Comprehensive Summary
This paper explores how electroencephalography (EEG) and machine learning methods have been utilized to classify patients with Obsessive-Compulsive Disorder. Following PRISMA guidelines, the authors used both classical and deep learning methods such as support vector machines (SVMs), convolutional neural networks (CNNs), and long short-term memory (LSTM) models. These models aimed to classify EEG signal patterns, primarily by analyzing frequency-domain features like theta, alpha, beta, and gamma bands, functional connectivity, and time–frequency characteristics. Deep neural networks like PsyNet and CNN-LSTM hybrids exceeded 90% accuracy and specificity and sensitivity values over 85%, though accuracies of most models ranged from 56.9% to 100%. According to their observations, the highest performing models all shared multi-band spectral analysis, artifact correlation, and multichannel EEG that focused on frontal, temporal, and occipital regions where theta, alpha, and gamma power were most discriminative for OCD. Despite this, many inconsistencies were identified such as lack of standardized preprocessing protocols, inconsistent validation methods, and no explainable AI tool which creates uncertainty about the usage of EEG for OCD. Because of these inconsistences, while EEG-based machine learning has shown high accuracy and promise as a diagnostic tool for OCD, it is not yet developed enough to relate to clinical practices.
Outcomes and Implications
OCD affects about 3.5% of the global population and therefore enables this review paper to hold medical significance as these EEG-based Machine learning systems could transform clinical assessment and management of OCD, which often goes undiagnosed or unaddressed. Current diagnostic criteria relies heavily on clinical interviews, self-reporting techniques, and behavioral observations which introduces significant bias. EEG biomarkers can improve the speed and accuracy of diagnosis by providing neurophysiological indicators of OCD-specific activity, specifically abnormal frontotemporal gamma oscillations and reduced alpha coherence. Clinically, such models could aid in early detection, monitor treatment response, or guide targeted neuromodulation therapies. High-performing models such as the CNN-LSTM and PsyNet architectures that achieved accuracies between 95% and 98%, identified EEG channels F7, T5, and O2 as discriminative, suggesting that these cortical regions play central roles in OCD pathophysiology. However, this review states that most models and studies lacked external replication and real-world testing, meaning clinical application remains early. To reach clinical readiness, future research must expand participant diversity beyond the predominantly Asian cohorts, employ standardized EEG protocols, and incorporate explainable AI tools to clarify how models reach diagnostic conclusions. Regardless of this, the high accuracy of some models allows the conclusion that EEG-ML diagnostics are promising for OCD to enhance its psychiatric diagnoses and assessment.