Computational methods for identifying biomarkers of cognitive decline in Parkinson's disease using Quantitative EEG

Neurological disorders are the leading cause of disability and the second leading cause of death worldwide. Among these diseases, neurodegenerative disorders constitute a considerable segment, the two most common ones being Alzheimer’s (AD) and Parkinson’s disease (PD), which are also the leading causes of dementia. Dementia is a severe burden on the quality of life of the patients and their caregivers. As there is still no cure for the cause of neurodegenerative diseases, dementia due to AD and PD can only be treated symptomatically to some extent. For any potential clinical studies aiming at finding curative agents for neurodegenerative dementias, it is critical to intervene at a very early stage of the disease, when a substantial part of neurons are still alive and can be preserved. Moreover, symptomatic interventions by lifestyle change or drug application are likely more effective when applied early in the disease course. For these reasons, new approaches to assess the risk of dementia in cognitively healthy or only mildly affected patients is essential.
This thesis focusses on finding potential diagnostic and prognostic biomarkers of cognitive decline in Parkinson’s disease patients, extensively investigating patterns in brain waves through electroencephalogram (EEG) recordings. In a real-world clinical study at one centre, it is common to have data from a limited number of patients. However, due to the nature of EEG data, the number of features extracted can be much higher than the number of patients, resulting in sparse data requiring suitable analysis methods. The different studies carried out for this project aimed at feature selection – both, for distinguishing PD patients from healthy controls as well as PD patients with Mild Cognitive Impairment from those without, and investigating the association of EEG features with cognitive tests. Through a collaboration, it was also possible to evaluate EEG as a potential marker for patients at the prodromal stage or before there are enough symptoms to have a clear clinical diagnosis of Parkinson’s disease.
The first significant outcome of this thesis was identifying optimal methods for feature selection and using penalized regression for shortlisting EEG spectral power features to classify PD patients from healthy individuals using high-density as well as standard 10-20 EEG recording systems. This was also tested on some prodromal PD patients. Theta spectral power came out as an important feature in both studies and was highly associated with dopamine depletion in the brain, as seen with DaTscan imaging on a subset of prodromal and PD patients. Another outcome was finding connectivity measures in delta and theta bands to be of high importance in identifying PD patients with Mild Cognitive Impairment and finding correlations between memory and connectivity in the theta band, attention and connectivity in the beta band. On following up on a subset of these patients for 5 years, theta spectral power and connectivity were found to have the strongest association with the change in cognition, in line with our hypothesis based on all the baseline studies. Our data suggest that theta activity can be a diagnostic marker for PD and prognostic marker for cognitive decline in Parkinson’s disease, eventually leading to dementia.