Mild cognitive impairment classification based on a deep learning-based approach using EEG data

With the increase in the life expectancy of a person in our society, more and more people are developing neurological diseases such as Alzheimer's. Advancements in biomedical data and computer capabilities allow for new approaches to detect and prevent neurocognitive problems. Mild cognitive impairment, which may be a precursor to Alzheimer's disease, is amenable to early identification by using computer-aided electroencephalography (EEG) analysis. EEG signals provide a consistent and affordable technique to investigate the various patterns between amnestic mild cognitive impairment (aMCI) and non-aMCI (naMCI). We collected brain wave signals using an EEG system (Brain Products GmbH, Munich, Germany) during a 2-level N-Back working memory test, where 34 active electrodes were employed. During the preprocessing step, the EEG signals will be cut into smaller frames and converted to the frequency domain using the short-time Fourier transform (STFT) transform to extract brainwave sub-bands (theta, alpha, and beta). Then, the extracted brainwave sub-bands will be converted to 2D images, where they will be used as input to train a convolutional neural network (CNN) model for patient classification. Experiment findings showed that the proposed method for subject classification provided good accuracy for the healthy, aMCI, and naMCI classes. The proposed method has yielded a high classification accuracy of 98.2%, 96.7% and 97.1% for HNC, aMCI, and naMCI, respectively.