Summary table of the train and test datasets.

The ABIDE dataset is publicly availablea, and contains images acquired at 17 sites, with a diverse set of acquisition settings and parameters. This heterogeneity makes it a good candidate to train machine learning models that can generalize well to novel samples from new sites. We selected DS030 [20] from OpenfMRIb as held-out dataset to evaluate the performance on data unrelated to the training set. A comprehensive table showing the heterogeneity of parameters within the ABIDE dataset and also DS030 is provided in Block 2 of S1 File.