Few diagnostic classifiers have been applied to OCD across multiple scanners and sites. Prior studies using structural MRI data to classify OCD using single-site samples yielded accuracies ranging from 0.72 to 0.9314. The wide range of performances observed in our individual site classification is in agreement with the published literature. Such a wide range may in part be explained by sample size, as larger samples tended to have higher AUC values16,19,34. However, this relationship does not necessarily hold true for large-scale multicenter studies, due to heterogeneity that arises from pooling samples with different scanning parameters, processing pipelines, inclusion criteria, demographic and clinical characteristics14,21. All these factors can impact the data and obscure a pattern of abnormalities shared by all patients. Monocenter studies that minimize heterogeneity may therefore yield higher classification performances, but limit the generalizability to new, unseen data and its use in clinical practice16,17. Thus, whereas small monocenter studies focus on answering a specific question about their patient population, large multicenter studies assume that a fundamental pattern of the disorder of interest can be detected despite the presence of
