Link to bioRxiv paper:
http://biorxiv.org/cgi/content/short/2023.04.13.536819v1?rss=1
Authors: Xie, X., Liu, H.
Abstract:
Visual inspection of histopathology slides via optical microscope is the routine medical examination for clinical diagnosis of tumors. In recent years, deep learning has demonstrated that computational pathology has potential to infer tumor-related genetic alterations and transcriptomic patterns. In this paper, we propose a weakly supervised contrastive learning framework to infer the protein level of tumor biomarkers from whole-slide images (WSIs) of breast cancer. We conduct contrastive learning-based pre-training on large-scale unlabeled breast cancer WSIs to extract histopathological features, which are then adapted to downstream tasks, including tumor diagnosis, prediction of protein level and drug response. Our extensive experiments show that our method outperforms other state-of-the-art models in tumor diagnostic task, and achieves high performance in predicting clinically relevant protein levels. To show the model interpretability, we spatially visualize the WSIs colored by attention scores of tiles, and find that the regions with high scores are highly consistent with the tumor and necrotic regions annotated by an experienced pathologist. Moreover, spatial transcriptomic data further verified that the heatmap generated by attention scores agree greatly with the spatial expression map of tumor biomarker genes. Our method achieves 0.79 AUC value in predicting the response of breast cancer patients to the drug trastuzumab. These findings show the remarkable potential of deep learning-based morphological feature is very indicative of clinically relevant protein levels, drug response and clinical outcomes.
Copy rights belong to original authors. Visit the link for more info
Podcast created by Paper Player, LLC
