A deep learning based framework for diagnosis of mild cognitive impairment

Abstract

Detecting mild cognitive impairment (MCI) from electroencephalography (EEG) data is a challenging problem as existing methods rely on machine learning based shallow architectures that are unable to successfully uncover relevant biomarkers from deep hidden layers of data. This study will design a deep learning-based framework including a Gated Recurrent Unit (GRU) model for effective detection of MCI participants from healthy control (HC) subjects utilizing EEG data. MCI is a gradual, irreversible neurodegenerative illness that is frequently a precursor to Alzheimer’s disease (AD) and can result in dementia in elderly. There is no cure or treatment to stop or reverse the course of MCI, early identification is critical for successful application of treatment strategies to enhance the quality of life of patients. The proposed framework consists of four steps: gathering raw EEG data, preprocessing of raw data (de-noising, segmentation, and down sampling), uncovering hidden features and classification of MCI subjects from normal subjects, and performance evaluation of the proposed model. The proposed GRU model has been compared with long short-term memory (LSTM), support vector machine (SVM), and K-nearest neighbor (KNN) classifiers. The stability of the framework is evaluated through 5-fold cross validation. The suggested GRU-based deep learning model serves as a reliable biomarker and aids technicians in developing a new automatic MCI detection method.