Screening of Mild Cognitive Impairment in Patients with Parkinson's Disease Using a Variational Mode Decomposition Based Deep-Learning

Parkinson's disease (PD) which is the second most common neurodegenerative disease in the United States is challenging for specialists to diagnose and grade. Prior to the onset of motor symptoms of PD, patients exhibit alteration in sleep architecture which plays a critical role in consolidating memory, a key cognitive process of the brain. Standard spectral and signal analysis techniques have been recently introduced to exploit the changes in the electroencephalography of sleep related to PD or its cognitive complications including dementia. However, the use of artificial intelligence for the automated detection of the progression of PD to mild cognitive impairment (MCI) or dementia in sleep EEG have not yet been investigated. In this paper, we introduce a novel highly accurate variational mode decomposition based deep-learning framework applied on sleep electroencephalography signals in order to classify PD subjects into patients exhibiting normal cognition (NC) or MCI. The proposed framework is capable of detecting MCI at a significantly high 4-fold cross validation accuracy, sensitivity, specificity and quadratic weighted Kappa score of almost 99% offering a rapid and supportive tool for specialists to monitor the progression of PD and ensure the early initiation of efficient therapeutic treatments that will accordingly improve the quality of life for patients and their caregivers.