Abstract
This paper presents a novel framework for hippocampal shape analysis using medial surfaces. By skeletonizing the 3D hippocampus into a 2D sheet and mapping local thickness, the authors provide a granular method to quantify shape variations. The approach reveals gender and age-related differences that traditional volumetry overlooks, such as specific local translations and rotations in females and stronger age-related volume correlations in males.
Introduction
The hippocampus (HC) is a critical brain structure for memory and emotion, often showing volume loss in conditions like Alzheimer’s, depression, and PTSD. While clinicians have long used volumetry (measuring total size) to track these changes, volume is a "blunt instrument." It indicates that a structure is smaller, but not where or how the shape has changed. Existing 3D shape models are often computationally unreliable or difficult to register across different subjects. This study aims to provide a more granular, statistically robust method to quantify 3D shape, specifically to investigate differences in how age and gender affect the HC.
Methodology
The authors employ a framework based on medial surfaces—essentially a 3D "skeleton" that runs through the middle of an object. The process transforms a complex 3D shape into a 2D "thickness map," allowing for standard pixel-by-pixel statistical comparisons across a population.
- Divergence-Based Extraction: They use a "grassfire flow" model where the boundary of the HC moves inward at a constant speed. The points where these fronts meet form the medial surface.
- Topology Preservation: To ensure the skeleton accurately represents the HC, a thinning process preserves "homotopy" (the object's basic connectivity), ensuring no holes or cavities are accidentally created or destroyed.
- Dimensionality Reduction: Because the HC's medial surface is mostly a single, flat-ish sheet, the researchers use Principal Component Analysis (PCA) to find the two main axes of this sheet and project the 3D medial surface onto a 2D plane.
- The Height Function: For every point on this 2D plane, they record the "radius" (the distance from the skeleton to the original boundary).
Dataset and Processing
The study utilized a dataset of 80 healthy subjects (39 men, 41 women) between the ages of 18 and 42. T1-weighted MR images were used as the primary data source.
The hippocampi were manually segmented—the "gold standard" for accuracy—and then normalized into a standard stereotaxic space to account for differences in head size and orientation.
Key Results and Insights
One of the most important findings was that the medial surface of the HC is dominated by one large sheet. Reconstructing the HC using only this sheet resulted in less than a 1% volume error, justifying the use of a 2D projection for statistical analysis.
Hemispheric and Gender Differences
- Asymmetry: The right HC was found to be significantly larger than the left ($4300$ vs $4072\text{ mm}^3$, $p < .001$). The right HC is also generally thicker than the left.
- Age Correlations: In men, there was a significant correlation between age and HC volume ($r = -0.44$), which was almost non-existent in women ($r = 0.01$).
- Beyond Shrinkage: Gender differences aren't just about size; they involve local translation and rotation. In females, the "head" of the HC moves laterally while the "tail" moves medially compared to males.
Local Shape Changes
- Females: Showed a "thickening" of the alveus (a layer of white fiber) with age in the medial anterior region.
- Males: Showed significant decreases in the anterior and posterior regions, specifically corresponding to the uncal recess and the inferior horn of the lateral ventricles.
Conclusion and Future Work
The medial surface framework provides a detailed view of hippocampal morphology that traditional volumetry misses. While the study relies on time-consuming manual segmentation, the ability to map local thickness onto a 2D plane offers a powerful tool for neuroimaging research. Future work could benefit from including total volume as a covariate in the shape regression to see if shape changes persist even when controlling for overall size, and integrating other shape features like boundary curvature into the height function.
Summary updated