Functional MRI connectivity based analysis that ranges between simple univariate methods to complex deep-learning pipelines has been employed to differentiate autistic patients from healthy controls on benchmark datasets such as ABIDE. However, the variability induced via multi-site acquisition of data may perturb the underlying prediction model with undesirable consequences. We illustrate that statistical elimination of scanner effects using COMBAT harmonization yields better results and also facilitates in gaining insights into the discriminative connectivity patterns that emerge post harmonization and which correlate with clinical markers.
This abstract and the presentation materials are available to members only; a login is required.