Deep learning-based large-scale pan-cancer multiple biomarkers prediction using RIDGE with pathological images

Abstract

Molecular profiling tests currently relies significantly on genomics or transcriptomics analysis, leading a prolonged turnaround time and higher costs procedure. This comprehensive pan-cancer study aims to develop a general, weakly supervised, and clinically interpretable deep learning (DL) approach for predicting large-scale biomarkers directly from hematoxylin and eosin (H&E)-stained whole slide images (WSIs). We introduce Rapid and Intelligent Detector for Genetic Estimation (RIDGE) , a deep learning-based artificial intelligence (AI) system for gene expression signatures and standard biomarkers automated prediction in annotation-free whole slide images. RIDGE was developed and validated using 4983 WSIs from 4680 patients across 12 solid tumors from The Cancer Genome Atlas (TCGA). Its feasibility was further confirmed in an external validation cohort of 221 WSIs from 105 colorectal cancer patients from The Clinical Proteomic Tumor Analysis Consortium (CPTAC), specifically for predicting microsatellite instability (MSI) status. RIDGE achieved an overall area under the curve (AUC) of 0.763 (95% CI, 0.724–0.802) across 12 cancer types. In the external CPTAC cohort, RIDGE demonstrated an AUC of 0.769 (95% CI 0.686–0.841) for MSI status prediction, underscoring its reproducibility. We observed that the morphological visual characteristics could be captured by our model, enabling the gene expression signatures and standard biomarkers highly detectable directly from WSIs. These findings highlight RIDGE’s potential to significantly expedite cancer screening and personalized therapy. Moreover, RIDGE offers an AI-assisted assessment approach with high predictive performance to elucidate and quantify genotype-phenotype relationships in cancer, demonstrating its promise in advancing the field of cancer diagnostics. Not applicable.