Deep learning supported discovery of biomarkers for clinical prognosis of liver cancer

Abstract Tissue biomarkers are crucial for cancer diagnosis, prognosis assessment, and treatment planning. However, there are few known biomarkers that are robust enough to show true analytical and clinical value. Deep learning (DL)-based computational pathology can be used as a strategy to predict survival, but the limited interpretability and generalizability prevent acceptance in clinical practice. Here we present an interpretable human-centric DL-guided framework called PathFinder (Pathological-biomarker-finder) that can help pathologists to discover new tissue biomarkers from well-performing DL models. By combining sparse multi-class tissue spatial distribution information of whole slide images (WSIs) with attribution methods, PathFinder can achieve localization, characterization, and verification of potential biomarkers, while guaranteeing state-of-the-art prognostic performance. Using PathFinder, we discovered that spatial distribution of necrosis in liver cancer, a long-neglected factor, has a strong relationship with patient prognosis. We therefore proposed two clinically independent indicators, including necrosis area fraction and tumor necrosis distribution, for practical prognosis, and verified their potentials in clinical prognosis according to Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK)-derived criteria. Our work demonstrates a successful example of introducing DL into clinical practice in a knowledge discovery way, and the approach may be adopted in identifying biomarkers in various cancer types and modalities.