Gleason Grading Prediction in Prostate Cancer Patients – Xinyu Hao

Accurate prediction of the Gleason Grade is crucial for the diagnosis and treatment planning of prostate cancer. Although recent multiple instance learning methods have made progress in this field, many of them have failed to exploit the relationships between features across different regions of whole slide images (WSIs). In this talk, I will present a novel MIL framework designed to explicitly incorporate domain-specific information. Our method employs a selective instance feature aggregation mechanism to enhance patch-level representations. To further align the model with expert knowledge at the patient level, we adopt a knowledge distillation strategy to improve prediction accuracy. This work highlights the importance of integrating pathology-specific insights into MIL frameworks for enhanced clinical decision support.