Objective home-based EEG prediction of aMCI: Identification of a predictive electrophysiological model of cognitive function in amnesic mild cognitive impairment.

Project Information

ID-earlyMCI

Grant agreement ID: 893823

DOI

10.3030/893823

Project closed

EC signature date 24 March 2020

Start date 1 October 2020

End date 30 September 2022

Funded under

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Total cost

€ 184 590,72

EU contribution

€ 184 590,72

Coordinated by

BRAINWAVEBANK LTD.
United Kingdom

Periodic Reporting for period 1 - ID-earlyMCI (Objective home-based EEG prediction of aMCI: Identification of a predictive electrophysiological model of cognitive function in amnesic mild cognitive impairment.)

Reporting period: 2020-10-01 to 2022-09-30

Summary of the context and overall objectives of the project

With age, skills related to cognitive function are affected, resulting in decreased speed of processing, working memory capacity, inhibitory function and long-term memory. With dementia, these skills are further impacted. Atypical decreases in function do not necessarily manifest in all aspects of cognitive function simultaneously, but specific aspects can decline years prior to the diagnosis of dementias, such as Alzheimer’s disease (AD). Ideally, we would want to identify these declines before the disease fully manifests in all aspects, as this is key for early diagnosis of amnesic Mild Cognitive Impairment (aMCI), the preceding stage of AD. However, identifying these early declines require large studies covering decades of potential neurodegeneration. A complementary approach is to identify atypical age-related changes in cognitive function in healthy aging. This approach derived from healthy aging (with expected small changes) could potentially be applied to aMCI and AD populations (with expected larger changes) to identify inflection points of cognitive function worsening.

Electrophysiology (EEG or “brainwave recording”) is a well-established clinical biomarker with a strong literature linking it to screening, detection and tracking of MCI, AD, and other dementias, as well as age-related changes. However, most current studies on cognition in aging suffer from lack of personalised measures, or lack of data across days, months, years to identify changes over time. Therefore, in this project a new measurement of typical and atypical cognition in aging was developed, which capitalised on repeated assessments accompanied with EEG recordings, collected by Cumulus Neuroscience, Ltd.

Cumulus Neuroscience has developed a medical-device certified system of wearable EEG paired with engaging tablet-based tasks, and a supporting cloud infrastructure to securely collect and extract insights from the resulting data, yielding neurocognitive markers that are sensitive even to the subtle effects of healthy aging. Cumulus has designed gamified versions of gold-standard neuro-psychological tests that can be used in conjunction with wireless EEG. With its high participant-adherence, high system usability and high data quality, the Cumulus’ platform is suitable for identifying predictive models of cognitive (dys-) function on a large scale.

This project aimed to identify a personalised predictive model of cognitive function in older adults using a home-based EEG system and advanced machine learning methods. This required the development and identification of data selection approaches suitable for dry-EEG data to achieve data of good quality; the identification of models that model practice effects from repeated measurements; the identification of an EEG-base model of cognitive function in older adults; and the identification of a predictive EEG-based model of cognitive function in older adults three years later.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

In this project, a series of developments were made which included cutting-edge pipelines for data selection to ensure data is of sufficient quality, reliability and signal quality identification to compare different data selection approaches, feature/variable extraction to select variables that will be used in the analysis, modelling of practice effects to test if there are behavioural improvements, and machine learning analysis using home-based EEG data to identify participants with higher or lower cognitive function.

In two previous studies, participants had performed a series of neurocognitive assessments to assess their cognitive function, and had performed gamified tasks at home while wearing the dry-EEG headset. With this data, and after a comprehensive data quality check and feature extraction, a machine learning pipeline was developed and tested. The machine learning pipeline included approaches that were easy to interpret, avoiding black-box approaches. With this pipeline, a model that classified participants into higher or lower cognitive function with moderate accuracy was identified. This suggests that home-based EEG can be used for identifying subtle cognitive differences in neurotypical older adults cross-sectionally (with measurements taken in a 6-12 week window).

Participants were invited to a follow-up study, where only neurocognitive assessments were performed. A similar machine learning pipeline was tested which aimed to predict the level of cognitive function three years after the home-based EEG had been taken. In the follow-up study, the prediction models did not perform as well as the cross-sectional model. However, this could be due to the small number of participants who came back three years later (55 participants out of 94 in the initial group). Future research could aim for a larger sample size, as well as expanding the set of variables included in the models.

Besides the technical outcomes of this project, developments of this project were presented to different audiences in Europe, the USA, Colombia and Bangladesh. In these presentations the potential of interdisciplinary work including neuroscience and engineering was highlighted. The audiences included undergraduate, graduate and postgraduate students, as well as established scientific peers. Two of the presentations are available online in English and Spanish. During this fellowship, the fellow supported activities of the Career Development Working Group of the Marie Curie Alumni Association.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

The advanced methods developed in this fellowship have broadened and strengthened Cumulus analytic capabilities, which will be instrumental in Cumulus’ goal of understanding brain health and cognitive illness for improving people’s lives. During this fellowship, the fellow provided support to a PhD student doing a project in collaboration between the company and Trinity College Dublin, and supervised the master thesis of a student of the MSc Data Analytics program of Queen’s University Belfast. Statistical plans that the fellow developed during this fellowship will be used in future analysis.

The combination of unique large data and advanced methods of this fellowship has strengthened Cumulus’ position as a leading research-based company in brain markers of cognitive function. Also, the fellow received specialised training which will improve their research and transferrable skills. Hence, this fellowship contributed to achieving excellence in European research and enhancing the skill set of EU’s job market which is in shortage of big data skills. The methods developed in this fellowship will inform future predictive models of cognitive function which will be beneficial to the EU’s large ageing population, in line with the United Nations’ 3rd Sustainable Development Goal of ‘Good Health and Wellbeing’ and Horizon 2020’s societal challenge on Health, Demographic Change and Wellbeing.

Summary of methods and results of one of the outcomes of the project.