When Views Disagree: Conflict-Aware Evidential Inference for Multi-View Mammography

Multi-view mammography uses multiple imaging perspectives to support breast cancer detection, but variations in image quality and anatomical coverage can lead to conflicting diagnostic evidence. In this talk, PhD candidate Rayhan Ahmed will present the Conflict–Evidential Inference Network (CEI-Net), a novel framework that treats each mammography view as a source of uncertain evidence and explicitly reasons about disagreement between views. By incorporating calibrated uncertainty into the decision-making process, CEI-Net improves both the reliability and interpretability of breast cancer assessment. The approach has demonstrated strong performance on leading mammography benchmarks and offers new insights into building trustworthy AI systems for clinical screening.

About the Speaker

Md Rayhan Ahmed is a PhD candidate in Computer Science at UBC whose research focuses on trustworthy medical AI, including uncertainty-aware and conflict-aware deep learning for medical image analysis. His interests include medical image segmentation, anomaly detection, model calibration, interpretability and explainable clinical AI.