Mapping Surface Area Regionalization of Hippocampal Formation in Early Development

2156 

Abstract Submission 

Ya Wang¹, Liangjun Chen¹, Zhengwang Wu¹, Sheng-Che Hung¹, Li Wang¹, Weili Lin¹, Gang Li¹

¹University of North Carolina at Chapel Hill, Chapel hill, NC 27599, USA

Ya Wang  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

Liangjun Chen  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

Zhengwang Wu  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

Sheng-Che Hung  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

Li Wang  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

Weili Lin  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

Gang Li  
University of North Carolina at Chapel Hill
Chapel hill, NC 27599, USA

The hippocampal formation, which is a convoluted, thin subcortical structure, involves multiple crucial cognitive functions relating to specific hippocampal subregions (Hunsaker and Kesner, 2013). Current research indicates the hippocampal formation exhibits a distinct functional connectivity-based parcellation, which differs from traditional histology-based partition along the medial-lateral axis (Genon, et al., 2021). However, a paucity of studies delves into hippocampal internal organization in terms of early dynamic development, which is essentially related to both the underlying microstructure and functional connectivity. We thus aim to explore the development-based surface area regionalization of hippocampal formation by using high-resolution longitudinal MRI scans covering the first two postnatal years.

We used 513 longitudinal T1w and T2w brain MR images (resolution = 0.8*0.8*0.8 mm³) from 231 healthy subjects during the first two postnatal years from Baby Connectome Project dataset (Howell et al., 2019). All scans were processed with the infant brain extraction and analysis toolbox, iBEAT V2.0 (Wang et al., 2023). To map the surface area regionalization of the hippocampal formation, we utilized our created 4D infant brain volumetric atlases (Chen et al., 2022), which provide manually delineated hippocampal labels. When building 4D infant atlases, we obtained the deformations between age-specific atlases and between each age-specific atlas and the age-matched scans (Avants et al., 2011; Chen et al., 2023). To obtain the vertex-to-vertex correspondences of hippocampal surfaces across subjects and ages, we reconstructed the hippocampal surface mesh representation in 0-month atlas and then warped it to each scan at each age by combining the corresponding deformations between age-specific atlases and between each age-specific atlas and the age-matched scans. For example, for a 2-month individual scan, we concatenated the deformation fields in the following order: 0-month atlas to 1-month atlas, 1-month atlas to 2-month atlas, 2-month atlas to 2-month scan. Finally, the local surface area of each vertex was calculated on each surface and a matrix was formed with each column denoting the local areas of all vertices of a hippocampal surface and each row denoting the local areas of all scans at the same vertex. Then, a data-driven and hypothesis-free method, non-negative matrix factorization (NMF) (Lee and Seung, 1999), was adopted to partition hippocampal surface into a set of developmentally heterogeneous regions by clustering co-developing vertices into same regions (Wang et al., 2019; Sotiras et al., 2017).

Silhouette coefficient and reconstruction error were jointly used to ascertain the optimal subregion number. Fig. 1 shows the values of silhouette coefficient and reconstruction error across different numbers of hippocampal subregions separately for bilateral hippocampi using NMF method. We opted for 7 as the proper hippocampal subregion number, considering the high local maximum of silhouette coefficient, relatively low reconstruction error, and more symmetric patterns in bilateral hippocampal subregions. The hippocampal surface partition and its spatial position relative to the brain were shown in Fig. 2. Two distinct developmental patterns of hippocampal surface area were discerned, with one pattern adhering to the tripartite subdivision, including hippocampal head (regions 1 and 5), body (regions 2, 4, 6, and 7), and tail (region 3). Another pattern followed a medial-lateral partition, with the medial portion consisting of regions 4, 5, and 6, while the lateral portion comprising regions 1, 2, and 7. Of note, there is no specific medial-lateral division for region 3.

This work firstly explores the surface area regionalization of hippocampal formation from developmental view and proposes important means for exploring the development of hippocampus-related cognition during early postnatal years.

Segmentation and Parcellation ²

Normal Development

Subcortical Structures ¹

Anatomical MRI

Data analysis

Development

NORMAL HUMAN

STRUCTURAL MRI

Sub-Cortical