The identification of mild cognitive impairment in Parkinson's disease using EEG and machine learning. (7th December 2020)

• Record Type: Journal Article
• Authors: Sweeney, Aoife; Devereux, Barry; Ong, Charlie; McKinley, John; Kearney, Seamus; Doherty, Karen; Foy, Julia; Murphy, Brian; McGuinness, Bernadette; Passmore, Anthony Peter
• Is Part Of: Alzheimer's & dementia. Volume 16(2020)Supplement 11
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-07
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.040432
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Electroencephalography (EEG) is an inexpensive, non‐invasive and faster method to assess cognition in aging clinical groups. In this study, we are investigating the feasibility of using a 'dry‐EEG' mobile headset to assess cognitive impairment in Parkinson's disease (PD) over a 12‐month period. Cognitive impairment is prevalent in PD with approximately 75% of patients developing dementia within 10 years (Aarsland and Kurz, 2010). The presence of mild cognitive impairment (PD‐MCI) has been associated with increased dementia risk (Pederson et al. 2013). Very little is known about the EEG correlates of PD‐MCI, however EEG has previously been used to predict outcomes in MCI associated with Alzheimer's disease. Method: 50 Parkinson's disease patients and 50 age and sex‐matched controls have completed a battery of baseline neuropsychological and EEG tasks. These tasks measured attention, language, memory, visuospatial and executive function domains. Resting state EEG was also recorded. Event‐related potential components and spectral analyses on the delta, theta and alpha oscillatory bands were computed. Differences between groups were assessed using randomisation tests with FDR‐corrected pairwise comparisons. Random forest, k‐nearest neighbours, support‐vector machine and logistic regression machine learning algorithms were also applied to these data. An identical 12‐month follow‐up assessment will be conducted. Result: Findings from the baseline assessments are presented. 52% of the Parkinson's disease cohort fulfilled the criteria for MCI. There was no difference in age, education or pre‐morbid IQ between groups. Machine learning classifiers derived from EEG and neuropsychological test metrics discriminated between PD‐MCI, PD normal cognition (PD‐NC) and healthy controls with over 80% accuracy. Conclusion: Cognitive impairment is prevalent in PD. Our findings show that it is feasible to use dry‐EEG to profile cognition in PD and can differentially discriminate between cognitively normal and cognitively impaired PD patients at an early stage. The features selected are in agreement with the dual syndrome hypothesis of cognition (Kehagia et al., 2013) and those identified in a 10 year follow‐up study (Williams‐Grey et al., 2013). The stability of these features and their ability to predict cognitive status at 12 months will be investigated at a follow‐up assessment.