Multi-Center Correction Functions for Magnetization Transfer Ratios of MRI Scans

Persistent differences among Magnetic Resonance Imaging (MRI) obtained with different MRI scanners or different pulse sequences is currently a major challenge worldwide, where large scale clinical trials, cross-sectional studies, or longitudinal studies are being conducted [1,2]. Routine analysis of thousands of brain and spinal cord MRI images is being conducted worldwide which were taken with various MRI scanners, pulse sequences, magnetic fields, and time points. Artificial objects, called “phantoms”, are widely being used to periodically calibrate and tune individual MRI scanners in order to minimize variation of MRI scans obtained by the same scanners at different time points. This also reduces the variation of the MRI scan outputs obtained with different scanners. When an MRI scan is ordered in real life, it is possible that there are only a few MRI scanners available at the given area, hence the choice of MRI settings may be limited. Although this is not likely to lead to a misdiagnosis that is based on a single MRI scan, it has a significant quantitative effect that impacts clinical measurement and may impair clinical decision making that is based on longitudinal MRI scans, which often contain scans from more than one scanner or setting. For the purpose of large scale cross-sectional and longitudinal data analysis, clinical trials, and following long term patient progression of specific neurological disorders, it is important to have MRI images which are taken under identical, or at least similar, conditions. Therefore, the existence of persistent, unwanted variation among scanners is a significant problem for which correction functions could be the solution. The key question in this article is “Can we obtain optimal predictive correction functions, which can be used to correct voxel based Magnetization Transfer Ratios (MTR) obtained at individual MRI scanners/pulse sequences in order to estimate what they would have been if the scans had been taken at different MRI scanners/pulse sequences”? Several approaches have been investigated for reducing interscanner variability and bias. Studies were performed for reproducibility issues and the implications of inter-scanner variability for clinical trials [3-10]. For instance, the Functional Imaging Biomedical Informatics Research Network (FBIRN) Consortium [7] investigated the reduction of the inter-scanner variability of activation in a multicenter MRI study by controlling for signal-to-fluctuation-noise-ratio differences. Several studies have identified the sources of MTR variations [11,12] and provided various correction schemes including correction for interscanner variations. However, there is no existing standard solution for standardizing multiple scanner MRI data and for obtaining predictive correction functions resulting from multiple MRI scanners. Moreover, there are a number of challenges when one attempts to achieve such goals.