Cheng Jiale, Zhao Fenqiang, Hu Dan, Cao Chao, Wu Zhengwang, Yuan Xinrui, Han Kangfu, Zhang Lu, Liu Tianming, Zhu Dajiang, Li Gang
Department of Radiology and Biomedical Research Imaging Center, University of North Carolina at Chapel Hill, NC, USA.
Joint Department of Biomedical Engineering, University of North Carolina at Chapel Hill and North Carolina State University, NC, USA.
Proc IEEE Int Symp Biomed Imaging. 2025 Apr;2025. doi: 10.1109/isbi60581.2025.10981138. Epub 2025 May 12.
The cortical 3-hinge gyrus (3HG) and its network (GyralNet) play key roles in understanding the regularity and variability of brain structure and function. However, existing cortical surface registration methods overlook these features, resulting in suboptimal alignment across subjects. Currently, no 3HG and GyralNet atlas exist for registration, and generation of the corresponding atlas requires extensive runtime using traditional methods. To enable better registration of these features, we introduce an unsupervised learning framework to jointly develop 3HGs and GyralNet atlas and register the individual cortical features onto the atlas. To incorporate the graph structure of 3HGs and GyralNet into the registration network, we convert them into surface distance maps, facilitating effective integration. To effectively learn large deformations, a multi-level spherical registration network based on spherical U-Net is introduced to perform registration in a coarse-to-fine manner. Experiments demonstrate our approach's ability to generate 3HGs and GyralNet atlas with detailed patterns and effectively improve registration accuracy.
皮质三铰链回（3HG）及其网络（回网络，GyralNet）在理解脑结构和功能的规律性及变异性方面发挥着关键作用。然而，现有的皮质表面配准方法忽略了这些特征，导致跨个体的对齐效果欠佳。目前，不存在用于配准的3HG和GyralNet图谱，并且使用传统方法生成相应图谱需要大量运行时间。为了实现对这些特征的更好配准，我们引入了一个无监督学习框架，以联合开发3HG和GyralNet图谱，并将个体皮质特征配准到该图谱上。为了将3HG和GyralNet的图结构纳入配准网络，我们将它们转换为表面距离图，便于有效整合。为了有效学习大变形，引入了基于球形U-Net的多级球形配准网络，以粗到细的方式进行配准。实验证明了我们的方法能够生成具有详细模式的3HG和GyralNet图谱，并有效提高配准精度。
