Exploring the role of dry-EEG and machine learning methods to profile mild cognitive impairment

  • Alison Toogood

Student thesis: Doctoral ThesisDoctor of Philosophy


Alzheimer’s Research UK reports 850,000 individuals are living with dementia in the UK today and this is estimated to increase to one million by 2025. Alzheimer’s Disease (AD), a type of dementia, accounts for approximately 60% of all diagnoses and as yet, there is no cure. Diagnosis rates in high income countries are below half and achieving acceptable levels of coverage and access to care is the challenge currently being faced (World Health Organization, 2015).

Mild cognitive impairment (MCI) is a pre-clinical form of AD (Petersen, 2011) and is manifested by mild symptoms affecting cognitive function, without interference in daily activities. Neurodegeneration, defined by synaptic dysfunction, occurs decades prior to any symptoms emerging (Albert et al., 2011) interrupting electrical communication between neurones.

A multi assessment approach using a wireless, dry-EEG system was used to explore the potential of recording brain activity outside the laboratory and in a more informal setting. Participants with MCI and neurotypical controls completed a neuropsychological assessment and computerized tasks designed to probe cognitive function where deficits are expected in MCI. Event related potential technique was applied to the EEG and features of interest were extracted for further analysis using supervised machine learning methods.

Analysis and Results:
Analysis was carried out on a group level and individual level. The latter involved training algorithms on a range of data models to identify patterns in the data with feature combinations predicting group membership of individuals with accuracy. Randomisation tests confirmed that there no differences found at a group level when considering the groups’ performance during episodic memory, language, executive function (working memory) or attention tasks.

For analysis at an individual level, improved model accuracies with the addition of ERP features to neuropsychological features were reported for the language and attention task. A combination of features, namely an ERP feature and neuropsychological feature were selected by the algorithms as features contributing to the best model.

The four EEG tasks in this study were designed to probe cognitive function known to deteriorate in individuals with MCI. Considering the trajectory of cognitive decline is essential when interpreting differences, or indeed no differences between the groups. However, a difference between the groups with respect to episodic memory would have been expected since it is the cognitive function upon which their diagnosis is based: this finding most likely points to the low power of the study. Feature selection was of interest across the tasks and provided insight into potential predictors of progression to AD relative to the cognitive function. However, due to the heterogeneity of MCI, it is with caution that these features are discussed since it is unknown who within the MCI group would progress to AD.

The study was a proof of concept exploring cognitive function using a wireless dry-EEG headset outside a laboratory or hospital setting. The benefits of this system included a quick and easy set-up without compromising the quality of data. This approach could provide an efficient and affordable way to extend research into more accessible locations such as health centres and at home.
Date of AwardJul 2021
Original languageEnglish
Awarding Institution
  • Queen's University Belfast
SponsorsNorthern Ireland Department for the Economy
SupervisorBernadette McGuinness (Supervisor) & Barry Devereux (Supervisor)


  • Electroencephalography
  • event related potential
  • mild cognitive impairment
  • N400
  • P300
  • subsequnet memory effect
  • ERP
  • MCI
  • EEG
  • episodic memory
  • language
  • working memory

Cite this