Abstracts

Rationale: There is an increasing number of brain atlases that, after coregistration to an individual’s MRI, can provide parcellations into multiple regions representing, respectively, sets of anatomical, cyto- and myelo-architectonic, and functionally homogeneous areas. While the alignment from atlas to subject is typically based on anatomy only (through use of a T1-weighted MRI), studies have shown reasonable consistency and repeatability in these maps. By defining multiple distinct parcellations on a single subject (one per atlas), researchers can relate individual features from MRI, PET, SPECT, or electrophysiology to each of these atlases. We describe open-source software that will compute these mappings.

Methods: We have implemented the multi-atlas mapping method using BrainSuite, software that applies a sequence of operations to T1-weighted MRIs to identify cortical and subcortical gray matter, extract a tessellated representation of the cortical surfaces, and then perform surface-constrained volumetric registration of a labelled brain atlas to the individual. For this purpose we use the USCBrain atlas, which parcellates the brain into a total of 130 ROIs. We then map all of the additional brain atlases (see Fig. 1) to the subject via an intermediate mapping to the USCBrain atlas. Most atlases are based either on the MNI or the HCP-grayordinate conventions. We perform a one-time mapping of both these coordinate systems to the USCBrain atlas space using a combination of FreeSurfer and BrainSuite. With these maps in place, we then transfer the labels from each atlas to the USCBrain atlas space. Once this atlas is coregistered to the individual we are then able to directly transfer each of the other atlases to the individual as illustrated in Fig. 1. We note that BrainSuite generates consistent cortical-surface and volumetric segmentations and labels using the USCBrain atlas. It is therefore a unique platform for extensions to multiple atlases, each of which can include either surface or volumetric parcellations. Furthermore, we are able to propagate surface labels into gyral white matter which can be useful in defining regional white-matter connectivity based on diffusion tractography.

Results: Mappings on a single individual for each of the atlases is shown in Fig. 1 on a smoothed cortical surface. As an indication of consistency across software, in Fig. 2 we show the USCBrain atlas and its mapping to an individual computed separately using Brainsuite and FreeSurfer. The multi-atlas software is implemented in MATLAB and compiled using the Compiler Toolbox on Windows, Mac, and Linux platforms, allowing users to run it without a MATLAB license. Typical execution times for the module per parcellation scheme is 1-2 minutes. Open-source code and binaries are available from https://github.com/ajoshiusc/svreg_multiparc.

Conclusions: We have developed a software module for BrainSuite that allows generation of multiple parcellations schemes on individual subject anatomy. The software is open source and integrated with the BrainSuite and BrainStorm software.

Funding: Please list any funding that was received in support of this abstract.: This work is supported by the following grants: R01 NS074980, W81XWH-18-1-0614, R01 NS089212, and R01 EB026299.