Normalization of Magnetization Transfer Ratio MRI For Multicentre Clinical Trials

Introduction Magnetization transfer ratio (MTR) magnetic resonance (MR) imaging is a promising technique for measuring in vivo demyelination and remyelination, providing a more direct measurement of activity in diseases like multiple sclerosis (MS). Incorporation of MTR in multicentre imaging protocols is feasible using stock sequences and MT pulses. However, such MTR values are sequence and scanner-dependent. Standardized protocols that minimize MTR differences across scanners require a level of control over the MT pulse characteristics that is unavailable in large, multicentre studies. In addition, significant shifts in MTR contrast have been observed in individual scanners after upgrades. Therefore, multicentre and longitudinal studies would benefit from MTR normalization to map raw MTR data onto a common scale. If this scale is semi-quantitative, comparison to published data becomes possible. Normalization to an external standard is cumbersome and depends upon the stability of the standard. Using an internal standard in studies of patients with diseases such as MS which may have substantial diffuse pathology is not ideal, as numerous studies have shown that normal-appearing tissues in MS exhibit abnormal MTR, and often it is these changes that are of particular interest. A method to account for centre-to-centre variance in MTR images as an integrated part of the analysis process has recently been proposed. In this method, MTR data from various centres is analyzed using a statistical model that includes the centre as a factor, in addition to any factors of interest, such as treatment vs. placebo. We currently propose a method to normalize MTR images directly using a reference scan of a normal subject acquired at each site. Methods Five MS patients and one control were scanned at each of Site 1 and Site 2. Additionally, all the patients scanned at Site 2 were subsequently scanned at least once at Site 1. For all subjects at all timepoints, gray matter (GM) and white matter (WM) probability maps were calculated. For each centre’s control subject, median MTR values were calculated for the > 85% probability GM (GMMTR-C) and WM (WMMTR-C). MTR data acquired at each site was normalized according to MTRnorm = MTRraw −GMMTR−C (WMMTR−C −GMMTR−C ) .