A Systematic Fairness Evaluation of Racial Bias in Alzheimer's Disease Diagnosis Using Machine Learning Models

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Informatics

Background:

• Alzheimer's disease (AD) poses a significant global health challenge, with projections indicating a substantial increase in cases.
• Machine learning (ML) models are increasingly used for AD diagnosis and progression prediction.
• Existing ML models may exhibit biases due to a lack of racial diversity in training datasets, impacting their generalizability.

Purpose of the Study:

• To investigate the fairness of ML models in AD diagnosis across different racial groups.
• To test the hypothesis that ML models trained on a single racial group perform poorly on other groups.
• To evaluate the effectiveness of fairness techniques in mitigating bias in ML models for AD.

Main Methods:

• Employed feature selection and model training techniques to enhance fairness.
• Investigated ML model performance across diverse racial demographics.
• Applied fairness interventions to reduce algorithmic bias.

Main Results:

• Confirmed that ML models trained on a single racial group underperform on other racial groups.
• Demonstrated that implementing fairness techniques significantly reduces bias in ML models for AD.
• Highlighted performance disparities linked to dataset demographics.

Conclusions:

• Emphasized the critical need for racial diversity in datasets used for training AD prediction models.
• Stressed the importance of developing and deploying fair ML models to ensure equitable healthcare outcomes.
• Advocated for continued research into bias mitigation strategies in medical AI.