Deformable surface mesh registration is a useful technique for various medical applications, such as intra-operative treatment guidance and intra- or inter-patient study. In this paper, we propose an automatic deformable mesh registration technique. The proposed method iteratively deforms a source mesh to a target mesh without manual feature extraction. Each iteration of the registration consists of two steps, automatic correspondence finding using robust point-matching (RPM) and local deformation using a radial basis function (RBF). The proposed RBF-based RPM algorithm solves the interlocking problems of correspondence and deformation using a deterministic annealing framework with fuzzy correspondence and RBF interpolation. Simulation tests showed promising results, with the average deviations decreasing by factors of 21.2 and 11.9, respectively. In the human model test, the average deviation decreased from 1.72±1.88 mm to 0.57±0.66 mm. We demonstrate the effectiveness of the proposed method by presenting some medical applications.
Bibliographical noteFunding Information:
This research was supported in part by the KIST, South Korea institutional program ( 2E26210, 2E26276 ), and the NIH, United States ( R01EB016777 ).
© 2016 Elsevier Ltd
- Automatic correspondence
- Deformable registration
- Mesh deformation
- Radial basis function
- Robust point matching