Registration for 3D model of human atlas based on vertex information

Authors: Jiao, P.F., Chang, J., Yang, X.S., Jiao, Y., Liu, F.D., Guo, S.H., Bai, R., Ouyang, J. and Zhang, J.J.

Journal: Yiyong Shengwu Lixue/Journal of Medical Biomechanics

Volume: 27

Issue: 5

Pages: 567-571

ISSN: 1004-7220


Objective: To develop a registration method for 3D human atlas models by using geometric information of the vertices so as to lay a foundation for statistical modeling of atlas. Methods: Based on CT images of the normal human, thirty 3D models of human atlases were created and marked by the manual selected points, including 1 standard module, 20 training sets and 9 testing samples. The training sets were first registered with the standard module, including calculation on geometric information of the individual vertex and optimization process of weight coefficients in the registration models. By minimizing the energy function defined with the Euclidean distances between the automatic registered points and the manual selected points in training sets, the optimized weight coefficients could be obtained. The testing samples were then registered with the standard module to calculate the Euclidean distances between the automatic registered points and the manual selected points. The results were then compared with the training sets to evaluate the stability of the registration method. Results: The registration function and the corresponding optimized weight coefficients were obtained, and the average errors for the training sets and testing samples were 1.983 mm and 2.045 mm, respectively. Further statistical analysis showed that there were no obvious differences in the error distributions among the training sets and testing samples. Conclusions: The accuracy and stability of the proposed registration method meet the requirement in medical applications, and it can provide automatic registration of points of interest on human atlas models and be used for element classification in statistical modeling.

Source: Scopus

Preferred by: Jian Chang, Jian Jun Zhang and Xiaosong Yang