Machine Learning Approach for Dementia Classification using EEG Signal Analysis

Alzheimer's disease belongs to the most quintessential types of dementia and it counts a huge percentage of dementia cases as more than 60 percent. Diagnosing dementia definitively and early is a huge challenge. This study's objective is to identify Alzheimer's Disease during its earliest stages like subjective cognitive decline and Mild cognitive impairment using EEG signal analysis.Methods: Data is provided by CRETH, includes the sample EEG signal recordings of 48 AD-patients, 79 MCI-patients, 34 SCD-patients, and 33 HC's. The EEG signals are analyzed by extracting features using FFT, CWT and PSD techniques and Machine learning algorithms like, k-Nearest Neighbor, Neural networks, Support vector machine and Random Forest , are used as classifiers.Results: Receiver-operating characteristic analysis of the EEG results 90%for HC ,81% for SCD, 90% for MCI and 89% for AD when CWT feature extraction for Beta band using KNN classifier is used corresponding to a sensitivity of 95%, specificity of 84%, and F1-score of 84%. Using PSD feature extraction of beta band with KNN classifier yielded 94 % of specificity, but sensitivity of 80% and F1-score of 81% is observed.