A Long Short-Term Memory Based Framework for Early Detection of Mild Cognitive Impairment From EEG Signals

Mild cognitive impairment (MCI) is an irreparable progressive neuro-degenerative disorder, which seems to be a precursor to Alzheimer's disease (AD) that may lead to dementia in elderly people. It is a major public health challenge for healthcare in the 21st century. Because there is no curative or therapy to halt or reverse the course of MCI, early identification is critical for successful treatment programs to enhance patients' quality of life. Currently, Electroencephalography (EEG) has been emerged as an efficient tool to investigate MCI. Traditional methods for finding MCI from EEG data use shallow machine learning-based architectures that cannot find important biomarkers in deep, hidden layers of the data and also have trouble dealing with a large amount of EEG data. To reduce this issue, this research will use EEG data to provide a deep learning-based framework using the Long Short-term Memory (LSTM) model for effective identification of MCI individuals from healthy volunteers (HV). The suggested framework consists of four phases: denoising, segmentation, downsampling, uncovering deep hidden features using the LSTM model, and identifying MCI patients with the sigmoid classifier. This study has designed 20 different LSTM models and investigated them to a publicly available MCI database to find out the best one. After performing 5-fold cross validation, the best model achieved 96.41% of accuracy, 96.55% of sensitivity and 95.95% of specificity. The proposed LSTM-based deep learning model provides a robust biomarker and guide technologists to create a new automatic diagnosis system for MCI detection.