Comprehensive Summary
This systematic review, conducted by Naderi and Jahanian-Najafabadi, explores the development of machine learning models with electroencephalogram (EEG) data to discover novel biomarkers for obsessive-compulsive disorder (OCD). Following PRISMA guidelines, 42 studies from Pubmed, Scopus, Web of Science, and IEEE databases were screened, of which 11 were included in the final analysis. From their results, they found that the usage of EEG data and machine learning to analyze OCD is relatively novel - only one paper was published on the topic prior to 2020. Furthermore, the authors found that a significant portion of the studies analyzed (54%) lacked methods for screening patients for OCD, which is partly due to their heavy reliance on public datasets, which may lack some critical information, such as that regarding demographics. Due to the high variability in experimental design, data summarization, and materials used among the 11 studies, it remains inconclusive whether the usage of machine learning and EEG data to analyze OCD biomarkers is effective. The authors suggest that standardized study designs and practices should be developed for future studies in this field of research, and future researchers should better account for the issue of model overfitting.
Outcomes and Implications
This study aims to assess whether there may be biomarkers predictive of OCD using machine learning and EEG data. OCD is a psychological disorder that affects ~3.5% of the global population, and the diagnosis of OCD is often significantly delayed or misdiagnosed for other disorders. Because of this, novel methods should be devised to more accurately and efficiently diagnose OCD – the usage of EEG data and machine learning is a promising field in this regard. To move forward, the authors propose that publicly-assessible EEG datasets should be expanded and standardized to encompass a larger set of demographics and increase reproducibility across studies examining different patient populations.