People-Centred Medical Image Analysis Proposes Fairer AI Integration

A new arXiv preprint argues that data-centric medical AI has failed to achieve widespread clinical adoption due to insufficient focus on fair performance across diverse patient populations and workflow integration. The authors identify performance biases as creating regulatory barriers and poorly integrated automation as disrupting clinical routines, degrading human-AI collaboration, and reducing clinician willingness to adopt AI tools. They note that prior work on workflow integration (e.g., Learning to Defer and Learning to Complement) and AI fairness has examined these challenges in isolation, overlooking their interdependence and practical constraints like restricted clinician availability. The paper proposes a people-centred approach to medical image analysis that jointly optimizes fairness and workflow integration.