Topography-Based Registration of Developing Cortical Surfaces in Infants Using Multidirectional Varifold Representation

Islem Rekik, Gang Li, Weili Lin, Dinggang ShenMICCAI 2015 · DOI

Abstract

Accurate registration of infant cortical surfaces is essential for longitudinal studies of early brain development. However, the rapid growth and complex folding of the neonatal brain pose significant challenges for traditional registration methods. This paper presents a novel framework using a multidirectional varifold representation within a Large Deformation Diffeomorphic Metric Mapping (LDDMM) framework. By combining surface normals with principal curvature directions and adaptively placing control points on gyral crests and sulcal fundi, the proposed method achieves superior alignment of complex anatomical structures compared to existing techniques.

Introduction

Mapping the development of the human cortex during the first months of life is a major frontier in neuroscience. Between birth and six months, the brain undergoes rapid growth and dramatic increases in cortical folding. To compare these changes across subjects or time points, researchers rely on surface registration—the process of warping one brain surface to match another.

Existing registration methods often suffer from limitations. Spherical inflation techniques can introduce significant data distortion. Other frameworks, such as those based on "currents" or standard "varifolds," typically only consider surface normals (the direction perpendicular to the surface). This approach fails to capture the tangential "flow" or orientation of the cortical folds themselves, making it difficult to accurately align the intricate patterns of sulci (valleys) and gyri (ridges) in the developing brain.

Methodology

The authors address these limitations through two primary innovations:

1. Multidirectional Varifold Representation

Instead of relying solely on surface normals, the proposed method decomposes the cortical surface into two distinct components: a normal varifold and a tangential varifold. The tangential component is derived from the principal curvature direction, which effectively captures the orientation of cortical folds. To mitigate noise in flatter regions where curvature is less defined, a variational diffusion equation is employed to smooth these directions across the entire surface, creating a continuous and informative field for registration.

Multidirectional Varifold Representation

2. Topography-Based Control Point Placement

Standard registration algorithms often place "control points"—the mathematical handles used to drive the deformation—uniformly or based on generic variability. In contrast, this framework adaptively places control points specifically on the most anatomically stable landmarks: the gyral crests (hills) and sulcal fundi (valleys). By utilizing supervertex partitioning to identify these regions, the authors ensure that the warping process is guided by the most critical features of the brain's topography.

Topography-Based Control Points

Experimental Results

The framework was evaluated using longitudinal inner cortical surfaces from 12 infants, each scanned at birth and again at 6 months. The method was benchmarked against three state-of-the-art techniques: diffeomorphic spectral matching, current-based matching, and standard varifold-based matching.

The proposed multidirectional varifold approach demonstrated significant improvements across several metrics:

  • Accuracy: It achieved the lowest vertex-wise distance error of 0.60 ± 0.65 mm.
  • Alignment: It reached an average Dice overlap of 0.88 ± 0.006 across 36 anatomical regions of interest.
  • Robustness: The performance gains were most pronounced in highly variable and convoluted regions, such as the precuneus and the rostral anterior cingulate cortex, where traditional methods often struggle to maintain anatomical consistency.

Conclusion

By integrating directional fold information and leveraging stable anatomical landmarks, the multidirectional varifold framework provides a more accurate and biologically grounded approach to infant brain registration. This method paves the way for the creation of high-resolution cortical atlases and more sensitive clinical studies of early neurodevelopmental disorders.

Summary updated

BibTeX
@inproceedings{rekik2015topography,
  author = {Islem Rekik and Gang Li and Weili Lin and Dinggang Shen},
  title = {{Topography-Based Registration of Developing Cortical Surfaces in Infants Using Multidirectional Varifold Representation}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2015},
  doi = {10.1007/978-3-319-24571-3\_28},
}
Figure