A Deep Learning Framework for Predicting Platinum Drug Sensitivity in Breast Cancer from Histopathological Images: Towards Precision Oncology

Abstract Breast cancer (BC) poses a significant global health challenge due to its high mortality and incidence rates. Although the use of platinum-based drugs in adjuvant chemotherapy has somewhat improved the complete response rate (pCR), the five-year survival rate for metastatic BC remains below 30%. Heterogeneous tumor microenvironments and patient variability lead to divergent drug sensitivities, underscoring the need for predictive tools to optimize treatment. Current organoid-based sensitivity assays face limitations in stability and timeliness, undermining clinical utility. This study analyzed tumor tissues from 63 breast cancer patients through dual approaches: organoid culture for platinum-drug sensitivity grading and histopathological imaging for feature extraction. Texture and nuclear characteristics from primary tumors were quantified using gray-level co-occurrence matrix (GLCM) and OpenCV algorithms. These features were integrated with deep learning models to establish a predictive framework for platinum-drug responsiveness directly from histopathological images. The developed system bypasses time-consuming organoid culture, enabling rapid sensitivity prediction to guide personalized chemotherapy. By correlating morphological patterns with drug response outcomes, this method enhances clinical decision-making for precision oncology, potentially improving survival through early treatment stratification. The approach addresses critical gaps in traditional methods by combining computational pathology with artificial intelligence, offering a scalable solution for real-world BC management.