Related Experiment Video
Updated: May 5, 2026
11:12
Controlling Parkinson's Disease With Adaptive Deep Brain Stimulation
Published on: July 16, 2014
25.1K
A Comparative Study of Machine Learning and Deep Learning Models for Automatic Parkinson's Disease Detection from
Sankhadip Bera1, Zong Woo Geem2, Young-Im Cho3
1Department of Information Technology, Jadavpur University, Jadavpur University Second Campus, Plot No. 8, Salt Lake Bypass, LB Block, Sector III, Salt Lake City, Kolkata 700106, India.
Diagnostics (Basel, Switzerland)
|March 28, 2025
Summary
This study introduces an innovative electroencephalogram (EEG) approach using machine learning and deep learning for early Parkinson's disease (PD) detection. The deep learning model achieved over 99% accuracy, offering a reliable, non-invasive diagnostic tool.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Computational Biology
Background:
- Parkinson's disease (PD) is a prevalent neurodegenerative disorder affecting over 1% of individuals over 60.
- Early PD detection is challenging due to unclear brain characterization, necessitating improved diagnostic techniques.
- Electroencephalogram (EEG) signals offer potential for developing efficient and reliable PD detection methods.
Purpose of the Study:
- To introduce an innovative EEG-based method for detecting and classifying Parkinson's disease (PD) patients.
- To compare the efficacy of machine learning (SVM) and deep learning (CNN) models for PD detection using EEG data.
- To establish a more accurate and non-invasive approach for early PD diagnosis.
Main Methods:
- Utilized two EEG datasets (UC San Diego Resting State EEG and IOWA) for analysis.
- Extracted spectral features from five key EEG frequency bands (alpha, beta, theta, gamma, delta).
- Employed Support Vector Machine (SVM) as a baseline classifier and a Convolutional Neural Network (CNN) as a deep learning approach.
Main Results:
- SVM achieved 82% and 94% accuracy on the two datasets, respectively, in a subject-dependent scenario.
- CNN significantly outperformed SVM, reaching accuracies exceeding 96% and 99% on the respective datasets.
- In a subject-independent environment, SVM achieved 68.09% accuracy, highlighting the need for generalizable models.
Conclusions:
- Advanced feature extraction combined with deep learning (CNN) provides a highly accurate, non-invasive method for PD diagnosis.
- The developed approach shows significant potential for reliable and efficient early detection of Parkinson's disease.
- Future research should focus on enhancing feature sets, increasing subject numbers, and improving model generalizability.
