Understanding alzheimer's disease diagnostic factors through machine learning

Abstract

Alzheimer’s disease (AD) is one of the leading public health concerns that continues to grow as the world’s population rapidly ages. It is therefore crucial to understand factors behind AD diagnosis and patient classification, where one of the leading perspectives in research today is the A/T/N framework. This is unsuitable for diagnostic implementation due to its arbitrariness and exclusion of other non-biological AD pathologies. Thus, the present study aims to examine other factors contributing to AD risk, such as demographics, cognitive and neuroimaging to provide a holistic perspective on AD diagnosis and risk. Support vector machine (SVM) modeling was used to train 6 different models in total, resulting in 1 combined model, and separate models for demographic, cognitive assessments, A/T/N, biological factors and structural connectivity as predictors respectively. The models were trained and tested, following which they were evaluated on their performance metrics. The model which performed the poorest used structural connectivity data as the only predictor, while the cognitive assessments model was the best. The combined factors model did not perform as well as expected, due to integration and dimensionality issues. This paper supports the idea that biological factors alone are not sufficient to make a proper diagnosis of AD, and that cognitive assessment is a key part of AD evaluation and diagnosis. It also aims to provide the basis for the development of a tool that can be used to aid classification in research settings.