Electroencephalogram-Driven Machine-Learning Scenario for Assessing Impulse Control Disorder Comorbidity in...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Parkinson's disease (PD) patients on dopamine agonist (DA) therapy may develop impulse control disorders (ICDs).
• Cognitive and motor comorbidities significantly increase disability and mortality in PD patients.
• Automatic assessment of ICD comorbidity is crucial for managing PD patients.

Purpose of the Study:

• To develop and validate an electroencephalogram (EEG)-driven machine learning approach for automatic ICD assessment in PD.
• To differentiate PD patients with ICD from those without, and estimate ICD severity.
• To explore the potential of a low-cost, wearable system for real-world application.

Main Methods:

• Utilized a Go/NoGo task to record EEG activity during cognitive and motor inhibition tasks.
• Employed a support vector machine (SVM) for ICD detection and support vector regression (SVR) for severity estimation.
• Trained and tested models on a dataset comprising PD patients with and without ICD, and healthy controls.

Main Results:

• The SVM model achieved 66.3% accuracy in differentiating PD patients with ICD from those without.
• The SVR model showed significantly higher severity scores in the ICD group compared to the PD group.
• The EEG-based approach demonstrated potential for distinguishing ICD presence and severity, outperforming random guessing.

Conclusions:

• An EEG-driven machine learning pipeline can effectively assess ICD comorbidity in PD patients.
• This technology may facilitate the development of wearable computer-aided diagnosis systems for early risk assessment.
• The findings support proactive monitoring for DA-triggered cognitive comorbidities in PD patients.