Applying Unsupervised Machine Learning Techniques to Resting-State BOLD Multicenter Neuroimaging of Pre-adolescent Complex Congenital Heart Disease Patients to Enhance Image Harmonization and Predict Motion Artifact Characteristics

Resting-state BOLD MR images are invaluable for evaluating the neurocognitive state of patients, particularly populations at high risk for neurodevelopmental impairment; however, BOLD images are highly susceptible to motion. The combination of machine learning and image reconstruction techniques during and after BOLD image acquisition holds great promise for harmonizing images and recovering motion-corrupted data. However, there is little information about the relationship between unsupervised ML techniques and characteristics of resting BOLD images. We examined resting state BOLD image harmonization and motion in a set of complex congenital heart disease case and healthy control adolescent subjects acquired through a multi-center study.