Classifying Obsessive-Compulsive Disorder from Resting-State EEG using Convolutional Neural Networks: A Pilot Study.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Classifying obsessive-compulsive disorder (OCD) using brain data is challenging.
• Resting-state electroencephalography (EEG) is a noninvasive, affordable neuroimaging technique.
• Traditional machine learning methods have limitations in predicting OCD from EEG.

Purpose of the Study:

• To investigate the efficacy of convolutional neural networks (CNNs) for classifying OCD using minimally processed EEG time-frequency representations.
• To compare CNN performance against traditional machine learning approaches like support vector machines (SVMs).
• To explore the impact of multimodal fusion, including clinical and demographic data, on classification accuracy.

Main Methods:

• Collected resting-state EEG data from 20 participants (10 OCD, 10 healthy controls).
• Transformed EEG segments into time-frequency representations using Morlet wavelets.
• Applied a 2D CNN classifier with leave-one-subject-out cross-validation and compared it to an SVM trained on spectral band power features.

Main Results:

• The CNN achieved 82.0% accuracy (AUC: 0.86), significantly outperforming the SVM baseline (49.0% accuracy, AUC: 0.45).
• Most clinical variables did not enhance classification accuracy beyond EEG data alone (80.0% accuracy).
• Incorporating education level improved classification performance to 85.0% accuracy (AUC: 0.89).

Conclusions:

• CNNs applied to resting-state EEG show significant potential for diagnosing OCD, surpassing traditional methods.
• Deep learning techniques hold promise for psychiatric applications, despite sample size limitations.
• Education level may serve as a valuable complementary feature for OCD classification, meriting further research in larger cohorts.