Hippocampus-specific fMRI group activation analysis using the continuous medial representation

Paul A. Yushkevich, John A. Detre, Dawn Mechanic-Hamilton, María A. Fernández-Seara, Kathy Z. Tang, Angela Hoang, Marc Korczykowski, Hui Zhang, James C. GeeNeuroImage 2007 · DOI

TL;DR

This paper introduces a novel shape-based approach for analyzing fMRI data specifically within the hippocampus. By focusing analysis on the structure's geometry, they achieve more sensitive and specific group activation maps compared to traditional whole-brain methods, particularly at lower smoothing levels. This has implications for better understanding brain function in regions with high anatomical variability.

What Problem Does This Paper Address?

Analyzing functional magnetic resonance imaging (fMRI) data across multiple subjects to understand brain activity typically involves registering all brains to a common template. However, this "whole-brain" approach often struggles with accurately normalizing smaller, complex structures like the hippocampus, leading to reduced sensitivity and difficulty in distinguishing activation in neighboring regions. Existing region-based analyses can be too simplistic, summarizing an entire structure into a single value. This paper aims to bridge this gap by developing a method for generating detailed, population-level statistical maps specifically for the hippocampus, improving both the accuracy of normalization and the specificity of the analysis.

Methods

The core of the proposed method is the "continuous medial representation" (cm-rep). This technique models the shape of a structure (in this case, the hippocampus) using a synthetic skeleton and a radial field that defines its thickness. This representation allows for the creation of a unique coordinate system that maps points within the hippocampus interior to a common space, based on their position relative to the medial skeleton and the boundary.

Key methodological contributions include:

  • Shape-Based Normalization: Instead of relying solely on image registration, the method uses geometrical correspondences derived from fitting cm-rep models to segmented hippocampi. This allows for more precise alignment of individual hippocampi.
  • Hippocampus-Specific Coordinate System: The cm-rep model enables the creation of a personalized coordinate system for each hippocampus, extending from its medial axis to its boundary. This allows for analysis of activation within the hippocampus's interior.
  • Non-Uniform Smoothing: A novel smoothing scheme is introduced that weights the contribution of each fMRI voxel based on its overlap with the hippocampus. This aims to reduce the influence of signal from outside the hippocampus, thereby increasing specificity without necessarily sacrificing sensitivity.

Datasets

The study utilized fMRI data from healthy volunteers. Specifically, datasets from 18 subjects (9 male, 9 female) were used for the validation experiment. The fMRI data were acquired using a 3 Tesla Siemens Trio Scanner with different voxel sizes (2.5x2.5x2.5 mm³ and 3x3x3 mm³). The experimental paradigm was a visual scene encoding task.

Key Insights and Ideas

  • Shape Matters for Normalization: The paper strongly suggests that accurately capturing the shape of a structure is crucial for effective normalization, especially for complex regions like the hippocampus. Traditional whole-brain methods, while powerful, can fail to adequately align these smaller structures.
  • Specificity Gains at Lower Smoothing: The hippocampus-specific approach demonstrates improved statistical significance at lower smoothing levels compared to whole-brain analysis. This indicates that by focusing the analysis and using shape-informed normalization, researchers can achieve greater specificity without a severe loss of sensitivity.
  • Non-Uniform Smoothing's Potential: The non-uniform smoothing scheme, tailored to the hippocampus's shape, shows promise in reducing the impact of surrounding tissue on activation measurements within the hippocampus. While it didn't dramatically increase sensitivity, it did not appear to lower it either, suggesting a potential for improved specificity.
  • Beyond Activation Peaks: The cm-rep coordinate system opens doors for more detailed analyses, such as comparing activation inside and outside the hippocampus, and potentially incorporating structural features like thickness into activation models.

Key Results

The study found that the hippocampus-specific random effects (RFX) analysis yielded more significant statistical maps than whole-brain analysis using SPM2, particularly at lower smoothing levels (e.g., 3mm FWHM). For instance, at 3mm FWHM smoothing, the adjusted p-values for the structure-specific method were consistently lower across various regions of interest within the hippocampus compared to whole-brain analysis. The paper also showed that the non-uniform smoothing scheme did not negatively impact sensitivity compared to uniform smoothing.

Limitations and Future Work

  • Manual Segmentation Dependency: In the current study, the hippocampus was manually segmented. The authors note that in future work, they plan to extend the cm-rep approach to fit deformable templates directly to structural MRI data, reducing the reliance on manual segmentation.
  • Segmentation Quality: The effectiveness of the shape-based normalization is dependent on the quality of the initial segmentations. The paper acknowledges that this can be a disadvantage if segmentations are of insufficient quality.
  • Broader Applicability: While the study focused on the hippocampus, the authors suggest that the cm-rep approach could be extended to other brain structures involved in memory and potentially other imaging modalities beyond fMRI.
  • Probabilistic Atlases: Future work includes fitting the cm-rep system to histology and post-mortem imaging data to create probabilistic atlases of the hippocampus and its subfields, which would allow for less manual processing in associating activation peaks with specific subregions.

Summary updated

BibTeX
@article{yushkevich2007hippocampus,
  author = {Paul A. Yushkevich and John A. Detre and Dawn Mechanic-Hamilton and María A. Fernández-Seara and Kathy Z. Tang and Angela Hoang and Marc Korczykowski and Hui Zhang and James C. Gee},
  title = {{Hippocampus-specific fMRI group activation analysis using the continuous medial representation}},
  journal = {NeuroImage},
  year = {2007},
  doi = {10.1016/j.neuroimage.2007.01.029},
}

Loading related content...

Figure