Deep learning-based predictive model for pathological complete response to neoadjuvant chemotherapy in breast cancer from biopsy pathological images: a multicenter study

• DOI: 10.3389/fphys.2024.1279982
• Publication Date: 2024-01-31T04:25:08Z
• Authors: Huancheng Zeng; Siqi Qiu; Shuxin Zhuang; Xiaolong Wei; Jundong Wu; Ranze Zhang; Kai Chen; Zhiyong Wu; Zhemin Zhuang
• Keywords: Neoadjuvant Therapy

Introduction: Early predictive pathological complete response (pCR) is beneficial for optimizing neoadjuvant chemotherapy (NAC) strategies breast cancer. The hematoxylin and eosin (HE)-stained slices of biopsy tissues contain a large amount information on tumor epithelial cells stromal. fusion image features clinicopathological expected to build model predict pCR NAC in Methods: We retrospectively collected total 440 cancer patients from three hospitals who underwent NAC. HE-stained were scanned form whole-slide images (WSIs), representative regions interest (ROI) each WSI selected at different magnifications. Based several deep learning models, we propose novel feature extraction method with Further, fused features, multimodal prediction based support vector machine (SVM) classifier was developed validated two additional validation cohorts (VCs). Results: Through experimental found that the SVM classifier, which uses VGG16 ×20 magnification, has best efficacy. area under curve (AUC) (DPM) 0.79, 0.73, 0.71 TC, VC1, VC2, respectively, all exceeded 0.70. AUCs clinical (CM), established by using 0.79 0.73 respectively. (DPCM) fusing improved AUC TC 0.84. VC2 0.78. Conclusion: Our study reveals pre-NAC can be used model. Combining further enhance efficacy