Multi-contrast anatomical subcortical structures parcellation

Abstract

The human subcortex is comprised of more than 450 individual nuclei which lie deep in the brain. Due to their small size and close proximity, up until now only 7% have been depicted in standard MRI atlases. Thus, the human subcortex can largely be considered as terra incognita. Here, we present a new open-source parcellation algorithm to automatically map the subcortex. The new algorithm has been tested on 17 prominent subcortical structures based on a large quantitative MRI dataset at 7 Tesla. It has been carefully validated against expert human raters and previous methods, and can easily be extended to other subcortical structures and applied to any quantitative MRI dataset. In sum, we hope this novel parcellation algorithm will facilitate functional and structural neuroimaging research into small subcortical nuclei and help to chart terra incognita.

Keywords: anatomical parcellation; human; neuroscience; quantitative MRI; subcortex.

Figure 1.. The 17 subcortical structures currently included in the parcellation algorithm in axial (A), sagittal (B), and coronal (C) views.

Figure 2.. The MASSP parcellation pipeline.

Atlas priors for interfaces between structures are combined to the MRI data, regularized via probability diffusion and topology correction, and the final structure posteriors are jointly estimated by region growing.

Figure 3.. Leave-one-out validation of the structures parcellated by MASSP, compared to the human rater with most neuroanatomical expertise.

Scores for the left and right side are computed separately and then combined into box-and-whisker plots.

Figure 4.. Inter-rater variability for the human expert raters.

Scores for the left and right side are computed separately and then combined into box-and-whisker plots.

Figure 5.. Parcellation with Freesurfer (top, on T1w image) and MASSP (bottom, on T2w image) on Human Connectome Project data.

MASSP priors were not derived from the contrasts, but transferred via a spatial mapping of the quantitative MRI intensities from AHEAD subjects.

Figure 6.. MASSP parcellation scores as a function of increasing number of subjects included in the atlas.

Scores for the left and right side are computed separately and then combined into box-and-whisker plots.

Figure 7.. MASSP parcellation scores over the lifespan.

Each matrix show the average Dice overlap (top), dilated Dice overlap (middle), and average surface distance (bottom) for using one age group as prior (’train’) to parcellate another age group (’test’).

Figure 8.. Regression of volume (log scale), structure thickness, R1, R2*, and QSM MRI parameters estimated using manual delineations versus MASSP automated parcellations.

Circles show individual data points, linear regression is indicated by a straight line, and 95% confidence interval is given as the shaded area. Pearson correlation coefficients are indicated when significant (p-value<0.01).

Figure 9.. MP2RAGEME maps and delineations: quantitative R1 (left), quantitative R2* (middle), QSM (right).

Manual delineations for the 17 structures of interest are overlaid on all images.

Figure 10.. Anatomical interface (A) and skeleton (B) priors derived from the 10 manually delineated subjects.

Figure 11.. Successive parcellation results: (A) voxel-wise posteriors and parcellation, (B) diffused posteriors and parcellation, (C) topology-corrected posteriors and final region-growing parcellation.