A Combination of Functional Principal Component Analysis and Long Short-Term Memory Networks for Alzheimer's Disease Prediction

Alzheimer's disease (AD) research has focused on understanding its causes, progression, and treatments, with an emphasis on early detection, particularly diagnosing mild cognitive impairment. Research AD data are typically longi-tudinal, sparse, and irregularly spaced, with high-dimensional features due to advances in neuroimaging techniques. Traditional classification methods have limitations on handling such data. This study proposes to combine multivariate functional principal component analysis (MFPCA) with long short-term memory (LSTM) networks to reconstruct variable functions over time and classify data. The empirical results show that this approach outperformed other methods, including the flexible discriminant analysis and the regularized discriminant analysis. These findings suggest that the proposed method holds great potential for the early detection of AD.