CMSC 2023 | Fixing inconsistent MRI data using image harmonization

Lianrui Zuo, MSE, The Johns Hopkins University, Baltimore, MD, describes the findings of a recent study on inconsistent MRI acquisition in longitudinal volumetric analysis. Researchers began this study after they noticed inconsistencies between MRI scans of the same patient with multiple sclerosis and realized that this was because they had visited a different scanner that was still following the same guidelines. This study had two objectives – to investigate the impact of inconsistent MRI acquisition across multiple sites by longitudinal volumetric analysis and to explore a potential solution to this. MRI data from 8 locations across the USA, including more than 200 patients, was used for three parallel analyses. The first focused on one site to show a real-world scenario where data are consistently acquired. The second used data from all eight sites, representing data that are not consistently acquired. The third used data from all eight sites that had undergone image harmonization – a type of machine learning that standardizes image contrast. They found that the first analysis achieved almost perfect performance, whereas the second analysis was significantly worse. The third analysis was almost identical to the first, suggesting that image harmonization is an important tool for data consistency that can be used in real-world scenarios. This interview took place at the Consortium of Multiple Sclerosis Centers (CMSC) Congress 2023 in Aurora, CO.

These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.