The NTU-DSI-122 template as a flexible platform for fiber orientation distribution registration of diffusion microstructure into stereotaxic space

Abstract

The registration of neuroimaging subjects to a common stereotaxic space allows for comparisons between timepoints, subjects, or with reference atlases of regions of interest. Typically, this has been performed by computing a similarity metric between the voxel-wise intensity values in the subject image and a reference image. Diffusion MRI is a method that provides far more detailed voxel-wise information than a simple scalar intensity value. This allows for the potential to use this additional information to perform registration with added within-tissue contrast. In this study, we present a novel use of the NTU-DSI-122 template as a fiber orientation distribution (FOD) template, for the purpose of registering subject dMRI images to stereotaxic space. The reliability and accuracy of this FOD-based registration method are compared to the intensity-based registration method ANTs by registering cellular microstructure maps from two separate cohorts to the NTU-DSI-122 template. The stochastic FOD-based method significantly outperformed the stochastic intensity-based metric on reliability and was able to more consistently register the same subject multiple times independently. The FOD-based method also significantly outperformed the intensity-based metric on registration accuracy by more completely aligning the microstructure maps to the template as measured by the Sorenson-Dice coefficient at multiple percentile thresholds. The NTU-DSI-122 template has the additional benefit of including multiple b-value shells between a wide range of feasible acquisition schemes, making the platform a flexible option for registering acquisitions of varying quality, including clinically acquired data.