Computational pathology infer clinically relevant protein levels and drug efficacy in breast cancer by weakly supervised contrastive learning

Abstract

Visual inspection of histopathology slides via optical microscope is the routine medical examination for clinical diagnosis of tumors. Recent studies have demonstrated that computational pathology not only automate the tumor diagnosis, but also showed great potential to uncover tumor-related genetic alterations and transcriptomic patterns. In this paper, we propose wsi2rppa, a weakly supervised contrastive learning framework to infer the protein levels of tumor biomarkers from whole slide images (WSIs) in breast cancer. We firstly conducted contrastive learning-based pre-training on tessellated tiles to extract histopathological features, which are then aggregated by attention pooling and adapted to downstream tasks. Our extensive experiments showed that our method achieved state-of-the-art performance in tumor diagnostic task, and also performed well in predicting clinically relevant protein levels and drug response. To show the model interpretability, we spatially visualized the WSIs colored the tiles by their attention scores, and found that the regions with high scores were highly consistent with the tumor and necrotic regions annotated by a 10-year experienced pathologist. Moreover, spatial transcriptomic data further verified that the heatmap generated by attention scores agree greatly with the spatial expression landscape of two typical tumor biomarker genes. In particular, our method achieved 0.79 AUC value in predicting the response of breast cancer patients to the drug trastuzumab treatment. These findings showed the remarkable potential of deep learning-based morphological feature is very indicative of clinically relevant protein levels, drug response and clinical outcomes.