Machine Learning for Alzheimer's Disease

"Machine Learning for Alzheimer's Disease: A Systematic Review of Diagnostic Performance and Ethical Challenges Using the ADNI Dataset"

Presented by Earlel Thiyagaratnam, PhD candidate in Health Information Science, as part of the 2024/25 Mediations Lecture Series.

All are welcome.

Attend in-person: FNB 4110
Attend online: Zoom link

Abstract: This study explores the use of machine learning (ML) for diagnosing Alzheimer’s Disease (AD), focusing on its accuracy and the challenges to its adoption. AD, the leading cause of dementia, lacks reliable early diagnostic tools, which delays timely intervention. ML offers potential by analyzing biomarkers from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset, which includes neuroimaging, genetic, and cognitive data. A review of 10 peer-reviewed studies (2020–2023) finds that ML models, particularly convolutional neural networks (CNNs), achieve diagnostic accuracy between 70% and 99% when using ADNI data. However, several concerns about data validity arise. The ADNI dataset’s internal validity is affected by biases in its longitudinal design and participant selection. Its external validity is limited by the underrepresentation of ethnic minorities and non-generalizable clinical protocols. Ethical issues, such as biases in data interpretation, lack of transparency in algorithms, and gaps in understanding patient experiences, further hinder clinical implementation. Regulatory frameworks like Canada’s Bill C-27 promote transparency and equity but lack mechanisms to enforce bias reduction in AI systems. Ultimately, while ML shows promise for diagnosing AD, its widespread adoption requires addressing these data biases, integrating diverse biomarkers, and ensuring policies that promote fair and transparent AI use in healthcare.