Obsessive-compulsive disorder detection using ensemble of scalp EEG-based convolutional neural network

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Obsessive-compulsive disorder (OCD) significantly impacts individuals' lives due to intrusive thoughts and compulsive behaviors.
• Early diagnosis of OCD is crucial for effective management and treatment.
• Current diagnostic methods can be time-consuming and may not capture the full spectrum of the disorder.

Purpose of the Study:

• To develop and evaluate deep learning models for the early diagnosis of OCD using Electroencephalography (EEG) signals.
• To compare the performance of individual pre-trained Convolutional Neural Network (CNN) models with an ensemble approach.
• To investigate the efficacy of transfer learning and optimization algorithms in EEG-based OCD detection.

Main Methods:

• Utilized raw scalp-EEG data from OCD patients.
• Developed and fine-tuned three pre-trained CNN models: EEGNet, Shallow ConvNet, and Deep ConvNet.
• Implemented an ensemble of these CNN models using weighted majority voting, with weights optimized by the Differential Evolution (DE) algorithm.

Main Results:

• Shallow ConvNet demonstrated strong individual performance with 85.91% accuracy.
• The ensemble model achieved superior results, reaching 87.03% accuracy, 82.21% sensitivity, and 96.69% specificity.
• The proposed hybrid model effectively extracted distinctive EEG features for accurate OCD identification.

Conclusions:

• Deep learning models, particularly an ensemble of CNNs, show significant promise for the early and accurate diagnosis of OCD using EEG signals.
• Transfer learning and optimization techniques enhance the diagnostic capabilities of EEG-based systems.
• The developed hybrid model offers a potential non-invasive tool for clinical OCD screening.