Classification of Apolipoprotein E ϵ4 Allele Carriers in Cognitively Healthy Older Adults and Alzheimer's Disease Patients Using Event-Related Potentials and Event-Related Spectral Perturbation

Alzheimer's disease (AD) is the most common neurodegenerative disease in older adults. With increasing life expectancy, the incidence of this disease is expected to increase over time. Individuals who are carriers of the ε4 allele of Apolipoprotein E (APOE ε4) have a greater chance of developing AD. Electroencephalography (EEG) is a relatively inexpensive and non-invasive method that can help diagnose neurodegenerative diseases. Event-Related Potentials (ERP) analysis is an EEG tool widely used in the literature to help identify AD. In this study, the Event-Related Spectral Perturbation (ERSP) tool was added, a novel application for identifying carriers of the APOE ε4 allele. We subdivided the individuals into four groups: control group of APOE ε4 carriers (CT-C), control group of non-APOE ε4 carriers (CT-N), AD carrying APOE ε4 (AD-C), and AD without APOE ε4 (AD-N). The extraction of ERP and ERSP features from EEG can be used as classifier features using Machine Learning (ML) techniques. The EEGLAB tool was used for the data preparation and feature extraction pipeline to search for statistically significant differences between the groups. Python programming language was also used with the aid of the scikit-learn library to perform feature selection and classification. Among the classifiers we used, we identified greater accuracy and technical advantage in the use of Support Vector Machine (SVM). We also found that ERSP was more effective than ERP in terms of the ML accuracy. Interestingly, this study also observed statistically significant differences in the pre-stimulus interval (-500 to 0 ms) for ERSP, and not only in the post-stimulus interval that is generally used for cognitive studies. This study proposed a machine learning classification method to identify APOE ε4 allele carriers in older adults with AD and cognitively healthy older adults. Using ERP alone, the performance was poor, but with the use of ERSP, the performance was much better, and with the fusion of ERP and ERSP, the accuracy values were even higher. Besides the classification accuracy, we added the F1 score as a performance metric.