Researchers have introduced a new metric to evaluate the fairness of medical imaging models in diabetic retinopathy screening. The Explanation Consistency Score (ECS) measures how consistent attribution maps are across different demographic groups. In experiments on diabetic retinopathy images, ECS was found to be relatively high and not significantly associated with performance disparities between ethnic groups. This suggests that predictive fairness and explanation consistency capture distinct aspects of model behavior. --- Why it matters: This work matters because it highlights the need for a more nuanced understanding of fairness in medical imaging models. By distinguishing between predictive fairness and explanation consistency, researchers can develop more effective strategies to address biases and improve model performance across different demographic groups. Source: https://arxiv.org/abs/2608.18759