Revolutionizing Cancer Diagnosis: How Deep Learning Predicts Continuous Biomarkers with Unprecedented Accuracy

The study introduces a self-supervised attention-based method for weakly supervised regression in digital pathology, which significantly improves biomarker prediction accuracy. Regression-based deep learning outperforms classification-based deep learning in predicting continuous biomarkers. The CAMIL regression method is used to predict molecular biomarkers from pathology slides, showing superior performance in predicting HRD status. Regression models have potential in enhancing prognostic capabilities and refining predictions from histologic slides.