PMC

Experimental design and the regseg tool. The proposed tool performs simultaneous segmentation and registration of dMRI features (the FA and the ADC maps) through a nonlinear mapping aligned with the phase-encoding (PE) axis of the echo-planar images (EPI). This data paper provides detailed information with figures, graphs and text of how the necessary “golden”-standard to validate regseg was obtained, and the mathematical foundations of the method.

Susceptibility distortions are challenging in dMRI. The artifact causes a misalignment of the structures of the brain (represented by contours overlaid on the T1-weighted -T1w- image of panel A) and the dMRI data (as depicted in panel B). In panel C we present a close-up of the frontal lobe of the diffusion image, where the warping of the echo-planar image (EPI) produces a mismatch with respect the “anatomically-correct” surfaces extracted from the T1w image. The warping is aligned with the phase-encoding (PE) direction of the image. In this case (panels B, C) the PE direction is the anterior-posterior axis. Since the distortion is related to the inhomogeneity of the field inside the scanner, some regions are not excessively affected by the artifact (white box in panel C). In this data paper, the methodology and instruments to generate “a priori” known distortions from real subjects that can be used as “golden”-standard in the validation of registration and segmentation processing tools for diffusion MRI.

The data for the verification and validation of the elements involved in the connectome extraction are valuable due to the absence of reference-standards. The analysis of structural connectivity networks extracted from dMRI data involves a convoluted processing flow comprising a large set of chained computational tools. Unit-test verification and validation of these tasks is crucial to assess the reliability of the whole process, and a challenging effort due to the lack of gold standards. In a joint registration and segmentation method that implicitly tackles with the susceptibility-derived distortion artifact is proposed, and evaluated on the surfaces as a surrogate of the goodness of the cortical parcellation. The involved elements in are denoted with orange-color boxes. In this paper, we provide the data and the software instruments used to generate a “golden”-standard required in the evaluation of the segmentation and registration task.