Predicting Parkinson's disease using gradient boosting decision tree models with electroencephalography signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Parkinson's disease (PD) diagnosis relies on symptomatic treatments and has limited accuracy with current methods.
• Existing electroencephalography (EEG)-based diagnostic tools face challenges with computational complexity and low accuracy.
• Early and accurate PD diagnosis is crucial for effective patient management.

Purpose of the Study:

• To develop a fast and accurate method for predicting Parkinson's disease (PD) using EEG signals.
• To evaluate the effectiveness of the Hjorth parameter and gradient boosting decision tree (GBDT) algorithm for PD detection.
• To improve upon the diagnostic accuracy of current PD assessment techniques.

Main Methods:

• Utilized an open EEG dataset comprising 41 PD patients and 41 healthy controls (HCs).
• Analyzed EEG signals from two distinct participant cohorts (University of New Mexico and University of Iowa).
• Explored optimal time segments and frequency ranges for Hjorth parameter analysis of PD-related EEG characteristics.

Main Results:

• The CatBoost-based model achieved 89.3% accuracy, 0.912 AUC, and 0.903 F-score in distinguishing PD patients from controls.
• The developed model demonstrated superior performance compared to previous studies and existing pathologic examination-based diagnoses (83.9% accuracy).
• Achieved a significant odds ratio of 115.5, indicating a strong predictive capability.

Conclusions:

• The proposed EEG analysis method using Hjorth parameters and GBDT offers a promising approach for enhancing PD diagnosis.
• This technique is expected to serve as an effective tool for improving the accuracy and efficiency of Parkinson's disease detection.
• The findings suggest a potential for wider clinical implementation of advanced EEG analysis in neurological disorder diagnostics.