Machine Learning and Deep Learning Approaches for Stage Classification of Alzheimer’s Disease using Near-Infrared Spectroscopy

Abstract

The prompt diagnosis of Alzheimer’s disease (AD) and its prodromal stages is vitally important in enabling patients, who may exhibit different patterns of neurodegenerative severity and progression risks, to take intervention and timely symptomatic treatments before brain damage occurs. Functional near-infrared spectroscopy (fNIRS) has been widely employed to support early-stage AD diagnoses in clinical settings. This study aims to validate the potential of a collection of Machine Learning (ML) classifiers coupled with diverse Re-sampling techniques and Deep Learning (DL) models. First, the hemodynamic responses were measured in the prefrontal cortex from four subject groups with a comprehensive experimental design for evaluating AD progression. Then, we adopted a set of ML classifiers combined with various sampling strategies and DL approaches to perform the multi-class classification problem on an extremely imbalanced dataset. We examined the changes in the concentration of oxygenated (HbO), deoxygenated (Hb), and total (THb) hemoglobin during the resting, cognitive, memory, and verbal tasks of the experimental protocol. Results indicated that several different ML classifiers coupled with sampling techniques are potentially capable of significantly enhancing classification performance. Of the 8 ML classifiers and 12 sets of resampled features, the highest accuracy was achieved with neural networks using the resampled SMOTETomek features (0.883 ± 0.010). Deep learning solution, meanwhile, remarkably yields the robust outperformance compared to conventional ML algorithms, CNN-LSTM in particular (0.909 ± 0.012). These findings validate the feasibility of implementing sampling strategies alongside ML algorithms and DL techniques in the imbalanced class distribution analysis, and demonstrate the great potential of fNIRS-based approaches in AD studies with can be further contributed to the development of AD diagnosis systems.