JISE


  [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]


Journal of Information Science and Engineering, Vol. 38 No. 1, pp. 253-271


Drosophila Brain Aligner: Registering 3D Point Clouds in 2D Parameterization Domain


Hao-Chiang Shao1,+, Lu-Hung Hsu2, Yung-Chang Chen2,
Ying-Chu Huang3 and Shih-Ting Huang4
1Department of Statistics and Information Science
Fu Jen Catholic University
New Taipei, 242 Taiwan

2Department of Eletrical Engineering
National Tsing Hua University
Hsinchu, 300044 Taiwan

3Service Systems Technology Center
Industrial Technology Research Institute
Hsinchu, 310401 Taiwan

4Department of Statistics
University of California, Irvine
Irvine, CA 92697, USA


In order to study brain functions and connectome, scientists need to register and warp source image data of some model organism, e.g. Drosophila, into a pre-defined standard atlas. Such registration and warping procedure is conventionally driven and constrained by point clouds extracted from contour edges of manually-segmented landmark tissues within a 2D+Z image volume. However, it is difficult to register two dense 3D point clouds. The fitness between spatial distributions of two point clouds cannot guarantee the matchness between 3D anatomical surfaces from which point clouds were sampled. Hence, to settle down this problem, we propose in this paper a strategy to register 3D point clouds of Drosophila brain in 2D parameterization domain. Our contributions are twofold. First, instead of registering point clouds directly, our method was designed to register two mesh surface models, each defined by a to-be-registered point clouds, so that the anatomical shape details described by the point cloud can be aligned after registration. Second, the proposed method performs registration/warping in a parameterization domain, and hence it no longer needs a rigid transformation to globally align and scale the input models. Experiments show that the surface-to-surface distance is reduced after registration and warping process. For models with an about 1100-voxel-long bounding box diagonal, the average surface-to-surface distance is reduced to about 0.1 voxel after registration. The proposed method is effective.


Keywords: point cloud registration, surface registration, parameterization, Drosophila brain, mesh surface model

  Retrieve PDF document (JISE_202201_13.pdf)