<ul><li>A study conducted a comprehensive fairness analysis of machine learning models for diagnosing Mild Cognitive Impairment (MCI) and Alzheimer's disease (AD) from MRI data, focusing on biases related to age, race, and gender.</li><li>The study assessed various fairness definitions and metrics to identify biases and compared bias mitigation strategies like pre-processing, in-processing, and post-processing methods.</li><li>A new composite measure was introduced to balance fairness and performance trade-off, considering the F1-score and equalized odds ratio, suitable for medical diagnostic applications.</li><li>Results showed biases in age and race, with mitigation strategies like Reject Option Classification improving equalized odds by 46% and 57% for race and gender biases, respectively, and adversarial debiasing achieving a 40% improvement for age bias.</li></ul>

Comparative assessment of fairness definitions and bias mitigation strategies in machine learning-based diagnosis of Alzheimer's disease from MR images

Comparative assessment of fairness definitions and bias mitigation strategies in machine learning-based diagnosis of Alzheimer's disease from MR images

Discover more