analyzed. Inclusion required histopathologically confirmed invasive breast cancer, completion of standardized NAC, pre- and post-treatment MRI (dynamic contrast-enhanced (DCE), T2WI, diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC)), and postoperative pathological confirmation of response, including pathologic complete response (pCR). T1, T2, DWI, ADC, and multiphase DCE images were integrated into a seven-channel input, alongside a separate early-phase DCE input. Four convolutional neural networks (VGG16_bn, ResNet50, DenseNet121, and SqueezeNet1_0) were trained under 2D and 2.5D frameworks with transfer learning. Grad-CAM was applied for interpretability, and clinical features (speculation, edema changes) were incorporated into a combined predictive model. Performance evaluation included ROC, AUC(Area Under the Curve), sensitivity, calibration curves, and Hosmer–Lemeshow tests. In the training cohort, the fusion model achieved the highest performance (AUC 0.955). In the independent test cohort, 2.5D ResNet50 performed best (AUC 0.897, accuracy 0.895, sensitivity 0.944). Fusion further improved AUC to 0.955, demonstrating superior generalizability and clinical potential. Biological sciences/Cancer Biological sciences/Computational biology and bioinformatics Health sciences/Medical research Health sciences/Oncology breast cancer MRI neoadjuvant chemotherapy 2.5D deep learning 2D deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 INTRODUCTION Breast cancer remains a leading threat to women's health worldwide and one of the primary causes of cancer-related mortality among women, representing a major public health challenge in the 21st century [ 1 ]. Neoadjuvant chemotherapy (NAC), an essential therapeutic strategy for locally advanced breast cancer, can effectively reduce tumor size, increase breast-conserving surgery rates, and provide valuable information for postoperative pathological assessment and individualized treatment planning [ 2 , 3 ]. However, the response to NAC varies substantially among breast cancer patients, and some individuals may show minimal benefit or even experience disease progression during treatment [4]. Therefore, accurate prediction of NAC response prior to treatment is critical for optimizing therapeutic decision-making, avoiding overtreatment, and improving patient outcomes. Magnetic resonance imaging (MRI), owing to its multiparametric and multimodal imaging capabilities, has been widely used for breast cancer screening, characterization, and treatment response assessment [ 5 ]. In particular, sequences such as DCE imaging, DWI, T2-weighted imaging, and ADC mapping provide complementary anatomical and functional information about breast tumors [6]. In recent years, artificial intelligence approaches such as radiomics and deep learning have offered new avenues for quantitative analysis and predictive modeling using breast MRI [7, 8]. Traditional radiomics methods rely on manually engineered features, are susceptible to variations in region of interest (ROI) delineation and feature selection algorithms, and often suffer from limited generalizability. In contrast, deep learning techniques—particularly convolutional neural networks (CNNs)—can automatically learn high-dimensional abstract representations from images without explicit feature engineering, achieving outstanding performance in tasks such as tumor detection, classification, and treatment response prediction [9]. Previous studies have attempted to develop 2D deep learning models based on MRI for predicting NAC response in breast cancer [ 10 ]. However, such models typically rely on single-slice images or single-modality inputs, thereby overlooking the spatial characteristics of the tumor and the role of the surrounding tumor microenvironment in treatment response. To address these limitations, a 2.5D strategy has recently been proposed, in which the largest cross-sectional slice of the tumor is combined with multimodal information to approximate three-dimensional representation, thereby enhancing the model’s ability to capture intratumoral heterogeneity [ 11 , 12 ]. Moreover, imaging features of the peritumoral region have been shown to be strongly associated with NAC response [ 13 ], yet they remain frequently overlooked in many conventional deep learning frameworks. Building on this background, the present study aimed to develop a 2.5D deep learning model that integrates multiple MRI modalities and to systematically compare its performance with that of conventional 2D models in predicting the response to NAC in breast cancer. In addition, the best-performing model is combined with key clinical features to construct an integrated analytical framework with enhanced predictive performance and practical clinical applicability. Figure 1 illustrates the overall workflow of this study. RESULTS Clinical characteristics A total of 187 breast cancer patients who received neoadjuvant chemotherapy (NAC) were included in this study, with 130 assigned to the training cohort and 57 to the test cohort. Table 1 summarizes the baseline characteristics of the two groups. No significant intergroup differences were observed regarding age, hormone receptor status (estrogen receptor (ER) and progesterone receptor (PR)), HER2 expression, Ki-67 level, molecular subtype, tumor type, axillary lymph node enlargement, edema grade, enhancement pattern, or spiculation signs (all p > 0.05), indicating that the two cohorts were well balanced and suitable for subsequent model training and validation. Table 1 Baseline characteristics of the study cohorts. feature_name label = ALL label = test label = train p -value Age 50.24 ± 10.37 49.11 ± 9.92 50.74 ± 10.56 0.101 ER 0.798 0 104(55.61) 33(57.89) 71(54.62) 1 83(44.39) 24(42.11) 59(45.38) PR 1.0 0 137(73.26) 42(73.68) 95(73.08) 1 50(26.74) 15(26.32) 35(26.92) Her2 \\nC_erb_2 0.232 0 96(51.34) 25(43.86) 71(54.62) 1 91(48.66) 32(56.14) 59(45.38) Ki_67 0.74 0 123(65.78) 36(63.16) 87(66.92) 1 64(34.22) 21(36.84) 43(33.08) Molecular_Subtype 0.546 1 42(22.46) 10(17.54) 32(24.62) 2 52(27.81) 16(28.07) 36(27.69) 3 60(32.09) 18(31.58) 42(32.31) 4 33(17.65) 13(22.81) 20(15.38) TIC0 0.319 1 126(67.38) 40(70.18) 86(66.15) 2 56(29.95) 17(29.82) 39(30.00) 3 5(2.67) Null 5(3.85) Tumor_Type 1.0 0 56(29.95) 17(29.82) 39(30.00) 1 131(70.05) 40(70.18) 91(70.00) Axillary_Lymph_Node_Enlargement 0.772 0 29(15.51) 10(17.54) 19(14.62) 1 158(84.49) 47(82.46) 111(85.38) Enhancement_Pattern 0.615 0 137(73.26) 44(77.19) 93(71.54) 1 50(26.74) 13(22.81) 37(28.46) Spiculation 0.414 0 67(35.83) 24(42.11) 43(33.08) 1 120(64.17) 33(57.89) 87(66.92) Edema0 0.579 1 29(15.51) 12(21.05) 17(13.08) 2 86(45.99) 25(43.86) 61(46.92) 3 28(14.97) 8(14.04) 20(15.38) 4 44(23.53) 12(21.05) 32(24.62) Edema1 0.582 0 1(0.53) null 1(0.77) 1 49(26.20) 18(31.58) 31(23.85) 2 85(45.45) 27(47.37) 58(44.62) 3 28(14.97) 7(12.28) 21(16.15) 4 24(12.83) 5(8.77) 19(14.62) Edema_Change 0.145 1 88(47.06) 30(52.63) 58(44.62) 2 61(32.62) 14(24.56) 47(36.15) Further univariate and stepwise multivariate logistic regression analyses were performed to identify clinical predictors associated with pathological complete response (pCR) (Table 2 ). The results showed that spiculation (OR = 0.266, 95% CI: 0.128–0.555, p = 0.003) and edema grade Edema1 (OR = 0.524, 95% CI: 0.329–0.836, p = 0.023) were significant negative predictors of pCR, suggesting that both may hold important value for treatment response prediction. Age also showed borderline significance ( p = 0.086). Table 2 Univariable analysis of clinical features. feature_name Log(OR) lower 95%CI upper 95%CI OR OR lower 95%CI OR upper 95%CI p_value PR -0.862 -1.802 0.078 0.422 0.165 1.081 0.131 ER -0.545 -1.317 0.227 0.580 0.268 1.255 0.245 Spiculation -1.323 -2.056 -0.589 0.266 0.128 0.555 0.003 Edema_Change -0.103 -0.425 0.219 0.902 0.654 1.245 0.599 Edema1 -0.646 -1.113 -0.179 0.524 0.329 0.836 0.023 Edema0 0.463 -0.004 0.930 1.589 0.996 2.535 0.103 Age 0.025 0.001 0.049 1.026 1.001 1.050 0.086 Performance of deep learning models Under the 2D and 2.5D input strategies, this study compared the predictive performance of four CNN models (VGG16BN, ResNet50, DenseNet121, and SqueezeNet1.0) on the training and independent test cohorts (Table 3 ). In the training cohort, the 2.5D ResNet50 model outperformed all other models, with an AUC of 0.919, an accuracy of 0.902, a sensitivity of 0.818, a specificity of 0.966, and an F1-score of 0.878. In the test cohort, the 2.5D ResNet50 model consistently maintained the best performance, achieving an AUC of 0.897 (95% CI: 0.7722–1.000), an accuracy of 0.895, and a sensitivity of 0.944, demonstrating strong generalization and stability. In contrast, the 2D models generally exhibited lower AUCs than the 2.5D models, with significant performance differences, suggesting that integrating multimodal image fusion and peritumoral features can enhance predictive performance. Table 3 Predictive performance of four CNNs in 2D and 2.5D frameworks. Signature Accuracy AUC 95% CI Sensitivity Specificity PPV NPV Precision Recall F1 Threshold Cohort DL25D_vgg16bn 0.824 0.859 0.7836–0.9343 0.659 0.948 0.906 0.786 0.906 0.659 0.763 0.476 train DL25D_resnet50 0.902 0.919 0.8578–0.9800 0.818 0.966 0.947 0.875 0.947 0.818 0.878 0.501 train DL25D_densenet121 0.725 0.752 0.6555–0.8477 0.591 0.828 0.722 0.727 0.722 0.591 0.650 0.490 train DL25D_squeezenet10 0.892 0.850 0.7608–0.9390 0.818 0.948 0.923 0.873 0.923 0.818 0.867 0.486 train DL2D_vgg16bn 0.833 0.833 0.7507–0.9162 0.932 0.759 0.745 0.936 0.745 0.932 0.828 0.500 train DL2D_resnet50 0.843 0.862 0.7859–0.9375 0.864 0.828 0.792 0.889 0.792 0.864 0.826 0.512 train DL2D_densenet121 0.833 0.845 0.7651–0.9254 0.750 0.897 0.846 0.825 0.846 0.750 0.795 0.466 train DL2D_squeezenet10 0.775 0.758 0.6567–0.8590 0.705 0.828 0.756 0.787 0.756 0.705 0.729 0.501 train DL25D_vgg16bn 0.763 0.756 0.5836–0.9275 0.722 0.800 0.765 0.762 0.765 0.722 0.743 0.462 test DL25D_resnet50 0.895 0.897 0.7722–1.0000 0.944 0.850 0.850 0.944 0.850 0.944 0.895 0.499 test DL25D_densenet121 0.816 0.767 0.6041–0.9293 0.611 1.000 1.000 0.741 1.000 0.611 0.759 0.506 test DL25D_squeezenet10 0.816 0.797 0.6411–0.9533 0.722 0.900 0.867 0.783 0.867 0.722 0.788 0.419 test DL2D_vgg16bn 0.711 0.700 0.5209–0.8791 0.667 0.750 0.706 0.714 0.706 0.667 0.686 0.496 test DL2D_resnet50 0.763 0.764 0.6029–0.9248 0.889 0.650 0.696 0.867 0.696 0.889 0.780 0.497 test DL2D_densenet121 0.737 0.675 0.4939–0.8561 0.611 0.850 0.786 0.708 0.786 0.611 0.687 0.502 test DL2D_squeezenet10 0.684 0.681 0.5140–0.8471 0.500 0.850 0.750 0.654 0.750 0.500 0.600 0.500 test DL2D: Deep learning 2D model; DL25D: Deep learning 2.5D model; PPV: positive predictive value; NPV: negative predictive value. Table 4 Performance metrics of clinical, 2D, 2.5D, and combined deep learning models. Signature Accuracy AUC 95% CI Sensitivity Specificity PPV NPV Precision Recall F1 Threshold Cohort Clinical 0.676 0.698 0.5971–0.7991 0.477 0.828 0.677 0.676 0.677 0.477 0.560 0.556 train DL25D 0.902 0.919 0.8578–0.9800 0.818 0.966 0.947 0.875 0.947 0.818 0.878 0.501 train DL2D 0.843 0.862 0.7859–0.9375 0.864 0.828 0.792 0.889 0.792 0.864 0.826 0.512 train Combined 0.931 0.955 0.9133–0.9973 0.955 0.914 0.894 0.964 0.894 0.955 0.923 0.443 train Clinical 0.632 0.654 0.4818–0.8265 0.944 0.350 0.567 0.875 0.567 0.944 0.708 0.308 test DL25D 0.895 0.897 0.7722–1.0000 0.944 0.850 0.850 0.944 0.850 0.944 0.895 0.499 test DL2D 0.763 0.764 0.6029–0.9248 0.889 0.650 0.696 0.867 0.696 0.889 0.780 0.497 test Combined 0.842 0.864 0.7378–0.9900 0.778 0.900 0.875 0.818 0.875 0.778 0.824 0.929 test DL2D: Deep-learning 2D model; DL25D: Deep-learning 2.5D model. Evaluation of the combined model and statistical comparison A combined model was constructed by integrating the deep-learning probabilities generated by the 2.5D ResNet50 model with clinically significant variables (spiculation sign, edema grade, and age) using the ExtraTrees regression algorithm. In the training cohort, the combined model achieved an AUC of 0.955 (95% CI: 0.9133–0.9973). Moreover, the model demonstrated good calibration, with the Hosmer–Lemeshow test showing no significant lack of fit in either the training or test cohort ( p > 0.05), and the calibration curves exhibiting close agreement between predicted and observed outcomes (Fig. 4 ). Decision curve analysis (DCA) indicated that the combined model provided the highest net benefit across most clinical decision thresholds (Fig. 5 ), further supporting its strong potential for clinical application. The DeLong test showed that the combined model achieved a significantly higher AUC compared to the clinical model and the 2D model ( p < 0.01), exceeding also that of the 2.5D model (p = 0.047) (Fig. 6 ). Grad-CAM visualization analysis To further enhance model interpretability and to examine the decision basis of the deep learning model in predicting the efficacy of NAC for breast cancer, we employed Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize the model’s salient regions. Grad-CAM propagates gradient information back to the final convolutional layer to generate heatmaps that highlight the image regions most influential for the model’s classification decisions, thereby revealing the visual basis underlying the network’s predictions. We selected representative pCR and non-pCR cases from both the training and test cohorts and compared the heatmap distributions generated by the 2.5D ResNet50 model with those produced by the 2D models (e.g., 2D VGG16BN). The results are presented in Fig. 7 . The analysis revealed that the Grad-CAM heatmaps generated by the 2.5D model were predominantly concentrated within the tumor parenchyma and its margins, and in some cases extended into adjacent non-neoplastic tissue and interstitial regions. This pattern suggests that the model may have captured microenvironmental features associated with treatment response, such as the tumor–stroma interface, vascular distribution, and edema spread. In contrast, the activation regions of the 2D models were often confined to the tumor core, with some cases exhibiting activation “shifts” or even “blank responses,” indicating limited ability to extract key features and suggesting instability in model decision-making. Moreover, in some non-pCR patients, the Grad-CAM maps of the 2.5D model highlighted salient regions located in irregularly enhanced peripheral tumor areas, spiculated structures, and surrounding ADC-abnormal regions. This observation suggests that the model may have detected focal structural or functional abnormalities potentially associated with chemotherapy resistance. Taken together with the preceding performance assessments, the Grad-CAM analysis further demonstrated from an interpretability perspective that the 2.5D architecture, through multimodal integration and enhanced spatial contextual modeling, improved the recognition of complex tumor phenotypes, thereby contributing to greater model stability and clinical credibility. DISCUSSION Key findings In this study, we systematically developed and compared multiple deep learning models, proposing a 2.5D CNN architecture based on multimodal breast MRI inputs and validating its effectiveness and stability in predicting the response to NAC in breast cancer. Our main observations can be outlined as follows: 1. The 2.5D model markedly outperformed the 2D models and demonstrated a substantially greater capacity for capturing contextual and spatially informative features. Compared with conventional 2D models that operate on single-view images, the 2.5D model achieved superior predictive performance in both the training and test cohorts, with particularly notable advantages in key metrics such as AUC, sensitivity, and F1 score. For example, using ResNet50, the 2.5D model achieved an AUC of 0.897 in the test cohort, substantially outperforming the 2D model (AUC = 0.764) and demonstrating higher accuracy as well as more stable generalization performance. The 2.5D model integrates seven imaging modalities—T1, T2WI, DWI, ADC, and early-, mid-, and late-phase DCE—thereby capturing multidimensional physiological information encompassing tumor anatomy, diffusion properties, and perfusion dynamics, while also enhancing the sensitivity of the model to temporal evolution and microenvironmental dynamics. This spatial–temporal multimodal integration mechanism substantially strengthened the model’s capacity to characterize tumor heterogeneity, microcirculatory alterations, and their associations with chemotherapy response. 2. Among the various network architectures evaluated, ResNet50 achieved the best overall performance, demonstrating strong generalization and stability. In this study, four classical CNN architectures—VGG16_bn [ 14 ],ResNet50 [15], DenseNet121 [16], and SqueezeNet1.0 [17]—were selected for comparison. Under both the 2D and 2.5D input strategies, ResNet50 achieved the highest AUC and F1 scores and exhibited the least performance fluctuation between the training and test cohorts, indicating superior generalization capability. By incorporating residual connections, ResNet50 effectively mitigates gradient vanishing and network degradation in deep architectures, facilitating more stable convergence during training and enhancing its capacity for feature abstraction and transfer learning [15]. In our study, ResNet50 not only preserved robust baseline feature extraction capabilities but also more precisely captured high-level treatment-related imaging features, demonstrating superior discriminative power particularly in cases characterized by peripheral tumor enhancement, ADC alterations, and pronounced intratumoral heterogeneity. 3. The combined model integrates clinical variables with imaging-derived deep-learning features, further improving predictive performance and enhancing model interpretability. Notably, although the combined model achieved the best performance in the training cohort (AUC = 0.955), this advantage did not persist in the independent test cohort, where the 2.5D ResNet50 model alone demonstrated the highest performance. This discrepancy may be attributable to several factors. First, the combined model effectively integrated the complementary information between deep imaging features and clinical variables during training, resulting in a higher degree of model fit. However, the distribution of clinical variables (e.g., spiculation and edema grade) may vary across centers, and their stability and generalizability are inferior to those of imaging features, which likely attenuated the overall predictive performance during external validation. Second, although the ExtraTrees algorithm, which has strong nonlinear modeling capacity, captured complex interactions between imaging features and clinical variables in the training set, its dependence on sample-specific clinical variables may have contributed to reduced stability when applied to new data. In contrast, the 2.5D ResNet50 model, which relies primarily on deep imaging phenotypes derived from multimodal MRI sequences, exhibited greater feature stability across centers and consequently demonstrated superior robustness and reproducibility in the independent test cohort. These findings suggest that deep imaging features may serve as more reliable predictors across populations and clinical centers, whereas the contribution of clinical variables to model fusion requires further validation and optimization in larger, multicenter datasets. Comparison with existing studies In recent years, the application of artificial intelligence in breast cancer imaging has advanced rapidly. In particular, multiple studies have explored the development of MRI-based deep-learning and radiomics models for predicting the response to NAC. However, existing studies share several limitations, including reliance on single-modality imaging, restricted input dimensionality, limited model generalizability, and the absence of integrated clinical features [18–20]. Regarding image input strategies, most existing studies have adopted two-dimensional (2D) deep-learning frameworks, typically constructing predictive models from the largest tumor cross-section on DCE or T2WI images. For example, Joo et al. developed a deep-learning model integrating subtraction T1-weighted images, T2-weighted images, and clinical variables to predict pCR prior to NAC. The model achieved an AUC of 0.888 in the validation cohort, significantly outperforming the clinical-only model (AUC = 0.827, p < 0.05) [ 21 ]. Although such approaches yield performance improvements, they still fail to comprehensively capture the spatial heterogeneity and dynamic enhancement patterns of the tumor. Some studies have attempted 2D multimodal fusion strategies, in which images from different MRI sequences are fed into the model and concatenated at the feature level to enhance predictive performance. For instance, Fan et al. extracted radiomics features from DCE-MRI, DWI, and T2WI images and performed multiparametric fusion, which markedly outperformed single-modality models in predicting NAC response [ 22 ]. Although this strategy improves spatial coverage, it remains constrained by its reliance on single-slice planar inputs and therefore cannot overcome the inherent limitations in representing tumor eco-structural complexity and heterogeneity. In contrast, the 2.5D deep-learning strategy proposed in this study substantially enhances the modeling capacity for multidimensional tumor phenotypes. By selecting the largest tumor cross-section and extracting the corresponding slice across multiple MRI modalities, the method achieves a “2D input, 3D-aware” modeling paradigm. This strategy preserves computational efficiency while markedly improving the model’s ability to capture complex structures such as tumor morphological boundaries and microenvironmental features. Consequently, it achieves substantial performance gains, with an AUC of 0.897 in the test cohort—surpassing most reported 2D models. In exploring model architectures, this study systematically compared four representative CNN architectures. Through multiple rounds of experimental evaluation, ResNet50 demonstrated the most favorable stability and performance, consistent with findings reported in previous studies. Peng et al. [23] reported that a ResNet50-based deep-learning model outperformed traditional radiomics approaches in predicting pCR to NAC using preoperative DCE-MRI. Their findings highlight that deep residual networks can effectively extract high-level imaging features associated with treatment response, thereby achieving superior generalizability and reliability in cross-cohort analyses. Collectively, these results further support the potential of ResNet50 as a backbone architecture for breast MRI-based predictive modeling, particularly in the context of NAC response assessment. In the study by Islam et al. [ 24 ], a ResNet50-based breast lesion classification model achieved a significantly improved AUC (0.866 ± 0.015) after the integration of a convolutional block attention module further demonstrating the strong potential of ResNet50 in breast disease–related imaging analysis. Our study further verifies its extensibility under a 2.5D input strategy and demonstrates enhanced cross-center generalizability. Notably, our study found a significant association between the emergence of a peritumoral edema pattern (edema1) after the first cycle of NAC and the achievement of pCR, suggesting that the presence or evolution of edema may reflect an early treatment response. Previous studies have incorporated peritumoral imaging features as key imaging markers or radiomic predictors of NAC response [25], and have reported their added value in predicting pCR [26]. However, other studies have reported that pre-treatment peritumoral edema is associated with poor prognosis, such as higher recurrence rates and reduced survival, suggesting that the prognostic implication of edema may depend on the timing of assessment (pre-treatment vs on-treatment/post-treatment follow-up) and its dynamic evolution [27]. From a pathobiological perspective, peritumoral edema may reflect local inflammatory responses, increased vascular permeability, or stromal alterations within the tumor microenvironment. When these changes emerge or intensify shortly after the initiation of chemotherapy, they may indicate chemotherapy-induced tumor cell necrosis, vascular permeability alterations, or immune-cell infiltration, thereby correlating with a higher likelihood of achieving pCR. Based on this hypothesis, short-interval imaging follow-up—for example, after one chemotherapy cycle—to monitor dynamic changes in peritumoral edema may serve as a feasible imaging biomarker for early treatment response assessment and therapy optimization. Model interpretability and clinical value In the application of artificial intelligence—particularly deep learning—to medical imaging, model interpretability remains one of the primary barriers to clinical translation. To enhance the transparency and trustworthiness of the proposed model, we incorporated Gradient-weighted Class Activation Mapping (Grad-CAM) [ 28 ] to spatially visualize the decision-making process of the deep learning network. The Grad-CAM heatmaps demonstrated that, when predicting pCR, the 2.5D ResNet50 model consistently focused on the tumor core, tumor margins, and adjacent tissues. In some cases, activation extended to the tumor–stroma interface and regions of abnormal enhancement, suggesting that the model captured intrinsic tumor features and learned microenvironmental characteristics associated with treatment response. These findings support the advantage and reliability of the 2.5D model in predicting NAC response from an interpretability perspective, while providing clinicians with an intuitive understanding of the output of the model. Limitations and future directions Although this study provides several innovative insights, it has certain limitations. First, the sample size was relatively limited and data were collected from only two institutions, which may not fully reflect variability across broader populations or imaging platforms. Future studies should incorporate larger, multicenter, and prospective cohorts to validate the generalizability of the model. Second, the model incorporated only MRI imaging and basic clinical variables, without integrating pathological findings, molecular subtypes, or multi-omics data. Incorporating multimodal biological information may further enhance predictive performance. Finally, the study did not assess the model’s sensitivity across different chemotherapy regimens (e.g., platinum-based vs. non–platinum-based). Subgroup analyses are warranted to further improve the model’s applicability in diverse therapeutic settings. Future research directions include establishing large-scale, standardized, multicenter datasets; developing more interpretable multimodal architectures; integrating imaging data with multi-omics information for joint modeling; and exploring clinically deployable intelligent tools to facilitate real-world implementation. CONCLUSION This study systematically compared the performance of 2D and 2.5D deep-learning models for predicting the response to NAC in breast cancer using breast MRI, and further compared four representative CNN architectures. The results indicate that the 2.5D multichannel model not only captures anatomical, diffusion- and perfusion-related information, but also enhances the model’s awareness of temporal evolution and dynamic tumor microenvironment features, yielding superior overall performance compared with 2D models in predicting pCR. Among the evaluated architectures, ResNet50 demonstrated the highest stability and predictive accuracy, consistent with previous breast MRI deep-learning studies and supporting the robustness of this architecture. Taken together, the findings confirm the effectiveness and feasibility of deep-learning approaches for NAC response prediction and highlight the critical role of multidimensional imaging features and peritumoral microenvironmental information in model performance. Future work integrating large-scale multicenter datasets with multimodal and multi-omics information may further enhance model generalizability and clinical utility, ultimately providing more precise tools for individualized NAC response assessment and therapeutic decision-making. MATERIALS AND METHODS 3.1 Patient cohort and data partitioning This multicenter retrospective study included 187 patients with breast cancer (mean age, 50.24 ± 10.3 years) from two tertiary medical centers, comprising 130 patients from Hospital A (Fuyang Cancer Hospital; 50.74 ± 10.56 years) and 57 patients from Hospital B (Fuyang People’s Hospital; 49.11 ± 9.92 years). The inclusion period spanned from January 2020 to December 2024. All enrolled patients underwent breast MRI examinations both before and after neoadjuvant chemotherapy (NAC), with image quality sufficient for analysis and complete pathological and follow-up data available. The study was approved by the institutional review boards of both centers, and informed consent was waived for all participants. Inclusion criteria included: histopathologically confirmed invasive breast cancer; receipt of NAC with a complete and standardized treatment regimen; availability of pre-treatment breast MRI, including dynamic contrast-enhanced (DCE), T2-weighted imaging (T2WI), and diffusion-weighted imaging (DWI) sequences, as well as apparent diffusion coefficient (ADC) maps; accessible postoperative pathological reference standard for assessing treatment response (e.g., presence or absence of pathological complete response, pCR); and MRI images without severe artifacts, allowing clear identification of the ROI. Exclusion criteria included: incomplete MRI examinations or severe motion artifacts; prior systemic treatments (chemotherapy, radiotherapy, or targeted therapy); unavailable postoperative pathology or incomplete follow-up data; coexistence of other major systemic diseases (e.g., malignancies or severe cardiopulmonary disorders); or tumors too small to allow reliable ROI delineation on MRI (Fig. 2 ). To enhance the clinical generalizability of the model, the dataset was partitioned according to its natural institutional distribution: data from Center 1 were used for model training, whereas data from Center 2 served as an independent validation cohort. Clinical information—including age, hormone receptor status, molecular subtype, Ki-67 index, enhancement pattern, spiculation, and edema characteristics—was extracted and cross-checked from the electronic medical record system. Image acquisition and processing Eleven patients underwent breast MRI according to standard clinical imaging protocols, including five major acquisitions: T1WI, T2WI, DWI, ADC mapping, and DCE imaging. DCE images were acquired in multiple post-contrast phases. Acquisition parameters, including slice thickness, repetition time/echo time (TR/TE), and field of view, followed the standardized protocols of each center to ensure adequate spatial and temporal resolution for assessing tumor morphology and enhancement kinetics. In MRI, low-frequency intensity non-uniformities, also known as bias field artifacts, commonly arise from scanner hardware limitations and variations in tissue magnetic susceptibility. These artifacts lead to inconsistent tissue intensities, thereby impairing reliable feature extraction and negatively affecting the performance of deep learning models. To enhance intensity consistency and improve model robustness, N4 bias field correction was applied to all MRI sequences prior to model input. To ensure spatial consistency across multimodal images and facilitate subsequent ROI extraction and model construction, all MRI sequences underwent standardized registration. Specifically, the early-phase DCE-MRI (DCE–mid phase) was designated as the fixed image, whereas T1WI, T2WI, DWI, ADC, and the remaining DCE phases (intermediate and delayed) were treated as moving images. Registration was performed with minimal structural deformation to align all sequences within a unified coordinate system, thereby preserving spatial correspondence of anatomical structures across modalities. After registration, two radiologists reviewed all images to confirm accurate alignment of key anatomical structures (tumor, glandular tissue, skin, and pectoral muscle) across modalities. ROI segmentation and cropping For all patients, ROIs were manually delineated on early-phase DCE images by two radiologists with expertise in breast imaging, using ITK-SNAP (version 3.8.0). In cases of substantial disagreement between the two readers, the final ROI was adjudicated by a senior breast radiologist with more than 20 years of clinical experience. The images and corresponding ROIs were rigidly registered across all modalities, followed by cropping of the matched regions to ensure consistency of multimodal inputs. To enhance the model’s learning of both the tumor core and its surrounding microenvironment, the largest cross-sectional slice was selected as the representative image for each patient, following previous studies [7–8]. Based on the minimum bounding rectangle of the ROI, an additional 10-pixel margin was expanded in all directions to preserve peritumoral contextual information. All images were then uniformly cropped to a fixed size of 224 × 224 pixels and subjected to intensity normalization to ensure consistency across model inputs. Multimodal image fusion To fully exploit the structural, functional, and dynamic enhancement information contained in different breast MRI sequences, a seven-channel input configuration was constructed, comprising T1WI, T2WI, DWI, ADC, and three DCE-MRI phases (early, intermediate, and delayed). All modalities were registered and cropped on the same tumor slice and ROI level to ensure spatial alignment and anatomical comparability across channels. This channel configuration maintains architectural simplicity while maximally integrating tumor morphological information (T1WI, T2WI), water-molecule diffusion characteristics (DWI, ADC), and hemodynamic alterations (DCE). Together, these modalities provide a rich and complementary phenotypic representation of the tumor. The synergistic integration of these multimodal inputs enhances the model’s ability to capture biological heterogeneity associated with NAC response. This approach addresses the limitations of single-modality methods in spatial context, temporal evolution, and microenvironment perception, ultimately offering a more comprehensive feature foundation for deep learning. Deep learning modeling Model architecture and input strategy. This study developed both 2D and 2.5D deep learning models to systematically compare their performance and representational capacity in predicting the therapeutic response to NAC in breast cancer. For the 2D models, we adopted a conventional single-modality input strategy, using the largest tumor slice from the early-phase DCE-MRI as the input to construct a standard two-dimensional CNN framework. The 2.5D models incorporated a multimodal and multi-temporal fusion strategy, employing a seven-channel input configuration that included T1WI, T2WI, DWI, ADC, and early-, mid-, and late-phase DCE-MRI images. All images were extracted and spatially registered on the basis of the largest tumor slice to ensure anatomical alignment and contextual consistency across modalities. This input strategy integrates multidimensional information—including anatomical structure, tissue diffusion properties, and perfusion dynamics—thereby providing a more comprehensive tumor phenotypic representation and enhancing the model’s ability to capture and model NAC-related response heterogeneity. In terms of network architecture, four classical CNN backbones—VGG16_bn, ResNet50, DenseNet121, and SqueezeNet1_0—were adopted as the foundational models. All models were implemented using the PyTorch 1.11.0 deep learning framework and initialized with ImageNet-pretrained weights to facilitate transfer learning, thereby improving convergence efficiency and mitigating overfitting associated with the relatively limited size of medical imaging datasets. Training and optimization strategy. A transfer-learning strategy was applied during model training, where all networks were fine-tuned from ImageNet-pretrained weights to preserve low-level feature extraction while enhancing task-specific feature representation. Stochastic Gradient Descent (SGD) was used as the optimizer with an initial learning rate of 0.01. A cosine-annealing schedule was employed to dynamically adjust the learning rate according to the following formulation: $$\\:{\\eta\\:}_{t}\\:=\\:{\\eta\\:}_{min}^{i}+\\frac{1}{2}({\\eta\\:}_{max}^{i}-{\\eta\\:}_{min}^{i})(1+cos(\\frac{{T}_{cur}}{{T}_{i}}\\pi\\:\\left)\\right)$$ In this formulation, \\(\\:{\\eta\\:}_{min}^{i}\\) = 0 denotes the minimum learning rate, whereas \\(\\:{\\eta\\:}_{max}^{i}\\) = 0.01 specifies the maximum learning rate. \\(\\:{T}_{i}\\) = 32 represents the total number of epochs in the training cycle. SGD was used as the optimizer, and the Softmax cross-entropy loss function was employed during training. Model evaluation metrics and validation. After training, all models were evaluated on an independent test set, with the area under the receiver operating characteristic curve (ROC-AUC) used as the primary performance metric. Additional metrics—including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1 score, precision, and recall—were also reported. The confidence interval of the AUC was estimated using the DeLong method. To further assess the stability and generalizability of the models, calibration curves were generated separately for the training and test sets, and the goodness of fit of the predicted probabilities was evaluated using the Hosmer–Lemeshow test. In addition, DCA was performed to evaluate the net clinical benefit of each model across a range of decision thresholds. Model interpretability via Grad-CAM. To enhance the clinical interpretability of the models, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to visualize the trained networks. By extracting activation heatmaps from the final convolutional layer, we examined whether the model’s attention regions overlapped with the tumor core and surrounding critical structures, thereby supporting the assessment of the rationality of the model’s predictive decisions. Fusion strategy and statistical analysis Clinical feature selection and modeling. To investigate the independent predictive value of clinical characteristics in assessing the response to NAC in breast cancer, univariate and multivariate regression analyses were first performed on all clinical variables. These included age, ER status, PR status, HER2 status, Ki-67 index, molecular subtype, enhancement pattern, spiculation, the degree and change of edema, axillary lymph node enlargement, and other variables, totaling 15 predictors. For each variable, the odds ratio (OR) and its 95% confidence interval (95% CI) for predicting pCR were calculated, and the corresponding p-values were recorded. Variables with a screening threshold of p < 0.10 were included in subsequent fusion-model analyses to reduce the risk of omitting potentially relevant predictors. Ultimately, significant variables such as spiculation, edema (Edema1), and age were incorporated into the clinical feature model (Clinical Signature). The performance of the clinical model was evaluated using the AUC to assess its independent predictive capability. Fusion model construction. To further improve the predictive accuracy and clinical applicability of the models, we extended the fusion framework between deep learning features and clinical characteristics. Specifically, the best-performing 2.5D ResNet50 and 2D ResNet50 models were selected, and their output probabilities (the “Deep Learning Signature”) were combined with the previously identified significant clinical variables—spiculation, edema grade, and age—to construct a combined model. For model development, the combined model was trained using the Extremely Randomized Trees (ExtraTrees) algorithm. Performance evaluation. The performance of all models—the clinical model, deep learning models, and the combined model—was evaluated separately on the training and test sets. The primary evaluation metrics included ROC-AUC analysis with 95% CI estimates, as well as sensitivity, specificity, accuracy, PPV, NPV, F1 score, precision, and recall. In addition, to assess the calibration of the predicted probabilities, calibration curves were generated, and the Hosmer–Lemeshow test was performed to evaluate statistical goodness of fit. A well-calibrated model should demonstrate close agreement between the predicted probabilities and the observed event rates ( p > 0.05). Model comparison and statistical testing. To determine whether the performance differences between models were statistically significant, pairwise comparisons of the AUC values were conducted using the DeLong test. A p -value < 0.05 was considered to indicate a significant difference in predictive performance between the two considered models. Furthermore, DCA was used to evaluate the net benefit of each model across different threshold probabilities, thereby assisting in determining the clinical decision-making value of the models in real-world settings. A higher net benefit curve indicates greater clinical utility of the model over a broader range of threshold probabilities. Statistical analysis The Shapiro–Wilk test was used to assess the normality of clinical variables. Depending on the distribution of continuous variables, either the t-test or the Mann–Whitney U test was applied. Categorical variables were analyzed using the chi-square (χ²) test. All data analyses were performed using the OnekeyAI platform (version 4.9.1) using Python 3.7.12. Statistical evaluations were conducted with Statsmodels 0.13.2. Machine learning implementations, including support vector machines (SVM), were carried out using Scikit-learn 1.0.2. The deep learning framework was developed with PyTorch 1.11.0 and optimized with CUDA 11.3.1 and cuDNN 8.2.1 to enhance computational performance. Declarations Ethics statement Research involving human participants has been approved by the Medical Ethics Committee of Fuyang People's Hospital. The study was conducted in accordance with local laws and institutional requirements. In accordance with national laws and institutional regulations, written informed consent was not required from participants or their legal guardians/next of kin. As all investigations involving human participants have been reviewed and approved by the Institutional Review Board of Fuyang People's Hospital (Affiliated to Anhui Medical University) (Approval No. [2025]4), the ethics committee has waived the requirement for written informed consent. Competing interests The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. Funding Statement The authors declare that this research and/or the publication of this article has received funding. This work was supported by the 2023 Fuyang Health and Medical Research Project (Grant No. FY2023-005). Author Contribution YW&FP&MK: Data curation, Writing– original draft, Software. BZ: Data curation, Writing original draft. YZ: Funding acquisition, Project administration, Resources, Writing– review & editing(YW, FP, and MK contributed equally to the work) Data Availability The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. References Bray, F. et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J. Clin. 74 , 229–263 (2024). Rastogi, P. et al. Preoperative Chemotherapy: Updates of National Surgical Adjuvant Breast and Bowel Project Protocols B-18 and B-27. JCO 26, 778–785 (2008). Vaidya, J. S. et al. Rethinking neoadjuvant chemotherapy for breast cancer. BMJ. j5913 (2018). Janssen, L. M. et al. Predicting response to neoadjuvant chemotherapy with liquid biopsies and multiparametric MRI in patients with breast cancer. npj Breast Cancer 10, 10 (2024). (5913). Wekking, D. et al. Breast MRI: Clinical Indications, Recommendations, and Future Applications in Breast Cancer Diagnosis. Curr. Oncol. Rep. 25 , 257–267 (2023). Kazama, T., Takahara, T. & Hashimoto, J. Breast Cancer Subtypes and Quantitative Magnetic Resonance Imaging: A Systemic Review. Life 12 , 490 (2022). Lu, G. et al. Deep learning radiomics based on multimodal imaging for distinguishing benign and malignant breast tumours. Front. Med. 11 , 1402967 (2024). Hu, Q. et al. A deep learning methodology for improved breast cancer diagnosis using multiparametric MRI. Sci. Rep. 10, 10536 Khosravi, P., Kazemi, E., Imielinski, M., Elemento, O. & Hajirasouliha, I. Deep Convolutional Neural Networks Enable Discrimination of Heterogeneous Digital Pathology Images. EBioMedicine 27, 317–328 (2018). Choi, R., Joel, M., Hui, M. & Aneja, S. Deep learning algorithm to predict pathologic complete response to neoadjuvant chemotherapy for breast cancer prior to treatment. JCO 40, 600–600 (2022). (2020). La Saint-Esteven, G. A 2.5D convolutional neural network for HPV prediction in advanced oropharyngeal cancer. Comput. Biol. Med. 142 , 105215 (2022). Wang, Y. et al. Assessment of prostate cancer aggressiveness through the combined analysis of prostate MRI and 2.5D deep learning models. Front. Oncol. 15, 1539537 Crombé, A. et al. MRI assessment of surrounding tissues in soft-tissue sarcoma during neoadjuvant chemotherapy can help predicting response and prognosis. Eur. J. Radiol. 109, 178–187 (2018). (2025). Simonyan, K. & Zisserman, A. Very Deep Convolutional Networks for Large-Scale Image Recognition . (2015). He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learning for Image Recognition. in IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 770–778 (IEEE, 2016). 770–778 (IEEE, 2016). (2016). 10.1109/cvpr.2016.90 Huang, G., Liu, Z., Van Der Maaten, L. & Weinberger, K. Q. Densely Connected Convolutional Networks. in IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2261–2269 (IEEE, 2017). 2261–2269 (IEEE, 2017). (2017). 10.1109/cvpr.2017.243 Iandola, F. N. et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and < 0.5MB model size. Choi, R., Joel, M., Hui, M. & Aneja, S. Deep learning algorithm to predict pathologic complete response to neoadjuvant chemotherapy for breast cancer prior to treatment. JCO 40, 600–600 (2022). Lo Gullo, R., Eskreis-Winkler, S., Morris, E. A. & Pinker, K. Machine learning with multiparametric magnetic resonance imaging of the breast for early prediction of response to neoadjuvant chemotherapy. The Breast 49, 115–122 (2020). (2016). Granzier, R. W. Y., van Nijnatten, T. J. A., Woodruff, H. C., Smidt, M. L. & Lobbes, M. B. I. Exploring breast cancer response prediction to neoadjuvant systemic therapy using MRI-based radiomics: A systematic review. Eur. J. Radiol. 121 , 108736 (2019). Joo, S. et al. Multimodal deep learning models for the prediction of pathologic response to neoadjuvant chemotherapy in breast cancer. Sci. Rep. 11 , 18800 (2021). Fan, M. et al. Multiparametric MRI radiomics fusion for predicting the response and shrinkage pattern to neoadjuvant chemotherapy in breast cancer. Front. Oncol. 13, 1057841 Peng, Y. et al. Pretreatment DCE-MRI-Based Deep Learning Outperforms Radiomics Analysis in Predicting Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast Cancer. Front. Oncol. 12, 846775 (2022). (2023). Islam, W. et al. Improving Performance of Breast Lesion Classification Using a ResNet50 Model Optimized with a Novel Attention Mechanism. Tomography. 8, 2411–2425 Cakir Pekoz, B. et al. Can peritumoral edema evaluated by Magnetic Resonance Imaging before neoadjuvant chemotherapy predict complete pathological response in breast cancer? Scott Med J 68, 121–128 (2023). Park, J. et al. Machine Learning Predicts Pathologic Complete Response to Neoadjuvant Chemotherapy for ER + HER2- Breast Cancer: Integrating Tumoral and Peritumoral MRI Radiomic Features. Diagnostics 13, 3031 (2023). Shigematsu, H. et al. Prognostic Value of MRI Assessment of Residual Peritumoral Edema in Breast Cancer Treated With Neoadjuvant Chemotherapy. Magnetic Resonance Imaging 61, 944–955 (2025). (2022). Selvaraju, R. R. et al. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization. Int. J. Comput. Vis. 128 , 336–359 (2020). Additional Declarations No competing interests reported. Supplementary Files CWSEditorialCertificate.pdf Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 19 Apr, 2026 Reviewers agreed at journal 29 Mar, 2026 Reviewers invited by journal 29 Mar, 2026 Editor invited by journal 01 Dec, 2025 Editor assigned by journal 29 Nov, 2025 Submission checks completed at journal 29 Nov, 2025 First submitted to journal 28 Nov, 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-8229046\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Article\",\"associatedPublications\":[],\"authors\":[{\"id\":615398572,\"identity\":\"d238fd0b-58b4-470e-adff-f12eafb77a8a\",\"order_by\":0,\"name\":\"Yalei Wang\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Affiliated Fuyang People's Hospital of Anhui Medical University.\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Yalei\",\"middleName\":\"\",\"lastName\":\"Wang\",\"suffix\":\"\"},{\"id\":615398573,\"identity\":\"7639c8c0-74e2-45d9-b3a0-88f15a539227\",\"order_by\":1,\"name\":\"Fuqiang Pan\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Affiliated Fuyang People's Hospital of Bengbu Medical University.\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Fuqiang\",\"middleName\":\"\",\"lastName\":\"Pan\",\"suffix\":\"\"},{\"id\":615398574,\"identity\":\"397a163e-8419-4eb2-92f7-f24bc6aee36c\",\"order_by\":2,\"name\":\"Mengqing Kang\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Affiliated Fuyang People's Hospital of Anhui Medical University.\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Mengqing\",\"middleName\":\"\",\"lastName\":\"Kang\",\"suffix\":\"\"},{\"id\":615398575,\"identity\":\"23756923-aec2-40cc-b264-4d83dbf242c9\",\"order_by\":3,\"name\":\"Baoqi Zhang\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The Affiliated Fuyang People's Hospital of Anhui Medical University.\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Baoqi\",\"middleName\":\"\",\"lastName\":\"Zhang\",\"suffix\":\"\"},{\"id\":615398576,\"identity\":\"e73f51ae-de47-4983-a6fd-88480365ce59\",\"order_by\":4,\"name\":\"Yang Zhang\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxklEQVRIiWNgGAWjYBAC9gYGBmYQg5+Z+fADorTwHIBqkWxnSzMgTYvBeR4FCeK0SOQYfy6ouWO3+TAPgwFDjU00MVoMjGcce5a87TDvgQcMx9JyGwhpsQdqSeZhO5xsdpgvwYCx4TBhLSBbDvP8O5xs3MxjIEGsFsNm3rbDdgbMRGvheVbMzNt3OEHiMDCQE4jxCw978ubPPN8O2/P3Hz784EONDWEtDAIZ4AhMBKtMIKgcBPiPPwBR9kQpHgWjYBSMgpEJAK4VPUeLAKa2AAAAAElFTkSuQmCC\",\"orcid\":\"\",\"institution\":\"The Affiliated Fuyang People's Hospital of Anhui Medical University.\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Yang\",\"middleName\":\"\",\"lastName\":\"Zhang\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-11-28 09:53:24\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-8229046/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-8229046/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":106068933,\"identity\":\"9200e73e-2312-4f76-abed-1cd5b7d4e0bd\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 06:21:35\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":186665,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eStudy workflow.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/f56b794a4f4b8a081bced6d5.png\"},{\"id\":106068934,\"identity\":\"0853bb09-12e5-4584-b6de-6f479be509f3\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 06:21:35\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":113074,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eFlowchart of patient inclusion and exclusion.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/414d210d8f89322465d6ba2e.png\"},{\"id\":106094221,\"identity\":\"85261129-0ba2-434a-90c2-cb5f9acd5cb1\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 11:41:49\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":137341,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eROC curves for different signatures across cohorts.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/fbdf991c54492175524693a6.png\"},{\"id\":106094360,\"identity\":\"187d6751-fa65-4230-8d42-da20d61b096b\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 11:42:16\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":137155,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eCalibration curves of different signatures for the test cohort.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/b54775cff77652db4a4238ac.png\"},{\"id\":106068936,\"identity\":\"9faee73c-55bc-4ac1-94f4-b52db9060d5f\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 06:21:35\",\"extension\":\"png\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":155742,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eDecision curves of different signatures in the test cohort.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/b0d48adfde82caa0845a8a1c.png\"},{\"id\":106068938,\"identity\":\"5eb5d1e3-a893-41e3-9c32-91b704c93198\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 06:21:35\",\"extension\":\"png\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":123025,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eDeLong test results for different signatures.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image6.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/46eeb313fbf560d0dd40f8a2.png\"},{\"id\":106068939,\"identity\":\"6568a2d7-3100-43ab-9332-37f5b25f877b\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 06:21:35\",\"extension\":\"png\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":792603,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003e（A，B）Comparison of Grad-CAM heatmaps from 2.5D and 2Dmodels.\\u003c/strong\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"image7.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/9676a4f06a25aa976351c0ad.png\"},{\"id\":106959469,\"identity\":\"120a60d0-becd-44e5-b034-0edbefa624f8\",\"added_by\":\"auto\",\"created_at\":\"2026-04-15 09:10:06\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":2983655,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/c8fa1b6f-1be1-4686-9ee0-b7d32a41c24e.pdf\"},{\"id\":106068932,\"identity\":\"37ded14d-9435-421f-b4cf-1c9573c80d57\",\"added_by\":\"auto\",\"created_at\":\"2026-04-03 06:21:35\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":30147,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"CWSEditorialCertificate.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8229046/v1/3e0030224dbd4401b4b0f930.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"2D and 2.5D Deep Learning Models for Neoadjuvant Chemotherapy Response Prediction in Breast Cancer\",\"fulltext\":[{\"header\":\"INTRODUCTION\",\"content\":\"\\u003cp\\u003eBreast cancer remains a leading threat to women's health worldwide and one of the primary causes of cancer-related mortality among women, representing a major public health challenge in the 21st century [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e]. Neoadjuvant chemotherapy (NAC), an essential therapeutic strategy for locally advanced breast cancer, can effectively reduce tumor size, increase breast-conserving surgery rates, and provide valuable information for postoperative pathological assessment and individualized treatment planning [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e]. However, the response to NAC varies substantially among breast cancer patients, and some individuals may show minimal benefit or even experience disease progression during treatment [4]. Therefore, accurate prediction of NAC response prior to treatment is critical for optimizing therapeutic decision-making, avoiding overtreatment, and improving patient outcomes.\\u003c/p\\u003e \\u003cp\\u003eMagnetic resonance imaging (MRI), owing to its multiparametric and multimodal imaging capabilities, has been widely used for breast cancer screening, characterization, and treatment response assessment [\\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e5\\u003c/span\\u003e]. In particular, sequences such as DCE imaging, DWI, T2-weighted imaging, and ADC mapping provide complementary anatomical and functional information about breast tumors [6]. In recent years, artificial intelligence approaches such as radiomics and deep learning have offered new avenues for quantitative analysis and predictive modeling using breast MRI [7, 8]. Traditional radiomics methods rely on manually engineered features, are susceptible to variations in region of interest (ROI) delineation and feature selection algorithms, and often suffer from limited generalizability. In contrast, deep learning techniques\\u0026mdash;particularly convolutional neural networks (CNNs)\\u0026mdash;can automatically learn high-dimensional abstract representations from images without explicit feature engineering, achieving outstanding performance in tasks such as tumor detection, classification, and treatment response prediction [9].\\u003c/p\\u003e \\u003cp\\u003ePrevious studies have attempted to develop 2D deep learning models based on MRI for predicting NAC response in breast cancer [\\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e]. However, such models typically rely on single-slice images or single-modality inputs, thereby overlooking the spatial characteristics of the tumor and the role of the surrounding tumor microenvironment in treatment response. To address these limitations, a 2.5D strategy has recently been proposed, in which the largest cross-sectional slice of the tumor is combined with multimodal information to approximate three-dimensional representation, thereby enhancing the model\\u0026rsquo;s ability to capture intratumoral heterogeneity [\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e]. Moreover, imaging features of the peritumoral region have been shown to be strongly associated with NAC response [\\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e], yet they remain frequently overlooked in many conventional deep learning frameworks. Building on this background, the present study aimed to develop a 2.5D deep learning model that integrates multiple MRI modalities and to systematically compare its performance with that of conventional 2D models in predicting the response to NAC in breast cancer. In addition, the best-performing model is combined with key clinical features to construct an integrated analytical framework with enhanced predictive performance and practical clinical applicability. Figure\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e illustrates the overall workflow of this study.\\u003c/p\\u003e\"},{\"header\":\"RESULTS\",\"content\":\"\\u003cdiv id=\\\"Sec3\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eClinical characteristics\\u003c/h2\\u003e \\u003cp\\u003eA total of 187 breast cancer patients who received neoadjuvant chemotherapy (NAC) were included in this study, with 130 assigned to the training cohort and 57 to the test cohort. Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e summarizes the baseline characteristics of the two groups. No significant intergroup differences were observed regarding age, hormone receptor status (estrogen receptor (ER) and progesterone receptor (PR)), HER2 expression, Ki-67 level, molecular subtype, tumor type, axillary lymph node enlargement, edema grade, enhancement pattern, or spiculation signs (all \\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.05), indicating that the two cohorts were well balanced and suitable for subsequent model training and validation.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003eBaseline characteristics of the study cohorts.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"5\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003efeature_name\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003elabel\\u0026thinsp;=\\u0026thinsp;ALL\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003elabel\\u0026thinsp;=\\u0026thinsp;test\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003elabel\\u0026thinsp;=\\u0026thinsp;train\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e\\u003cem\\u003ep\\u003c/em\\u003e-value\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAge\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e50.24\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;10.37\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e49.11\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;9.92\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e50.74\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;10.56\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.101\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eER\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.798\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e104(55.61)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e33(57.89)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e71(54.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e83(44.39)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e24(42.11)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e59(45.38)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003ePR\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.0\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e137(73.26)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e42(73.68)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e95(73.08)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e50(26.74)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e15(26.32)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e35(26.92)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eHer2\\u003cem\\u003e\\\\nC_erb_2\\u003c/em\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.232\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e96(51.34)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e25(43.86)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e71(54.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e91(48.66)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e32(56.14)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e59(45.38)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eKi_67\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.74\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e123(65.78)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e36(63.16)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e87(66.92)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e64(34.22)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e21(36.84)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e43(33.08)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eMolecular_Subtype\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.546\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e42(22.46)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e10(17.54)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e32(24.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e52(27.81)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e16(28.07)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e36(27.69)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e60(32.09)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e18(31.58)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e42(32.31)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e4\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e33(17.65)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e13(22.81)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e20(15.38)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eTIC0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.319\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e126(67.38)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e40(70.18)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e86(66.15)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e56(29.95)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e17(29.82)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e39(30.00)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e5(2.67)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eNull\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e5(3.85)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eTumor_Type\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.0\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e56(29.95)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e17(29.82)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e39(30.00)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e131(70.05)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e40(70.18)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e91(70.00)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAxillary_Lymph_Node_Enlargement\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.772\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e29(15.51)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e10(17.54)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e19(14.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e158(84.49)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e47(82.46)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e111(85.38)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEnhancement_Pattern\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.615\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e137(73.26)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e44(77.19)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e93(71.54)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e50(26.74)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e13(22.81)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e37(28.46)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eSpiculation\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.414\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e67(35.83)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e24(42.11)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e43(33.08)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e120(64.17)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e33(57.89)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e87(66.92)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEdema0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.579\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e29(15.51)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e12(21.05)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e17(13.08)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e86(45.99)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e25(43.86)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e61(46.92)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e28(14.97)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e8(14.04)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e20(15.38)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e4\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e44(23.53)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e12(21.05)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e32(24.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEdema1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.582\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e1(0.53)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003enull\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e1(0.77)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e49(26.20)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e18(31.58)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e31(23.85)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e85(45.45)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e27(47.37)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e58(44.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e28(14.97)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e7(12.28)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e21(16.15)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e4\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e24(12.83)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e5(8.77)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e19(14.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEdema_Change\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.145\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e88(47.06)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e30(52.63)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e58(44.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e61(32.62)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e14(24.56)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e47(36.15)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003eFurther univariate and stepwise multivariate logistic regression analyses were performed to identify clinical predictors associated with pathological complete response (pCR) (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e). The results showed that spiculation (OR\\u0026thinsp;=\\u0026thinsp;0.266, 95% CI: 0.128\\u0026ndash;0.555, \\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.003) and edema grade Edema1 (OR\\u0026thinsp;=\\u0026thinsp;0.524, 95% CI: 0.329\\u0026ndash;0.836, \\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.023) were significant negative predictors of pCR, suggesting that both may hold important value for treatment response prediction. Age also showed borderline significance (\\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.086).\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003eUnivariable analysis of clinical features.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"8\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c8\\\" colnum=\\\"8\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003efeature_name\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eLog(OR)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003elower 95%CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eupper 95%CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003eOR\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003eOR lower 95%CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003eOR upper 95%CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003ep_value\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003ePR\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e-0.862\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-1.802\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.078\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.422\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.165\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e1.081\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.131\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eER\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e-0.545\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-1.317\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.227\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.580\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.268\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e1.255\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.245\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eSpiculation\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e-1.323\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-2.056\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e-0.589\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.266\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.128\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.555\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.003\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEdema_Change\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e-0.103\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-0.425\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.219\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.902\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.654\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e1.245\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.599\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEdema1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e-0.646\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-1.113\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e-0.179\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.524\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.329\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.836\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.023\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEdema0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.463\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-0.004\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.930\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.589\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.996\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e2.535\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.103\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAge\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.025\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.001\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.049\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e1.026\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e1.001\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e1.050\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.086\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003c/div\\u003e\\n\\u003ch3\\u003ePerformance of deep learning models\\u003c/h3\\u003e\\n\\u003cp\\u003eUnder the 2D and 2.5D input strategies, this study compared the predictive performance of four CNN models (VGG16BN, ResNet50, DenseNet121, and SqueezeNet1.0) on the training and independent test cohorts (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e). In the training cohort, the 2.5D ResNet50 model outperformed all other models, with an AUC of 0.919, an accuracy of 0.902, a sensitivity of 0.818, a specificity of 0.966, and an F1-score of 0.878. In the test cohort, the 2.5D ResNet50 model consistently maintained the best performance, achieving an AUC of 0.897 (95% CI: 0.7722\\u0026ndash;1.000), an accuracy of 0.895, and a sensitivity of 0.944, demonstrating strong generalization and stability. In contrast, the 2D models generally exhibited lower AUCs than the 2.5D models, with significant performance differences, suggesting that integrating multimodal image fusion and peritumoral features can enhance predictive performance.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab3\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 3\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003ePredictive performance of four CNNs in 2D and 2.5D frameworks.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"13\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c8\\\" colnum=\\\"8\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c9\\\" colnum=\\\"9\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c10\\\" colnum=\\\"10\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c11\\\" colnum=\\\"11\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c12\\\" colnum=\\\"12\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c13\\\" colnum=\\\"13\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eSignature\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eAccuracy\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eAUC\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e95% CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003eSensitivity\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003eSpecificity\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003ePPV\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003eNPV\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003ePrecision\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003eRecall\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003eF1\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003eThreshold\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003eCohort\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_vgg16bn\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.824\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.859\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7836\\u0026ndash;0.9343\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.659\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.948\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.906\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.786\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.906\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.659\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.763\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.476\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_resnet50\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.902\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.919\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.8578\\u0026ndash;0.9800\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.966\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.947\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.875\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.947\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.878\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.501\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_densenet121\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.725\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.752\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.6555\\u0026ndash;0.8477\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.591\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.828\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.722\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.727\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.722\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.591\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.650\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.490\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_squeezenet10\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.892\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7608\\u0026ndash;0.9390\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.948\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.923\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.873\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.923\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.867\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.486\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_vgg16bn\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.833\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.833\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7507\\u0026ndash;0.9162\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.932\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.759\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.745\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.936\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.745\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.932\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.828\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.500\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_resnet50\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.843\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.862\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7859\\u0026ndash;0.9375\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.864\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.828\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.792\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.889\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.792\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.864\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.826\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.512\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_densenet121\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.833\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.845\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7651\\u0026ndash;0.9254\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.750\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.897\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.846\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.825\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.846\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.750\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.795\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.466\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_squeezenet10\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.775\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.758\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.6567\\u0026ndash;0.8590\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.705\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.828\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.756\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.787\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.756\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.705\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.729\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.501\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_vgg16bn\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.763\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.756\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.5836\\u0026ndash;0.9275\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.722\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.800\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.765\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.762\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.765\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.722\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.743\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.462\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_resnet50\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.895\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.897\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7722\\u0026ndash;1.0000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.895\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.499\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_densenet121\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.816\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.767\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.6041\\u0026ndash;0.9293\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.611\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e1.000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e1.000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.741\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e1.000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.611\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.759\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.506\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D_squeezenet10\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.816\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.797\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.6411\\u0026ndash;0.9533\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.722\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.900\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.867\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.783\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.867\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.722\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.788\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.419\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_vgg16bn\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.711\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.700\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.5209\\u0026ndash;0.8791\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.667\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.750\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.706\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.714\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.706\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.667\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.686\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.496\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_resnet50\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.763\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.764\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.6029\\u0026ndash;0.9248\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.889\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.650\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.696\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.867\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.696\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.889\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.780\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.497\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_densenet121\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.737\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.675\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.4939\\u0026ndash;0.8561\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.611\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.786\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.708\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.786\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.611\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.687\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.502\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D_squeezenet10\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.684\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.681\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.5140\\u0026ndash;0.8471\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.500\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.750\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.654\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.750\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.500\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.600\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.500\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003ctfoot\\u003e \\u003ctr\\u003e\\u003ctd colspan=\\\"13\\\"\\u003eDL2D: Deep learning 2D model; DL25D: Deep learning 2.5D model; PPV: positive predictive value; NPV: negative predictive value.\\u003c/td\\u003e\\u003c/tr\\u003e \\u003c/tfoot\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab4\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 4\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003ePerformance metrics of clinical, 2D, 2.5D, and combined deep learning models.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"13\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c8\\\" colnum=\\\"8\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c9\\\" colnum=\\\"9\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c10\\\" colnum=\\\"10\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c11\\\" colnum=\\\"11\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c12\\\" colnum=\\\"12\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c13\\\" colnum=\\\"13\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eSignature\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eAccuracy\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eAUC\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e95% CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003eSensitivity\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003eSpecificity\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003ePPV\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003eNPV\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003ePrecision\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003eRecall\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003eF1\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003eThreshold\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003eCohort\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eClinical\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.676\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.698\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.5971\\u0026ndash;0.7991\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.477\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.828\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.677\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.676\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.677\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.477\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.560\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.556\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.902\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.919\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.8578\\u0026ndash;0.9800\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.966\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.947\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.875\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.947\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.878\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.501\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.843\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.862\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7859\\u0026ndash;0.9375\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.864\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.828\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.792\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.889\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.792\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.864\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.826\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.512\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eCombined\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.931\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.955\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.9133\\u0026ndash;0.9973\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.955\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.914\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.894\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.964\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.894\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.955\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.923\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.443\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etrain\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eClinical\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.632\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.654\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.4818\\u0026ndash;0.8265\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.350\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.567\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.875\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.567\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.708\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.308\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL25D\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.895\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.897\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7722\\u0026ndash;1.0000\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.850\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.944\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.895\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.499\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eDL2D\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.763\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.764\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.6029\\u0026ndash;0.9248\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.889\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.650\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.696\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.867\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.696\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.889\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.780\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.497\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eCombined\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.842\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.864\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.7378\\u0026ndash;0.9900\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.778\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e0.900\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e0.875\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e0.818\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e0.875\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e0.778\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c11\\\"\\u003e \\u003cp\\u003e0.824\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c12\\\"\\u003e \\u003cp\\u003e0.929\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c13\\\"\\u003e \\u003cp\\u003etest\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003ctfoot\\u003e \\u003ctr\\u003e\\u003ctd colspan=\\\"13\\\"\\u003eDL2D: Deep-learning 2D model; DL25D: Deep-learning 2.5D model.\\u003c/td\\u003e\\u003c/tr\\u003e \\u003c/tfoot\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e\\n\\u003ch3\\u003eEvaluation of the combined model and statistical comparison\\u003c/h3\\u003e\\n\\u003cp\\u003eA combined model was constructed by integrating the deep-learning probabilities generated by the 2.5D ResNet50 model with clinically significant variables (spiculation sign, edema grade, and age) using the ExtraTrees regression algorithm. In the training cohort, the combined model achieved an AUC of 0.955 (95% CI: 0.9133\\u0026ndash;0.9973). Moreover, the model demonstrated good calibration, with the Hosmer\\u0026ndash;Lemeshow test showing no significant lack of fit in either the training or test cohort (\\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.05), and the calibration curves exhibiting close agreement between predicted and observed outcomes (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e).\\u003c/p\\u003e \\u003cp\\u003eDecision curve analysis (DCA) indicated that the combined model provided the highest net benefit across most clinical decision thresholds (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e), further supporting its strong potential for clinical application. The DeLong test showed that the combined model achieved a significantly higher AUC compared to the clinical model and the 2D model (\\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.01), exceeding also that of the 2.5D model (p\\u0026thinsp;=\\u0026thinsp;0.047) (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e).\\u003c/p\\u003e\\n\\u003ch3\\u003eGrad-CAM visualization analysis\\u003c/h3\\u003e\\n\\u003cp\\u003eTo further enhance model interpretability and to examine the decision basis of the deep learning model in predicting the efficacy of NAC for breast cancer, we employed Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize the model\\u0026rsquo;s salient regions.\\u003c/p\\u003e \\u003cp\\u003eGrad-CAM propagates gradient information back to the final convolutional layer to generate heatmaps that highlight the image regions most influential for the model\\u0026rsquo;s classification decisions, thereby revealing the visual basis underlying the network\\u0026rsquo;s predictions. We selected representative pCR and non-pCR cases from both the training and test cohorts and compared the heatmap distributions generated by the 2.5D ResNet50 model with those produced by the 2D models (e.g., 2D VGG16BN). The results are presented in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003e.\\u003c/p\\u003e \\u003cp\\u003eThe analysis revealed that the Grad-CAM heatmaps generated by the 2.5D model were predominantly concentrated within the tumor parenchyma and its margins, and in some cases extended into adjacent non-neoplastic tissue and interstitial regions. This pattern suggests that the model may have captured microenvironmental features associated with treatment response, such as the tumor\\u0026ndash;stroma interface, vascular distribution, and edema spread. In contrast, the activation regions of the 2D models were often confined to the tumor core, with some cases exhibiting activation \\u0026ldquo;shifts\\u0026rdquo; or even \\u0026ldquo;blank responses,\\u0026rdquo; indicating limited ability to extract key features and suggesting instability in model decision-making.\\u003c/p\\u003e \\u003cp\\u003eMoreover, in some non-pCR patients, the Grad-CAM maps of the 2.5D model highlighted salient regions located in irregularly enhanced peripheral tumor areas, spiculated structures, and surrounding ADC-abnormal regions. This observation suggests that the model may have detected focal structural or functional abnormalities potentially associated with chemotherapy resistance.\\u003c/p\\u003e \\u003cp\\u003eTaken together with the preceding performance assessments, the Grad-CAM analysis further demonstrated from an interpretability perspective that the 2.5D architecture, through multimodal integration and enhanced spatial contextual modeling, improved the recognition of complex tumor phenotypes, thereby contributing to greater model stability and clinical credibility.\\u003c/p\\u003e\"},{\"header\":\"DISCUSSION\",\"content\":\"\\u003cdiv id=\\\"Sec8\\\" class=\\\"Section2\\\"\\u003e\\n\\u003ch2\\u003eKey findings\\u003c/h2\\u003e\\n\\u003cp\\u003eIn this study, we systematically developed and compared multiple deep learning models, proposing a 2.5D CNN architecture based on multimodal breast MRI inputs and validating its effectiveness and stability in predicting the response to NAC in breast cancer. Our main observations can be outlined as follows:\\u003c/p\\u003e\\n\\u003cp\\u003e1. The 2.5D model markedly outperformed the 2D models and demonstrated a substantially greater capacity for capturing contextual and spatially informative features. Compared with conventional 2D models that operate on single-view images, the 2.5D model achieved superior predictive performance in both the training and test cohorts, with particularly notable advantages in key metrics such as AUC, sensitivity, and F1 score. For example, using ResNet50, the 2.5D model achieved an AUC of 0.897 in the test cohort, substantially outperforming the 2D model (AUC\\u0026thinsp;=\\u0026thinsp;0.764) and demonstrating higher accuracy as well as more stable generalization performance.\\u003c/p\\u003e\\n\\u003cp\\u003eThe 2.5D model integrates seven imaging modalities\\u0026mdash;T1, T2WI, DWI, ADC, and early-, mid-, and late-phase DCE\\u0026mdash;thereby capturing multidimensional physiological information encompassing tumor anatomy, diffusion properties, and perfusion dynamics, while also enhancing the sensitivity of the model to temporal evolution and microenvironmental dynamics. This spatial\\u0026ndash;temporal multimodal integration mechanism substantially strengthened the model\\u0026rsquo;s capacity to characterize tumor heterogeneity, microcirculatory alterations, and their associations with chemotherapy response.\\u003c/p\\u003e\\n\\u003cp\\u003e2. Among the various network architectures evaluated, ResNet50 achieved the best overall performance, demonstrating strong generalization and stability. In this study, four classical CNN architectures\\u0026mdash;VGG16_bn [\\u003cspan class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e],ResNet50 [15], DenseNet121 [16], and SqueezeNet1.0 [17]\\u0026mdash;were selected for comparison. Under both the 2D and 2.5D input strategies, ResNet50 achieved the highest AUC and F1 scores and exhibited the least performance fluctuation between the training and test cohorts, indicating superior generalization capability.\\u003c/p\\u003e\\n\\u003cp\\u003eBy incorporating residual connections, ResNet50 effectively mitigates gradient vanishing and network degradation in deep architectures, facilitating more stable convergence during training and enhancing its capacity for feature abstraction and transfer learning [15]. In our study, ResNet50 not only preserved robust baseline feature extraction capabilities but also more precisely captured high-level treatment-related imaging features, demonstrating superior discriminative power particularly in cases characterized by peripheral tumor enhancement, ADC alterations, and pronounced intratumoral heterogeneity.\\u003c/p\\u003e\\n\\u003cp\\u003e3. The combined model integrates clinical variables with imaging-derived deep-learning features, further improving predictive performance and enhancing model interpretability.\\u003c/p\\u003e\\n\\u003cp\\u003eNotably, although the combined model achieved the best performance in the training cohort (AUC\\u0026thinsp;=\\u0026thinsp;0.955), this advantage did not persist in the independent test cohort, where the 2.5D ResNet50 model alone demonstrated the highest performance. This discrepancy may be attributable to several factors. First, the combined model effectively integrated the complementary information between deep imaging features and clinical variables during training, resulting in a higher degree of model fit. However, the distribution of clinical variables (e.g., spiculation and edema grade) may vary across centers, and their stability and generalizability are inferior to those of imaging features, which likely attenuated the overall predictive performance during external validation. Second, although the ExtraTrees algorithm, which has strong nonlinear modeling capacity, captured complex interactions between imaging features and clinical variables in the training set, its dependence on sample-specific clinical variables may have contributed to reduced stability when applied to new data. In contrast, the 2.5D ResNet50 model, which relies primarily on deep imaging phenotypes derived from multimodal MRI sequences, exhibited greater feature stability across centers and consequently demonstrated superior robustness and reproducibility in the independent test cohort. These findings suggest that deep imaging features may serve as more reliable predictors across populations and clinical centers, whereas the contribution of clinical variables to model fusion requires further validation and optimization in larger, multicenter datasets.\\u003c/p\\u003e\\n\\u003c/div\\u003e\\n\\u003ch3\\u003eComparison with existing studies\\u003c/h3\\u003e\\n\\u003cp\\u003eIn recent years, the application of artificial intelligence in breast cancer imaging has advanced rapidly. In particular, multiple studies have explored the development of MRI-based deep-learning and radiomics models for predicting the response to NAC. However, existing studies share several limitations, including reliance on single-modality imaging, restricted input dimensionality, limited model generalizability, and the absence of integrated clinical features [18\\u0026ndash;20].\\u003c/p\\u003e\\n\\u003cp\\u003eRegarding image input strategies, most existing studies have adopted two-dimensional (2D) deep-learning frameworks, typically constructing predictive models from the largest tumor cross-section on DCE or T2WI images. For example, Joo et al. developed a deep-learning model integrating subtraction T1-weighted images, T2-weighted images, and clinical variables to predict pCR prior to NAC. The model achieved an AUC of 0.888 in the validation cohort, significantly outperforming the clinical-only model (AUC\\u0026thinsp;=\\u0026thinsp;0.827, \\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05) [\\u003cspan class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e]. Although such approaches yield performance improvements, they still fail to comprehensively capture the spatial heterogeneity and dynamic enhancement patterns of the tumor. Some studies have attempted 2D multimodal fusion strategies, in which images from different MRI sequences are fed into the model and concatenated at the feature level to enhance predictive performance. For instance, Fan et al. extracted radiomics features from DCE-MRI, DWI, and T2WI images and performed multiparametric fusion, which markedly outperformed single-modality models in predicting NAC response [\\u003cspan class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e]. Although this strategy improves spatial coverage, it remains constrained by its reliance on single-slice planar inputs and therefore cannot overcome the inherent limitations in representing tumor eco-structural complexity and heterogeneity.\\u003c/p\\u003e\\n\\u003cp\\u003eIn contrast, the 2.5D deep-learning strategy proposed in this study substantially enhances the modeling capacity for multidimensional tumor phenotypes. By selecting the largest tumor cross-section and extracting the corresponding slice across multiple MRI modalities, the method achieves a \\u0026ldquo;2D input, 3D-aware\\u0026rdquo; modeling paradigm. This strategy preserves computational efficiency while markedly improving the model\\u0026rsquo;s ability to capture complex structures such as tumor morphological boundaries and microenvironmental features. Consequently, it achieves substantial performance gains, with an AUC of 0.897 in the test cohort\\u0026mdash;surpassing most reported 2D models.\\u003c/p\\u003e\\n\\u003cp\\u003eIn exploring model architectures, this study systematically compared four representative CNN architectures. Through multiple rounds of experimental evaluation, ResNet50 demonstrated the most favorable stability and performance, consistent with findings reported in previous studies. Peng et al. [23] reported that a ResNet50-based deep-learning model outperformed traditional radiomics approaches in predicting pCR to NAC using preoperative DCE-MRI. Their findings highlight that deep residual networks can effectively extract high-level imaging features associated with treatment response, thereby achieving superior generalizability and reliability in cross-cohort analyses.\\u003c/p\\u003e\\n\\u003cp\\u003eCollectively, these results further support the potential of ResNet50 as a backbone architecture for breast MRI-based predictive modeling, particularly in the context of NAC response assessment. In the study by Islam et al. [\\u003cspan class=\\\"CitationRef\\\"\\u003e24\\u003c/span\\u003e], a ResNet50-based breast lesion classification model achieved a significantly improved AUC (0.866\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;0.015) after the integration of a convolutional block attention module further demonstrating the strong potential of ResNet50 in breast disease\\u0026ndash;related imaging analysis. Our study further verifies its extensibility under a 2.5D input strategy and demonstrates enhanced cross-center generalizability.\\u003c/p\\u003e\\n\\u003cp\\u003eNotably, our study found a significant association between the emergence of a peritumoral edema pattern (edema1) after the first cycle of NAC and the achievement of pCR, suggesting that the presence or evolution of edema may reflect an early treatment response. Previous studies have incorporated peritumoral imaging features as key imaging markers or radiomic predictors of NAC response [25], and have reported their added value in predicting pCR [26]. However, other studies have reported that pre-treatment peritumoral edema is associated with poor prognosis, such as higher recurrence rates and reduced survival, suggesting that the prognostic implication of edema may depend on the timing of assessment (pre-treatment vs on-treatment/post-treatment follow-up) and its dynamic evolution [27]. From a pathobiological perspective, peritumoral edema may reflect local inflammatory responses, increased vascular permeability, or stromal alterations within the tumor microenvironment. When these changes emerge or intensify shortly after the initiation of chemotherapy, they may indicate chemotherapy-induced tumor cell necrosis, vascular permeability alterations, or immune-cell infiltration, thereby correlating with a higher likelihood of achieving pCR. Based on this hypothesis, short-interval imaging follow-up\\u0026mdash;for example, after one chemotherapy cycle\\u0026mdash;to monitor dynamic changes in peritumoral edema may serve as a feasible imaging biomarker for early treatment response assessment and therapy optimization.\\u003c/p\\u003e\\n\\u003ch3\\u003eModel interpretability and clinical value\\u003c/h3\\u003e\\n\\u003cp\\u003eIn the application of artificial intelligence\\u0026mdash;particularly deep learning\\u0026mdash;to medical imaging, model interpretability remains one of the primary barriers to clinical translation. To enhance the transparency and trustworthiness of the proposed model, we incorporated Gradient-weighted Class Activation Mapping (Grad-CAM) [\\u003cspan class=\\\"CitationRef\\\"\\u003e28\\u003c/span\\u003e] to spatially visualize the decision-making process of the deep learning network. The Grad-CAM heatmaps demonstrated that, when predicting pCR, the 2.5D ResNet50 model consistently focused on the tumor core, tumor margins, and adjacent tissues. In some cases, activation extended to the tumor\\u0026ndash;stroma interface and regions of abnormal enhancement, suggesting that the model captured intrinsic tumor features and learned microenvironmental characteristics associated with treatment response. These findings support the advantage and reliability of the 2.5D model in predicting NAC response from an interpretability perspective, while providing clinicians with an intuitive understanding of the output of the model.\\u003c/p\\u003e\\n\\u003cdiv id=\\\"Sec11\\\" class=\\\"Section2\\\"\\u003e\\n\\u003ch2\\u003eLimitations and future directions\\u003c/h2\\u003e\\n\\u003cp\\u003eAlthough this study provides several innovative insights, it has certain limitations. First, the sample size was relatively limited and data were collected from only two institutions, which may not fully reflect variability across broader populations or imaging platforms. Future studies should incorporate larger, multicenter, and prospective cohorts to validate the generalizability of the model. Second, the model incorporated only MRI imaging and basic clinical variables, without integrating pathological findings, molecular subtypes, or multi-omics data. Incorporating multimodal biological information may further enhance predictive performance. Finally, the study did not assess the model\\u0026rsquo;s sensitivity across different chemotherapy regimens (e.g., platinum-based vs. non\\u0026ndash;platinum-based). Subgroup analyses are warranted to further improve the model\\u0026rsquo;s applicability in diverse therapeutic settings.\\u003c/p\\u003e\\n\\u003cp\\u003eFuture research directions include establishing large-scale, standardized, multicenter datasets; developing more interpretable multimodal architectures; integrating imaging data with multi-omics information for joint modeling; and exploring clinically deployable intelligent tools to facilitate real-world implementation.\\u003c/p\\u003e\\n\\u003c/div\\u003e\"},{\"header\":\"CONCLUSION\",\"content\":\"\\u003cp\\u003eThis study systematically compared the performance of 2D and 2.5D deep-learning models for predicting the response to NAC in breast cancer using breast MRI, and further compared four representative CNN architectures. The results indicate that the 2.5D multichannel model not only captures anatomical, diffusion- and perfusion-related information, but also enhances the model\\u0026rsquo;s awareness of temporal evolution and dynamic tumor microenvironment features, yielding superior overall performance compared with 2D models in predicting pCR. Among the evaluated architectures, ResNet50 demonstrated the highest stability and predictive accuracy, consistent with previous breast MRI deep-learning studies and supporting the robustness of this architecture.\\u003c/p\\u003e \\u003cp\\u003eTaken together, the findings confirm the effectiveness and feasibility of deep-learning approaches for NAC response prediction and highlight the critical role of multidimensional imaging features and peritumoral microenvironmental information in model performance. Future work integrating large-scale multicenter datasets with multimodal and multi-omics information may further enhance model generalizability and clinical utility, ultimately providing more precise tools for individualized NAC response assessment and therapeutic decision-making.\\u003c/p\\u003e\"},{\"header\":\"MATERIALS AND METHODS\",\"content\":\"\\u003cp\\u003e \\u003cb\\u003e3.1 Patient cohort and data partitioning\\u003c/b\\u003e \\u003c/p\\u003e \\u003cp\\u003eThis multicenter retrospective study included 187 patients with breast cancer (mean age, 50.24\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;10.3 years) from two tertiary medical centers, comprising 130 patients from Hospital A (Fuyang Cancer Hospital; 50.74\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;10.56 years) and 57 patients from Hospital B (Fuyang People\\u0026rsquo;s Hospital; 49.11\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;9.92 years). The inclusion period spanned from January 2020 to December 2024. All enrolled patients underwent breast MRI examinations both before and after neoadjuvant chemotherapy (NAC), with image quality sufficient for analysis and complete pathological and follow-up data available. The study was approved by the institutional review boards of both centers, and informed consent was waived for all participants. Inclusion criteria included: histopathologically confirmed invasive breast cancer; receipt of NAC with a complete and standardized treatment regimen; availability of pre-treatment breast MRI, including dynamic contrast-enhanced (DCE), T2-weighted imaging (T2WI), and diffusion-weighted imaging (DWI) sequences, as well as apparent diffusion coefficient (ADC) maps; accessible postoperative pathological reference standard for assessing treatment response (e.g., presence or absence of pathological complete response, pCR); and MRI images without severe artifacts, allowing clear identification of the ROI. Exclusion criteria included: incomplete MRI examinations or severe motion artifacts; prior systemic treatments (chemotherapy, radiotherapy, or targeted therapy); unavailable postoperative pathology or incomplete follow-up data; coexistence of other major systemic diseases (e.g., malignancies or severe cardiopulmonary disorders); or tumors too small to allow reliable ROI delineation on MRI (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e).\\u003c/p\\u003e \\u003cp\\u003e To enhance the clinical generalizability of the model, the dataset was partitioned according to its natural institutional distribution: data from Center 1 were used for model training, whereas data from Center 2 served as an independent validation cohort. Clinical information\\u0026mdash;including age, hormone receptor status, molecular subtype, Ki-67 index, enhancement pattern, spiculation, and edema characteristics\\u0026mdash;was extracted and cross-checked from the electronic medical record system.\\u003c/p\\u003e \\u003cdiv id=\\\"Sec14\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eImage acquisition and processing\\u003c/h2\\u003e \\u003cp\\u003eEleven patients underwent breast MRI according to standard clinical imaging protocols, including five major acquisitions: T1WI, T2WI, DWI, ADC mapping, and DCE imaging. DCE images were acquired in multiple post-contrast phases. Acquisition parameters, including slice thickness, repetition time/echo time (TR/TE), and field of view, followed the standardized protocols of each center to ensure adequate spatial and temporal resolution for assessing tumor morphology and enhancement kinetics.\\u003c/p\\u003e \\u003cp\\u003eIn MRI, low-frequency intensity non-uniformities, also known as bias field artifacts, commonly arise from scanner hardware limitations and variations in tissue magnetic susceptibility. These artifacts lead to inconsistent tissue intensities, thereby impairing reliable feature extraction and negatively affecting the performance of deep learning models. To enhance intensity consistency and improve model robustness, N4 bias field correction was applied to all MRI sequences prior to model input.\\u003c/p\\u003e \\u003cp\\u003eTo ensure spatial consistency across multimodal images and facilitate subsequent ROI extraction and model construction, all MRI sequences underwent standardized registration. Specifically, the early-phase DCE-MRI (DCE\\u0026ndash;mid phase) was designated as the fixed image, whereas T1WI, T2WI, DWI, ADC, and the remaining DCE phases (intermediate and delayed) were treated as moving images. Registration was performed with minimal structural deformation to align all sequences within a unified coordinate system, thereby preserving spatial correspondence of anatomical structures across modalities. After registration, two radiologists reviewed all images to confirm accurate alignment of key anatomical structures (tumor, glandular tissue, skin, and pectoral muscle) across modalities.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec15\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eROI segmentation and cropping\\u003c/h2\\u003e \\u003cp\\u003eFor all patients, ROIs were manually delineated on early-phase DCE images by two radiologists with expertise in breast imaging, using ITK-SNAP (version 3.8.0). In cases of substantial disagreement between the two readers, the final ROI was adjudicated by a senior breast radiologist with more than 20 years of clinical experience. The images and corresponding ROIs were rigidly registered across all modalities, followed by cropping of the matched regions to ensure consistency of multimodal inputs.\\u003c/p\\u003e \\u003cp\\u003eTo enhance the model\\u0026rsquo;s learning of both the tumor core and its surrounding microenvironment, the largest cross-sectional slice was selected as the representative image for each patient, following previous studies [7\\u0026ndash;8]. Based on the minimum bounding rectangle of the ROI, an additional 10-pixel margin was expanded in all directions to preserve peritumoral contextual information. All images were then uniformly cropped to a fixed size of 224 \\u0026times; 224 pixels and subjected to intensity normalization to ensure consistency across model inputs.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec16\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eMultimodal image fusion\\u003c/h2\\u003e \\u003cp\\u003eTo fully exploit the structural, functional, and dynamic enhancement information contained in different breast MRI sequences, a seven-channel input configuration was constructed, comprising T1WI, T2WI, DWI, ADC, and three DCE-MRI phases (early, intermediate, and delayed). All modalities were registered and cropped on the same tumor slice and ROI level to ensure spatial alignment and anatomical comparability across channels. This channel configuration maintains architectural simplicity while maximally integrating tumor morphological information (T1WI, T2WI), water-molecule diffusion characteristics (DWI, ADC), and hemodynamic alterations (DCE). Together, these modalities provide a rich and complementary phenotypic representation of the tumor. The synergistic integration of these multimodal inputs enhances the model\\u0026rsquo;s ability to capture biological heterogeneity associated with NAC response. This approach addresses the limitations of single-modality methods in spatial context, temporal evolution, and microenvironment perception, ultimately offering a more comprehensive feature foundation for deep learning.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec17\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eDeep learning modeling\\u003c/h2\\u003e \\u003cp\\u003e \\u003cb\\u003eModel architecture and input strategy.\\u003c/b\\u003e This study developed both 2D and 2.5D deep learning models to systematically compare their performance and representational capacity in predicting the therapeutic response to NAC in breast cancer. For the 2D models, we adopted a conventional single-modality input strategy, using the largest tumor slice from the early-phase DCE-MRI as the input to construct a standard two-dimensional CNN framework. The 2.5D models incorporated a multimodal and multi-temporal fusion strategy, employing a seven-channel input configuration that included T1WI, T2WI, DWI, ADC, and early-, mid-, and late-phase DCE-MRI images. All images were extracted and spatially registered on the basis of the largest tumor slice to ensure anatomical alignment and contextual consistency across modalities. This input strategy integrates multidimensional information\\u0026mdash;including anatomical structure, tissue diffusion properties, and perfusion dynamics\\u0026mdash;thereby providing a more comprehensive tumor phenotypic representation and enhancing the model\\u0026rsquo;s ability to capture and model NAC-related response heterogeneity.\\u003c/p\\u003e \\u003cp\\u003eIn terms of network architecture, four classical CNN backbones\\u0026mdash;VGG16_bn, ResNet50, DenseNet121, and SqueezeNet1_0\\u0026mdash;were adopted as the foundational models. All models were implemented using the PyTorch 1.11.0 deep learning framework and initialized with ImageNet-pretrained weights to facilitate transfer learning, thereby improving convergence efficiency and mitigating overfitting associated with the relatively limited size of medical imaging datasets.\\u003c/p\\u003e \\u003cp\\u003e \\u003cb\\u003eTraining and optimization strategy.\\u003c/b\\u003e A transfer-learning strategy was applied during model training, where all networks were fine-tuned from ImageNet-pretrained weights to preserve low-level feature extraction while enhancing task-specific feature representation. Stochastic Gradient Descent (SGD) was used as the optimizer with an initial learning rate of 0.01. A cosine-annealing schedule was employed to dynamically adjust the learning rate according to the following formulation:\\u003cdiv id=\\\"Equa\\\" class=\\\"Equation\\\"\\u003e\\u003cdiv format=\\\"TEX\\\" class=\\\"mathdisplay\\\" id=\\\"FileID_Equa\\\" name=\\\"EquationSource\\\"\\u003e\\n$$\\\\:{\\\\eta\\\\:}_{t}\\\\:=\\\\:{\\\\eta\\\\:}_{min}^{i}+\\\\frac{1}{2}({\\\\eta\\\\:}_{max}^{i}-{\\\\eta\\\\:}_{min}^{i})(1+cos(\\\\frac{{T}_{cur}}{{T}_{i}}\\\\pi\\\\:\\\\left)\\\\right)$$\\u003c/div\\u003e\\u003c/div\\u003e\\u003c/p\\u003e \\u003cp\\u003eIn this formulation, \\u003cspan class=\\\"InlineEquation\\\"\\u003e\\u003cspan class=\\\"mathinline\\\"\\u003e\\\\(\\\\:{\\\\eta\\\\:}_{min}^{i}\\\\)\\u003c/span\\u003e\\u003c/span\\u003e = 0 denotes the minimum learning rate, whereas \\u003cspan class=\\\"InlineEquation\\\"\\u003e\\u003cspan class=\\\"mathinline\\\"\\u003e\\\\(\\\\:{\\\\eta\\\\:}_{max}^{i}\\\\)\\u003c/span\\u003e\\u003c/span\\u003e = 0.01 specifies the maximum learning rate. \\u003cspan class=\\\"InlineEquation\\\"\\u003e\\u003cspan class=\\\"mathinline\\\"\\u003e\\\\(\\\\:{T}_{i}\\\\)\\u003c/span\\u003e\\u003c/span\\u003e = 32 represents the total number of epochs in the training cycle. SGD was used as the optimizer, and the Softmax cross-entropy loss function was employed during training.\\u003c/p\\u003e \\u003cp\\u003e \\u003cb\\u003eModel evaluation metrics and validation.\\u003c/b\\u003e After training, all models were evaluated on an independent test set, with the area under the receiver operating characteristic curve (ROC-AUC) used as the primary performance metric. Additional metrics\\u0026mdash;including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1 score, precision, and recall\\u0026mdash;were also reported. The confidence interval of the AUC was estimated using the DeLong method.\\u003c/p\\u003e \\u003cp\\u003eTo further assess the stability and generalizability of the models, calibration curves were generated separately for the training and test sets, and the goodness of fit of the predicted probabilities was evaluated using the Hosmer\\u0026ndash;Lemeshow test. In addition, DCA was performed to evaluate the net clinical benefit of each model across a range of decision thresholds.\\u003c/p\\u003e \\u003cp\\u003e \\u003cb\\u003eModel interpretability via Grad-CAM.\\u003c/b\\u003e To enhance the clinical interpretability of the models, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to visualize the trained networks. By extracting activation heatmaps from the final convolutional layer, we examined whether the model\\u0026rsquo;s attention regions overlapped with the tumor core and surrounding critical structures, thereby supporting the assessment of the rationality of the model\\u0026rsquo;s predictive decisions.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec18\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eFusion strategy and statistical analysis\\u003c/h2\\u003e \\u003cp\\u003e \\u003cb\\u003eClinical feature selection and modeling.\\u003c/b\\u003e To investigate the independent predictive value of clinical characteristics in assessing the response to NAC in breast cancer, univariate and multivariate regression analyses were first performed on all clinical variables. These included age, ER status, PR status, HER2 status, Ki-67 index, molecular subtype, enhancement pattern, spiculation, the degree and change of edema, axillary lymph node enlargement, and other variables, totaling 15 predictors. For each variable, the odds ratio (OR) and its 95% confidence interval (95% CI) for predicting pCR were calculated, and the corresponding p-values were recorded.\\u003c/p\\u003e \\u003cp\\u003eVariables with a screening threshold of p\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.10 were included in subsequent fusion-model analyses to reduce the risk of omitting potentially relevant predictors. Ultimately, significant variables such as spiculation, edema (Edema1), and age were incorporated into the clinical feature model (Clinical Signature). The performance of the clinical model was evaluated using the AUC to assess its independent predictive capability.\\u003c/p\\u003e \\u003cp\\u003e \\u003cb\\u003eFusion model construction.\\u003c/b\\u003e To further improve the predictive accuracy and clinical applicability of the models, we extended the fusion framework between deep learning features and clinical characteristics. Specifically, the best-performing 2.5D ResNet50 and 2D ResNet50 models were selected, and their output probabilities (the \\u0026ldquo;Deep Learning Signature\\u0026rdquo;) were combined with the previously identified significant clinical variables\\u0026mdash;spiculation, edema grade, and age\\u0026mdash;to construct a combined model. For model development, the combined model was trained using the Extremely Randomized Trees (ExtraTrees) algorithm.\\u003c/p\\u003e \\u003cp\\u003e \\u003cb\\u003ePerformance evaluation.\\u003c/b\\u003e The performance of all models\\u0026mdash;the clinical model, deep learning models, and the combined model\\u0026mdash;was evaluated separately on the training and test sets. The primary evaluation metrics included ROC-AUC analysis with 95% CI estimates, as well as sensitivity, specificity, accuracy, PPV, NPV, F1 score, precision, and recall.\\u003c/p\\u003e \\u003cp\\u003eIn addition, to assess the calibration of the predicted probabilities, calibration curves were generated, and the Hosmer\\u0026ndash;Lemeshow test was performed to evaluate statistical goodness of fit. A well-calibrated model should demonstrate close agreement between the predicted probabilities and the observed event rates (\\u003cem\\u003ep\\u003c/em\\u003e\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.05).\\u003c/p\\u003e \\u003cp\\u003e \\u003cb\\u003eModel comparison and statistical testing.\\u003c/b\\u003e To determine whether the performance differences between models were statistically significant, pairwise comparisons of the AUC values were conducted using the DeLong test. A \\u003cem\\u003ep\\u003c/em\\u003e-value\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05 was considered to indicate a significant difference in predictive performance between the two considered models.\\u003c/p\\u003e \\u003cp\\u003eFurthermore, DCA was used to evaluate the net benefit of each model across different threshold probabilities, thereby assisting in determining the clinical decision-making value of the models in real-world settings. A higher net benefit curve indicates greater clinical utility of the model over a broader range of threshold probabilities.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec19\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003eStatistical analysis\\u003c/h2\\u003e \\u003cp\\u003eThe Shapiro\\u0026ndash;Wilk test was used to assess the normality of clinical variables. Depending on the distribution of continuous variables, either the t-test or the Mann\\u0026ndash;Whitney U test was applied. Categorical variables were analyzed using the chi-square (χ\\u0026sup2;) test.\\u003c/p\\u003e \\u003cp\\u003eAll data analyses were performed using the OnekeyAI platform (version 4.9.1) using Python 3.7.12. Statistical evaluations were conducted with Statsmodels 0.13.2. Machine learning implementations, including support vector machines (SVM), were carried out using Scikit-learn 1.0.2. The deep learning framework was developed with PyTorch 1.11.0 and optimized with CUDA 11.3.1 and cuDNN 8.2.1 to enhance computational performance.\\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e \\u003ch2\\u003eEthics statement\\u003c/h2\\u003e \\u003cp\\u003eResearch involving human participants has been approved by the Medical Ethics Committee of Fuyang People's Hospital. The study was conducted in accordance with local laws and institutional requirements. In accordance with national laws and institutional regulations, written informed consent was not required from participants or their legal guardians/next of kin. As all investigations involving human participants have been reviewed and approved by the Institutional Review Board of Fuyang People's Hospital (Affiliated to Anhui Medical University) (Approval No. [2025]4), the ethics committee has waived the requirement for written informed consent.\\u003c/p\\u003e \\u003c/p\\u003e\\u003cp\\u003e \\u003ch2\\u003eCompeting interests\\u003c/h2\\u003e \\u003cp\\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\\u003c/p\\u003e \\u003c/p\\u003e\\u003cp\\u003e \\u003ch2\\u003ePublisher\\u0026rsquo;s note\\u003c/h2\\u003e \\u003cp\\u003e All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.\\u003c/p\\u003e \\u003c/p\\u003e\\u003ch2\\u003eFunding Statement\\u003c/h2\\u003e \\u003cp\\u003eThe authors declare that this research and/or the publication of this article has received funding. This work was supported by the 2023 Fuyang Health and Medical Research Project (Grant No. FY2023-005).\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eYW\\u0026amp;FP\\u0026amp;MK: Data curation, Writing\\u0026ndash; original draft, Software. BZ: Data curation, Writing original draft. YZ: Funding acquisition, Project administration, Resources, Writing\\u0026ndash; review \\u0026amp; editing(YW, FP, and MK contributed equally to the work)\\u003c/p\\u003e\\u003ch2\\u003eData Availability\\u003c/h2\\u003e\\u003cp\\u003eThe raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003eBray, F. et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. \\u003cem\\u003eCA Cancer J. Clin.\\u003c/em\\u003e \\u003cb\\u003e74\\u003c/b\\u003e, 229\\u0026ndash;263 (2024).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eRastogi, P. et al. Preoperative Chemotherapy: Updates of National Surgical Adjuvant Breast and Bowel Project Protocols B-18 and B-27. \\u003cem\\u003eJCO\\u003c/em\\u003e 26, 778\\u0026ndash;785 (2008).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eVaidya, J. S. et al. Rethinking neoadjuvant chemotherapy for breast cancer. \\u003cem\\u003eBMJ.\\u003c/em\\u003e j5913 (2018). Janssen, L. M. et al. Predicting response to neoadjuvant chemotherapy with liquid biopsies and multiparametric MRI in patients with breast cancer. \\u003cem\\u003enpj Breast Cancer\\u003c/em\\u003e 10, 10 (2024). (5913).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWekking, D. et al. Breast MRI: Clinical Indications, Recommendations, and Future Applications in Breast Cancer Diagnosis. \\u003cem\\u003eCurr. Oncol. Rep.\\u003c/em\\u003e \\u003cb\\u003e25\\u003c/b\\u003e, 257\\u0026ndash;267 (2023).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKazama, T., Takahara, T. \\u0026amp; Hashimoto, J. Breast Cancer Subtypes and Quantitative Magnetic Resonance Imaging: A Systemic Review. \\u003cem\\u003eLife\\u003c/em\\u003e \\u003cb\\u003e12\\u003c/b\\u003e, 490 (2022).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLu, G. et al. Deep learning radiomics based on multimodal imaging for distinguishing benign and malignant breast tumours. \\u003cem\\u003eFront. Med.\\u003c/em\\u003e \\u003cb\\u003e11\\u003c/b\\u003e, 1402967 (2024).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHu, Q. et al. A deep learning methodology for improved breast cancer diagnosis using multiparametric MRI. Sci. Rep. 10, 10536 Khosravi, P., Kazemi, E., Imielinski, M., Elemento, O. \\u0026amp; Hajirasouliha, I. Deep Convolutional Neural Networks Enable Discrimination of Heterogeneous Digital Pathology Images. \\u003cem\\u003eEBioMedicine\\u003c/em\\u003e 27, 317\\u0026ndash;328 (2018). Choi, R., Joel, M., Hui, M. \\u0026amp; Aneja, S. Deep learning algorithm to predict pathologic complete response to neoadjuvant chemotherapy for breast cancer prior to treatment. \\u003cem\\u003eJCO\\u003c/em\\u003e 40, 600\\u0026ndash;600 (2022). (2020).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLa Saint-Esteven, G. A 2.5D convolutional neural network for HPV prediction in advanced oropharyngeal cancer. \\u003cem\\u003eComput. Biol. Med.\\u003c/em\\u003e \\u003cb\\u003e142\\u003c/b\\u003e, 105215 (2022).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWang, Y. et al. Assessment of prostate cancer aggressiveness through the combined analysis of prostate MRI and 2.5D deep learning models. Front. Oncol. 15, 1539537 Cromb\\u0026eacute;, A. et al. MRI assessment of surrounding tissues in soft-tissue sarcoma during neoadjuvant chemotherapy can help predicting response and prognosis. \\u003cem\\u003eEur. J. Radiol.\\u003c/em\\u003e 109, 178\\u0026ndash;187 (2018). (2025).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSimonyan, K. \\u0026amp; Zisserman, A. \\u003cem\\u003eVery Deep Convolutional Networks for Large-Scale Image Recognition\\u003c/em\\u003e. (2015).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHe, K., Zhang, X., Ren, S. \\u0026amp; Sun, J. Deep Residual Learning for Image Recognition. in \\u003cem\\u003eIEEE Conference on Computer Vision and Pattern Recognition (CVPR)\\u003c/em\\u003e 770\\u0026ndash;778 (IEEE, 2016). 770\\u0026ndash;778 (IEEE, 2016). (2016). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003e10.1109/cvpr.2016.90\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/cvpr.2016.90\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHuang, G., Liu, Z., Van Der Maaten, L. \\u0026amp; Weinberger, K. Q. Densely Connected Convolutional Networks. in \\u003cem\\u003eIEEE Conference on Computer Vision and Pattern Recognition (CVPR)\\u003c/em\\u003e 2261\\u0026ndash;2269 (IEEE, 2017). 2261\\u0026ndash;2269 (IEEE, 2017). (2017). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003e10.1109/cvpr.2017.243\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/cvpr.2017.243\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eIandola, F. N. et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and \\u0026lt;\\u0026thinsp;0.5MB model size. Choi, R., Joel, M., Hui, M. \\u0026amp; Aneja, S. Deep learning algorithm to predict pathologic complete response to neoadjuvant chemotherapy for breast cancer prior to treatment. \\u003cem\\u003eJCO\\u003c/em\\u003e 40, 600\\u0026ndash;600 (2022). Lo Gullo, R., Eskreis-Winkler, S., Morris, E. A. \\u0026amp; Pinker, K. Machine learning with multiparametric magnetic resonance imaging of the breast for early prediction of response to neoadjuvant chemotherapy. \\u003cem\\u003eThe Breast\\u003c/em\\u003e 49, 115\\u0026ndash;122 (2020). (2016).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGranzier, R. W. Y., van Nijnatten, T. J. A., Woodruff, H. C., Smidt, M. L. \\u0026amp; Lobbes, M. B. I. Exploring breast cancer response prediction to neoadjuvant systemic therapy using MRI-based radiomics: A systematic review. \\u003cem\\u003eEur. J. Radiol.\\u003c/em\\u003e \\u003cb\\u003e121\\u003c/b\\u003e, 108736 (2019).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eJoo, S. et al. Multimodal deep learning models for the prediction of pathologic response to neoadjuvant chemotherapy in breast cancer. \\u003cem\\u003eSci. Rep.\\u003c/em\\u003e \\u003cb\\u003e11\\u003c/b\\u003e, 18800 (2021).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFan, M. et al. Multiparametric MRI radiomics fusion for predicting the response and shrinkage pattern to neoadjuvant chemotherapy in breast cancer. Front. Oncol. 13, 1057841 Peng, Y. et al. Pretreatment DCE-MRI-Based Deep Learning Outperforms Radiomics Analysis in Predicting Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast Cancer. \\u003cem\\u003eFront. Oncol.\\u003c/em\\u003e 12, 846775 (2022). (2023).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eIslam, W. et al. Improving Performance of Breast Lesion Classification Using a ResNet50 Model Optimized with a Novel Attention Mechanism. Tomography. 8, 2411\\u0026ndash;2425 Cakir Pekoz, B. et al. Can peritumoral edema evaluated by Magnetic Resonance Imaging before neoadjuvant chemotherapy predict complete pathological response in breast cancer? \\u003cem\\u003eScott Med J\\u003c/em\\u003e 68, 121\\u0026ndash;128 (2023). Park, J. et al. Machine Learning Predicts Pathologic Complete Response to Neoadjuvant Chemotherapy for ER\\u0026thinsp;+\\u0026thinsp;HER2- Breast Cancer: Integrating Tumoral and Peritumoral MRI Radiomic Features. \\u003cem\\u003eDiagnostics\\u003c/em\\u003e 13, 3031 (2023). Shigematsu, H. et al. Prognostic Value of MRI Assessment of Residual Peritumoral Edema in Breast Cancer Treated With Neoadjuvant Chemotherapy. \\u003cem\\u003eMagnetic Resonance Imaging\\u003c/em\\u003e 61, 944\\u0026ndash;955 (2025). (2022).\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSelvaraju, R. R. et al. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization. \\u003cem\\u003eInt. J. Comput. Vis.\\u003c/em\\u003e \\u003cb\\u003e128\\u003c/b\\u003e, 336\\u0026ndash;359 (2020).\\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\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"scientific-reports\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"scirep\",\"sideBox\":\"Learn more about [Scientific Reports](http://www.nature.com/srep/)\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Scientific Reports\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Scientific Reports\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true},\"keywords\":\"breast cancer, MRI, neoadjuvant chemotherapy, 2.5D deep learning, 2D deep learning\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-8229046/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-8229046/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eNeoadjuvant chemotherapy (NAC) is central to managing locally advanced breast cancer, and accurate response assessment is essential for guiding clinical decision-making. This study developed and compared 2D and 2.5D deep learning models using multimodal breast Magnetic resonance imaging (MRI) to predict NAC response. A retrospective cohort of 187 patients (mean age 50.24 years) treated at two different hospitals between January 2020\\u0026thinsp;\\u0026minus;\\u0026thinsp;December 2024 was analyzed. Inclusion required histopathologically confirmed invasive breast cancer, completion of standardized NAC, pre- and post-treatment MRI (dynamic contrast-enhanced (DCE), T2WI, diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC)), and postoperative pathological confirmation of response, including pathologic complete response (pCR). T1, T2, DWI, ADC, and multiphase DCE images were integrated into a seven-channel input, alongside a separate early-phase DCE input. Four convolutional neural networks (VGG16_bn, ResNet50, DenseNet121, and SqueezeNet1_0) were trained under 2D and 2.5D frameworks with transfer learning. Grad-CAM was applied for interpretability, and clinical features (speculation, edema changes) were incorporated into a combined predictive model. Performance evaluation included ROC, AUC(Area Under the Curve), sensitivity, calibration curves, and Hosmer\\u0026ndash;Lemeshow tests. In the training cohort, the fusion model achieved the highest performance (AUC 0.955). In the independent test cohort, 2.5D ResNet50 performed best (AUC 0.897, accuracy 0.895, sensitivity 0.944). Fusion further improved AUC to 0.955, demonstrating superior generalizability and clinical potential.\\u003c/p\\u003e\",\"manuscriptTitle\":\"2D and 2.5D Deep Learning Models for Neoadjuvant Chemotherapy Response Prediction in Breast Cancer\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2026-04-03 06:21:30\",\"doi\":\"10.21203/rs.3.rs-8229046/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2026-04-19T20:41:11+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"324047499508739900448776837329973832454\",\"date\":\"2026-03-29T15:36:52+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewersInvited\",\"content\":\"\",\"date\":\"2026-03-29T14:14:17+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvited\",\"content\":\"\",\"date\":\"2025-12-01T17:36:53+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorAssigned\",\"content\":\"\",\"date\":\"2025-11-29T05:11:54+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"checksComplete\",\"content\":\"\",\"date\":\"2025-11-29T05:11:41+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"submitted\",\"content\":\"Scientific Reports\",\"date\":\"2025-11-28T09:34:36+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"scientific-reports\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"scirep\",\"sideBox\":\"Learn more about [Scientific Reports](http://www.nature.com/srep/)\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Scientific Reports\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Scientific Reports\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"2a3fcbff-e3d2-42e6-855c-4bbe4df32504\",\"owner\":[],\"postedDate\":\"April 3rd, 2026\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"under-review\",\"subjectAreas\":[{\"id\":65481543,\"name\":\"Biological sciences/Cancer\"},{\"id\":65481544,\"name\":\"Biological sciences/Computational biology and bioinformatics\"},{\"id\":65481545,\"name\":\"Health sciences/Medical research\"},{\"id\":65481546,\"name\":\"Health sciences/Oncology\"}],\"tags\":[],\"updatedAt\":\"2026-04-03T06:21:31+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2026-04-03 06:21:30\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-8229046\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-8229046\",\"identity\":\"rs-8229046\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}