This tool aims to automatically model the topological folding structure of the human hippocampus, and computationally unfold the hippocampus to segment subfields and generate hippocampal and dentate gyrus surfaces.

The overall workflow can be summarized in the following steps:

  1. Pre-processing MRI images including non-uniformity correction, resampling to 0.3mm isotropic subvolumes, registration and cropping to coronal-oblique subvolumes around each hippocampus

  2. Automatic segmentation of hippocampal tissues and surrounding structures via deep convolutional neural network U-net (nnU-net), models available for T1w, T2w, hippocampal b500 dwi, and neonatal T1w, and post-processing with fluid label-label registration to a high-resolution average template

  3. Imposing of coordinates across the anterior-posterior and proximal-distal dimensions of hippocampal grey matter via solving Laplace’s equation, and using an equivolume solution for laminar coordinates

  4. Generating transformations between native and unfolded spaces using scattered interpolation on the hippocampus and dentate gyrus coordinates separately

  5. Applying these transformations to generate surfaces meshes of the hippocampus and dentate gyrus, and extraction of morphological surface-based features including thickness, curvature, and gyrification index, sampled at the midthickness surface, and mapping subfield labels from the histological BigBrain atlas of the hippocampus

  6. Generating high-resolution volumetric segmentations of the subfields using the same transformations and volumetric representations of the coordinates.

Full Documentation: here

Relevant papers: