Parkinson’s Disease Brain States and Functional Connectivity: A Machine Learning Analysis of Neuroimaging Data

Abstract

[en] The goal of this study is to identify and characterize brain states as a function of the motivation with which the task was performed (the presence of avatars and their skill at performing the task). To this end, we developed a series of machine learning algorithms capable of capturing differences between the EEG data recorded at each condition. We used metrics of local activity, such as electrode power, of similarity (correlation between electrodes), and of network functional connectivity (co-variance across electrodes) and use them to cluster brain states and to identify network connectivity patterns typical of each motivated state. Studies in the field of computational neuroscience involve the analysis of brain dynamics across specific brain areas to study the mechanisms underlying brain activity. This particular study aims at discovering how brain activity is affected by social motivation by computational means. To this end, we analyzed a dataset of electro-encephalographic (EEG) data recorded previously during a reward-driven decision-making experiment performed by Parkinson patients. The goal of the experiment was to select and perform a reaching movement from an origin cue to one of two possible wide rectangular targets. Reward was contingent upon arrival precision. Social motivation was manipulated by simulating avatar partners of varying skill with whom our participants played. Competition with the avatar was explicitly discouraged. Our results show that the presence of different avatars yielded distinct brain states, characterized by means of functional connectivity and local activity. Specifically, we observed that motivation related states were best identified for the highest frequency band (gamma band) of the EEGs. In summary, this study has shown that brain states can be characterized by level of motivation with a high degree of accuracy, independently of the presence of medication.