Classification of Preclinical-Alzheimer's risk group from EEG data and psychological testing data using machine learning.
Abastract
This study aimed to classify older adults with risks of preclinical Alzheimer’s disease from normal older individuals by using the Free and Cued Selective Reminding Test (FCSRT) as a cognitive marker to examined anterograde episodic memory which is an initial cognitive sign in early stage of AD. However, individuals in the preclinical stage might not initially show noticeable cognitive deficits as the pathology progresses in their brain. Therefore, the concept of neural compensation was examined in this study. By adding the Wisconsin Card Sorting Test (WCST) to increase brain activities, we hypothesized that it would activate neural compensation and increased the brain activity level. So, EEG was used as a biomarker to record the brain activities. Then the data from FCSRT and EEG were used to train two types of machine learning (SVM and Decision tree) with 3 data sets: FCSRT, EEG, and both. The result showed that the decision tree model outperformed the other model in classifying results. The decirion tree model trained by FCSRT alone predicted that every participant was at risks while trained by EEG alone results in all participants were normal. Only when combining both FCSRT and EEG data, the decision tree model was able to classify participants at risk from normal participants.