(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: To improve computational efficiency of general simultaneous motion estimation and image reconstruction (G-SMEIR), we pursue a fast image domain motion estimation method.
Methods: To obtain an accurate deformation vector field (DVF) demands a large number of iterations of the original demons (ORD) method. Two variants, multi-resolution demons (MRD) and multi-step demons (MSD), are adapted to alleviate this problem. In MRD, images are downsampled to the lowest resolution for an initial DVF estimation, which is upsampled as an input for the DVF estimation at higher resolution. This procedure repeats until reaching the original resolution. In MSD, the DVFs of any pair of adjacent phases are obtained first. Then the sum of two adjacent DVFs is used as the initial input to DVF of two phases separated by one phase. The same principle can be applied for further away phases. We compared ORD, MRD and MSD using a 4D XCAT phantom based on root mean square error (RMSE) and structural similarity index (SSIM) between the reference phase image and the registered image.
Results: To achieve the similar performance, ORD needs 500 iterations at the original resolution (10.05 s GPU computing for two 256x256x100 volumes, averaged for 50 repetitions). MRD of four resolution levels needs 100 iterations at each level, which is equivalent to around 188 iterations (3.72 s). MSD only needs 100 iterations (4.08 s = 2.98 s for demons + 1.10 s for DVF sum). The best RMSE and SSIM for phase 3 to phase 1 registration are 5.66x10?4 and 0.9929 for ORD (500 iterations), 5.73x10?4 and 0.9932 for MRD, 5.61x10?4 and 0.9929 for MSD. MRD also yields the registered image visually closest to the reference image.
Conclusion: Both of MRD (2.7 times faster) and MSD (2.5 times faster) can significantly improve the computational efficiency for motion estimation of G-SMEIR.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the U.S. National Institutes of Health under Grant No. NIH/NCI R15CA199020-01A1 and NIH/NIBIB R03EB021600-01A1.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
