A1115
Title: Linear effect of inter-scanner variability: Insights from paired cross-scanner T1-weighted images in elderly subjects
Authors: Dana Tudorascu - University of Pittsburgh (United States) [presenting]
Abstract: Collecting structural MRI data across sites increases statistical power and enables the generalization of research outcomes; however, due to the variety of imaging acquisition, inter-scanner variability hinders the direct comparability of multi-scanner MRI data. Thus, many harmonization methods have been proposed to reduce inter-scanner variability in the image domain. Although proposed methods, especially incorporating deep learning techniques, have achieved promising performance, interpretability and understanding of inter-scanner variability were still limited. A small sample of eighteen cognitively normal participants is investigated, each scanned on four different 3T scanners, including GE, Philips, Siemens-Prisma and Siemens-Trio, during a short period of time (at most a few weeks apart). A statistical harmonization method, ComBat, was applied and extended to the image domain and investigated the linear effect of inter-scanner variability and image quality metrics. Furthermore, it is attempted to harmonize cross-scanner images by removing the estimated site effect. Besides estimating parametric maps of side effects, image quality metrics were calculated using MRIQC and similarity index to investigate the manifestation of scanner-related variation. Voxel-based morphometry using CAT12 was used to estimate cortical volumetric measures to compare the difference before and after the harmonization.
