Comparison of ComBat harmonization methods for longitudinal magnetic resonance imaging data in a travelling subject cohort
Sophie Richter1, Stefan Winzeck1,2, Marta M Correia3, Evgenios N Kornaropoulos4, Anne Manktelow1, Joanne Outtrim1, Doris Chatfield1, Jussi Posti5, Olli Tenovuo5, Guy B Williams6, David K Menon1, and Virginia F J Newcombe1
1Division of Anaesthesia, University of Cambridge, Cambridge, United Kingdom, 2BioMedIA Group, Department of Computing, Imperial College, London, United Kingdom, 3MRC Cognition and Brain Sciences Unit, University of Cambridge, Cambridge, United Kingdom, 4Diagnostic Radiology, Lund University, Lund, Sweden, 5Department of Neurosurgery and Turku Brain Injury Centre, Turku University Hospital and University of Turku, Turku, Finland, 6Wolfson Brain Imaging Centre, Department of Clinical Neurosciences, University of Cambridge, Cambridge, United Kingdom
The trend in neuroimaging towards multi-site studies requires validated harmonization approaches to eradicate scanner differences which mask the biological effect of interest. Here, the harmonization algorithm ComBat and its modification for longitudinal data (LongComBat) were compared on a large travelling subject sample (n=23 for structural MRI and n=31 for diffusion tensor MRI).
In structural data scanner difference are not apparent in unharmonized data but can be created by harmonization. For DTI data, scanner differences in unharmonized data are large and, both ComBat and LongComBat successfully diminished those in most regions of interest, with LongComBat achieving slightly lower false positive rates.
This abstract and the presentation materials are available to members only;
a login is required.