Single channel EEG time-frequency features to detect Mild Cognitive Impairment

Detection of Mild Cognitive Impairment (MCI), a possible biomarker for the Alzheimer disease (AD), has a huge importance for the management of the elderly's healthcare. As the treatment of AD is very costly and the early discovery is beneficial, a low-cost, early detection mechanism is needed. In this study, Electroencephalography (EEG) data from seventeen subjects was used to determine if brain activity could be used to distinguish MCI with cognitively normal individuals and compare classification performance. Event-related brain potentials (ERP) were recorded in response to auditory speech stimuli. We extracted spectral and temporal features of the ERPs and built a MCI detection process with Support vector machine (SVM), Logistic Regression (LR), and Random Forest (RF). We have compared behavioral response against EEG based time domain features, frequency domain features, and top-ranked features of time-frequency domains. Four feature groups of our study demonstrate that the ranked time and frequency domain features of the EEG perform better than behavioral features and other EEG/ERP response metrics. Our results demonstrate the performance of the detection of MCI with a cross-validation accuracy of 87.9%, sensitivity 85%, specificity 90%, and F-score 94%. Ability to objectively and reliably detect MCI at early might lead to efficacious treatment of AD and related disorders.