Dry-electroencephalography and machine learning methods for differential profiling of mild cognitive impairment and dementia in Parkinson's disease -a feasibility study.

A Sweeney, B Devereux, S Kearney, J McKinley, C Ong, J Anderson, B Murphy, B McGuinness, P Passmore

Research output: Contribution to conferencePosterpeer-review


Introduction: Cognitive impairment is prevalent in Parkinson's disease (PD), with 50% of patients developing dementia (PDD) within 10 years (Williams-Gray et al., 2013). The presence of mild cognitive impairment (MCI), particularly visuospatial dysfunction, has been associated with increased PDD risk. EEG has previously been shown to have predictive value for patient outcomes in MCI associated with Alzheimer's disease (AD-MCI), yet there is little information on the electrophysiological correlates of PD-MCI. Dry-EEG offers an inexpensive, non-invasive and faster method to assess cognition in aging clinical groups. However, as dry-EEG signal suffers from higher levels of noise, the feasibility of using this method to track cognition in these patient groups must first be established. Methods: A 12-month follow-up study will be conducted. PD patients (n=60) and matched controls will complete a comprehensive neuropsychological assessment and a battery of EEG tasks. These tasks will assess resting state activity, attention, language, memory and visuospatial cognitive domains. The PD patient cohort will consist of PDNC (normal cognition) and PD-MCI patients. A cross-sectional study of Alzheimer's disease (AD), Lewy-body dementia (LBD) and PDD patients (n=30) at one time-point using the same assessment tasks will also be conducted. Analysis Approach: The primary outcomes for this study will be the ability of dry-EEG to differentially discriminate between these patient groups and to track changes in PD cognitive status over 12 months. Previous studies of cognitive dysfunction in PD have identified EEG slowing (for example decreased alpha/theta ratio) and delayed or absent ERP components (Seer et al., 2016). For each task, power spectral analysis will be computed and the relevant ERP components (such as the P300 and MMN from the auditory oddball task) will be analysed. Differences between groups will be assessed using randomisation tests with FDR-corrected pairwise comparisons. Correlation analysis with validated clinical measures such as the MoCA will also be performed. Machine learning algorithms such as the 'Elastic Net' (Zou & Hastie, 2005), random forests and k-nearest neighbour models will be used to identify EEG features which discriminate between patient groups and which are associated with cognitive outcomes (i.e. stability or worsening) at follow-up.
Original languageEnglish
Publication statusPublished - 23 May 2019
EventBNA Festival of Neuroscience 2019 -
Duration: 23 May 201923 May 2019


ConferenceBNA Festival of Neuroscience 2019


  • Alzheimer disease
  • E1A associated p300 protein
  • Parkinson disease
  • adult
  • aging
  • attention
  • cohort analysis
  • conference abstract
  • controlled study
  • correlation analysis
  • cross-sectional study
  • diffuse Lewy body disease
  • elastic tissue
  • electroencephalogram
  • electroencephalography
  • endogenous compound
  • evoked response
  • feasibility study
  • female
  • follow up
  • human
  • k nearest neighbor
  • language
  • major clinical study
  • male
  • memory
  • mild cognitive impairment
  • noise
  • protein fingerprinting
  • random forest
  • randomization
  • rest
  • spectroscopy


Dive into the research topics of 'Dry-electroencephalography and machine learning methods for differential profiling of mild cognitive impairment and dementia in Parkinson's disease -a feasibility study.'. Together they form a unique fingerprint.

Cite this