Hippocampal subfields revealed through unfolding and unsupervised clustering of laminar and morphological features in 3D BigBrain

Abstract

The internal structure of the human hippocampus is challenging to map using histology or neuroimaging due to its complex archicortical folding. Here, we aimed to overcome this challenge using a unique combination of three methods. First, we leveraged a histological dataset with unprecedented 3D coverage, 3D BigBrain. Second, we imposed a computational unfolding framework that respects the topological continuity of hippocampal subfields, which are traditionally defined by laminar composition. Third, we adapted neocortical parcellation techniques to map the hippocampus with respect to not only laminar but also morphological features. Unsupervised clustering of these features revealed subdivisions that closely resemble ground-truth manual subfield segmentations. Critically, we also show that morphological features alone are sufficient to derive most hippocampal subfield boundaries. Moreover, some features showed differences within subfields along the hippocampal longitudinal axis. Our findings highlight new characteristics of internal hippocampal structure, and offer new avenues for its characterization with in-vivo neuroimaging.