Room: Exhibit Hall | Forum 5
Purpose: To evaluate the performance of 3D model-based global-to-local deformable algorithms for surface registration of multi organs in prostate cancer, by comparing a 3D global affine transformation-based algorithm.
Methods: This study included 220 prostate cancer patients who underwent IMRT or VMAT in prone position. Three registration algorithms were employed. An algorithm A was based on a 3D global affine transformation-based algorithm. Algorithms B and C were the extended algorithms combining 3D global affine transformation-based method with local deformable registration algorithms (the Laplacian surface deformation for B and the finite element method for C). First, the contours of the prostate, seminal vesicle, rectum and bladder on CT images were converted to PLY file format. Next, statistical shape models of these organs were independently constructed from the PLY files for first 20 patients. Finally, surface registration was done for the PLY files of the remaining 200 patients, using the three algorithms. The Hausdorff distance (HD) was calculated for the surfaces of the organs of interest. The experiment was conducted on a personal computer with 32 GB RAM, Intel Xeon E5-2687W v4 (Dual-core CPU, 3.0 GHz).
Results: In A, the 75th percentile HD was 3.9 mm for prostate and greater than 7.8 mm for other organs. Meanwhile, the 75th percentile HD was less than 0.7 mm in B. The 75th percentile HD was 0.3 mm for prostate in C, which was comparable to that in B. However, the 75th percentile HDs were greater than 1.0 mm in C, except for prostate. A median computation time to complete registration was less than 16, 101, and 30 s in A, B and C, respectively.
Conclusion: The 3D model-based global-to-local deformable algorithms completed surface registration in a short time and had sufficient registration accuracy.