Objective: The objective of this study was to determine the resting-state cortical oscillations and the machine learning model that best classifies Parkinson’s disease (PD) patients with freezing of gait (PDFOG+) and PD patients without FOG (PDFOG–) using cortical oscillatory features.

Background: FOG is one of the most debilitating motor symptoms in the late stage of Parkinson’s disease (PD) as it may lead to falls and impact quality of life. The pathophysiology of FOG is poorly understood in PD; however, our previous reports have suggested the presence of abnormal theta and beta oscillations in the cortico-basal ganglia networks in PDFOG+ compared to PDFOG–. However, cortical oscillations have not yet been extensively investigated to distinguish between PDFOG+ and PDFOG– using machine learning models based on resting-state scalp electroencephalography (EEG) recordings.

Method: EEG recordings of 83 PD patients (42 PDFOG+ / 41 PDFOG–) and 41 healthy age-matched controls were collected during resting-state condition for 3 minutes. We segmented EEG signals into 3 seconds epochs and converted time-domain signals into frequency domain. We exported the mean normalized power values from each frequency band. We used normalized power values from each epoch for classification models. We implemented our models on all frequency bands and channels. We used six different machine learning algorithms to classify PDFOG+ from other groups: Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), Bayes, and Deep Neural Network (DNN). We employed k-fold cross-validation approach for validating the results.

Results: Our machine learning classifying methods demonstrate that DNN model with frontal cortical theta (4-7 Hz) oscillations and combined oscillatory power of theta and beta (13-30 Hz) bands differentiate PDFOG+ from PDFOG– and healthy controls with higher accuracy, precision, recall, and F1-score values compared to other models, frequency bands, and cortical regions.

Conclusion: Our study leads to the understanding of the cortical characteristics of PDFOG+ during the resting-state condition, that can help in improving the objective classification of PDFOG+. Future studies to further improve and validate the performances of our models in clinical practice are warranted.

« Back to 2022 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/cortical-oscillatory-feature-based-classification-of-parkinsons-disease-with-freezing-gait/