T2 relaxometry for myelin water fraction: an ex vivo brain imaging study

Abstract

Abstract Introduction: Myelin water imaging provides a non-invasive approach for indirectly investigating myelin content of the brain by magnetic resonance Imaging (MRI). Myelin water fraction (MWF) reflects the fraction of water signal associated with water trapped between myelin layers. Postmortem imaging provides a unique opportunity to validate MWF metrics against gold-standard postmortem neuropathology. However, there is a need for validated sequences that would be applicable in postmortem settings. In this study, we validated and compared the applicability of turbo spin echo (TSE) and gradient-and-spin-echo (GRASE) techniques for myelin water imaging in formaldehyde fixed postmortem human brains. Methods: 40 postmortem human brain hemispheres were scanned with TSE and GRASE sequences. Acquired data were reconstructed using a non-parametric multicomponent T2 relaxometry approach using a consistent framework to support comparison between GRASE- and TSE-derived measures. Agreement between derived measures was quantified at both voxel and regional level. Maps derived from the full 32-echo GRASE reconstruction were compared against a truncated 14-echo reconstruction to examine the contribution of the longer echoes to the derived maps. Finally, a random forest regression model was trained based on TSE data to predict the GRASE-derived MWF maps. Results: Both sequences provided robust multicomponent T2 characterization across the brain, and the derived measures showed anatomical patterns consistent with expected differences across tissue types. TSE and GRASE derived maps had moderate to strong agreement at voxel (0.41< {rho} < 0.82, all pFDR<0.001) and regional levels (0.87< {rho} < 0.98, all pFDR<0.001), with consistently stronger agreement observed for regional metrics. Echo reduction comparisons between the full 32-echo GRASE reconstruction and a truncated 14-echo reconstruction showed minimal impact of echo truncation (all {rho} > 0.98, pFDR < 0.001). The nonlinear random forest regression model trained on TSE maps was able to accurately reproduce the MWF maps derived from GRASE (r = 0.91, RMSE = 8.87). Discussion: TSE and GRASE32 capture strongly related multicomponent T2 information in fixed postmortem human brain tissue, but with systematic and tissue-dependent quantitative differences across derived measures. These differences suggest that direct interchangeability should not be assumed and that cross-sequence mapping or calibration is required when quantitative equivalence is desired.