White matter atlas of the human spinal cord with estimation of partial volume effect.

Template-based analysis has proven to be an efficient, objective and reproducible way of extracting relevant information from multi-parametric MRI data. Using common atlases, it is possible to quantify MRI metrics within specific regions without the need for manual segmentation. This method is therefore free from user-bias and amenable to group studies. While template-based analysis is common procedure for the brain, there is currently no atlas of the white matter (WM) spinal pathways. The goals of this study were: (i) to create an atlas of the white matter tracts compatible with the MNI-Poly-AMU template and (ii) to propose methods to quantify metrics within the atlas that account for partial volume effect. The WM atlas was generated by: (i) digitalizing an existing WM atlas from a well-known source (Gray's Anatomy), (ii) registering this atlas to the MNI-Poly-AMU template at the corresponding slice (C4 vertebral level), (iii) propagating the atlas throughout all slices of the template (C1 to T6) using regularized diffeomorphic transformations and (iv) computing partial volume values for each voxel and each tract. Several approaches were implemented and validated to quantify metrics within the atlas, including weighted-average and Gaussian mixture models. Proof-of-concept application was done in five subjects for quantifying magnetization transfer ratio (MTR) in each tract of the atlas. The resulting WM atlas showed consistent topological organization and smooth transitions along the rostro-caudal axis. The median MTR across tracts was 26.2. Significant differences were detected across tracts, vertebral levels and subjects, but not across laterality (right-left). Among the different tested approaches to extract metrics, the maximum a posteriori showed highest performance with respect to noise, inter-tract variability, tract size and partial volume effect. This new WM atlas of the human spinal cord overcomes the biases associated with manual delineation and partial volume effect. Combined with multi-parametric data, the atlas can be applied to study demyelination and degeneration in diseases such as multiple sclerosis and will facilitate the conduction of longitudinal and multi-center studies.