Improved stability of whole brain surface parcellation with multi-atlas segmentation

Yuankai Huo, Shunxing Bao, Prasanna Parvathaneni, Bennett A. Landman

Introduction

In the field of neuroimaging, brain surface parcellation—the division of the cortical surface into distinct anatomical or functional regions—is a fundamental step for quantitative analysis. Accurate parcellation allows researchers to perform region-of-interest (ROI) studies, measuring localized cortical thickness, surface area, and volume. While FreeSurfer has long served as the industry standard, its "surface-to-volume" approach can encounter stability issues. Researchers at Vanderbilt University have introduced MaCRUISEsp, an extension of the Multi-atlas CRUISE (MaCRUISE) method, designed to provide high-stability whole-brain surface parcellation through a robust "volume-to-surface" strategy.

Methodology: The MaCRUISEsp Pipeline

The MaCRUISEsp framework transitions from raw 3D MRI data to fully labeled brain surfaces through a multi-stage process that emphasizes topological integrity and spatial consistency.

The MaCRUISEsp pipeline overview, from raw T1w MRI input to final parcellated surfaces.

1. Volume Segmentation Based Surface Parcellation (VSBSP)

The pipeline begins with high-quality volumetric segmentation, dividing the 3D voxels of the brain into 132 distinct labels using multi-atlas techniques. These labels are then projected onto the "central surface"—the mid-layer of the gray matter—using a nearest-neighbor approach. This central surface serves as a stable anchor for the subsequent steps.

2. Topological Correction (TC)

Direct projection from volume to surface often introduces "noise," such as fragmented regions or disconnected label "islands." To address this, MaCRUISEsp employs a Topological Correction step. This ensures that each anatomical label consists of exactly one connected component on the surface mesh. An iterative "nearest neighbor filling" strategy is used to merge small, disconnected components into the primary region, ensuring anatomical plausibility.

3. Surface Label Propagation

Once the central surface is cleanly parcellated, the labels are propagated inward to the white matter (inner) surface and outward to the pial (outer) surface. This propagation uses Euclidean distance calculations restricted by the surface's normal plane. This restriction is critical to prevent "label bleeding" across narrow sulci, where two different brain folds are physically close but anatomically distinct.

Experimental Evaluation

To assess the reliability of MaCRUISEsp, the researchers utilized the Kirby21 dataset, a gold standard for scan-rescan reproducibility studies. The dataset includes 21 subjects (11 male, 10 female) who were each scanned twice. The primary metric for evaluation was the Dice Similarity Coefficient (DSC), which measures the spatial overlap between the parcellations of the two scans; a DSC of 1.0 indicates perfect reproducibility.

Results and Discussion

The evaluation demonstrated that MaCRUISEsp offers exceptional stability, particularly for the central and inner surfaces.

Visual comparison of unparcellated MaCRUISE surfaces and the labeled MaCRUISE-SP outputs across inner, central, and outer layers.
  • Central Surface Stability: MaCRUISEsp achieved a median DSC of 0.948, indicating extremely high consistency across scans.
  • Comparative Performance: On the inner surface, MaCRUISEsp (DSC 0.929) outperformed the FreeSurfer baseline (DSC 0.905).
  • Granularity: Notably, MaCRUISEsp utilized the BrainCOLOR protocol providing 98 cortical labels, compared to FreeSurfer’s 75 labels. Achieving higher DSC scores with a more granular labeling scheme underscores the robustness of the MaCRUISEsp approach.

However, the study also identified areas for improvement. FreeSurfer maintained an edge on the outer (pial) surface (DSC 0.881 vs. MaCRUISEsp's 0.835). This suggests that the current label propagation method may require further refinement to better navigate the complex geometry of the brain's outermost folds.

Conclusion

MaCRUISEsp represents a significant advancement in brain surface parcellation, offering a highly stable alternative to traditional methods. By prioritizing the central surface and enforcing topological constraints, it provides researchers with a reliable tool for detailed ROI analysis. Future work will likely focus on improving outer surface propagation and integrating these parcellations into multi-modal analyses, such as Diffusion Tensor Imaging (DTI) mapping.

Summary updated

Abstract

Whole brain segmentation and cortical surface parcellation are essential in understanding the brain’s anatomicalfunctional relationships. Multi-atlas segmentation has been regarded as one of the leading segmentation methods for the whole brain segmentation. In our recent work, the multi-atlas technique has been adapted to surface reconstruction using a method called Multi-atlas CRUISE (MaCRUISE). The MaCRUISE method not only performed consistent volumesurface analyses but also showed advantages on robustness compared with the FreeSurfer method. However, a detailed surface parcellation was not provided by MaCRUISE, which hindered the region of interest (ROI) based analyses on surfaces. Herein, the MaCRUISE surface parcellation (MaCRUISEsp) method is proposed to perform the surface parcellation upon the inner, central and outer surfaces that are reconstructed from MaCRUISE. MaCRUISEsp parcellates inner, central and outer surfaces with 98 cortical labels respectively using a volume segmentation based surface parcellation (VSBSP), following a topological correction step. To validate the performance of MaCRUISEsp, 21 scanrescan magnetic resonance imaging (MRI) T1 volume pairs from the Kirby21 dataset were used to perform a reproducibility analyses. MaCRUISEsp achieved 0.948 on median Dice Similarity Coefficient (DSC) for central surfaces. Meanwhile, FreeSurfer achieved 0.905 DSC for inner surfaces and 0.881 DSC for outer surfaces, while the proposed method achieved 0.929 DSC for inner surfaces and 0.835 DSC for outer surfaces. Qualitatively, the results are encouraging, but are not directly comparable as the two approaches use different definitions of cortical labels.
BibTeX
@inproceedings{huo2018improved,
  author = {Yuankai Huo and Shunxing Bao and Prasanna Parvathaneni and Bennett A. Landman},
  title = {{Improved stability of whole brain surface parcellation with multi-atlas segmentation}},
  booktitle = {Image Processing},
  year = {2018},
  doi = {10.1117/12.2281509},
}
Figure