Electroencephalography-based machine learning for cognitive profiling in Parkinson's disease: Preliminary results

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Cognitive symptoms are prevalent in Parkinson's disease (PD).
• Characterizing cognitive profiles is crucial for predicting cognitive decline.
• Identifying predictors of cognitive worsening is essential for patient management.

Purpose of the Study:

• To explore the utility of resting-state electroencephalography (EEG) combined with data-mining for cognitive profile characterization in PD.
• To develop and validate machine learning models for classifying cognitive impairment severity.

Main Methods:

• Utilized dense EEG data from 118 Parkinson's disease patients.
• Performed spectral power analysis across 7 frequency bands.
• Employed support vector machines and k-nearest neighbors algorithms for model training and testing on 100 and 18 patients, respectively.

Main Results:

• Achieved high classification accuracies: 84% for support vector machines and 88% for k-nearest neighbors.
• Accurate classification was noted for groups requiring precise diagnosis for healthcare planning.
• Classification accuracy was lower for groups with severe cognitive deficits and smaller sample sizes.

Conclusions:

• EEG features, derived from a non-expensive and accessible modality, can serve as a screening tool.
• This approach can effectively identify the severity of cognitive impairment in Parkinson's disease patients.
• The findings support the use of EEG for early detection and management of cognitive issues in PD.