Mitigating dataset bias in Alzheimer’s Disease diagnostic prediction

Abstract

In recent years, artificial intelligence (AI) is being increasingly explored for enhancing the efficiency of diagnosis and prediction of Alzheimer’s disease (AD) through the analysis of neuroimaging data such as magnetic resonance imaging (MRI). However, many predictive models are developed using a limited number of datasets, raising concerns regarding dataset bias, representativeness, and the generalizability of model performance across diverse populations. This project examines the current landscape of AI applications in AD diagnosis through a narrative review. Simultaneously, a prototype machine learning model was developed using MRI image classification to predict AD, providing experimental results for a fine-tuned ResNet-34 architecture. Despite demonstrating high accuracy levels, the findings highlight the ethical challenges associated with deploying AI in real-world environments. To address these findings, we propose an ethical framework grounded in the principles of justice, non-maleficence, transparency, and accountability, aimed to promote equity and responsible development in AL-based diagnostic tools for AD.