A Comparison of Deformable Image Registration Algorithms as They Apply to 4DCT Lung Images
B Loughery*1,2, W Song3, (1) San Diego State University, San Diego, CA, (2) Wayne State University, Detroit, MI, (3) University of California, San Diego, La Jolla, CASU-E-J-77 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To evaluate four intensity-based deformable image registration (DIR) algorithms as they apply to 4DCT lung images.
Methods: The four algorithms evaluated were one- and two-way optical flow, and one- and two-way Demons. All algorithms were coded in-house using MATLAB. 4DCT datasets of ten patients were selected randomly from clinic for the study. Each code was used to register the end-of-exhalation (50% phase) image to all other nine phases. This was repeated for all ten patients. The relative performance of each algorithm was evaluated by calculating the sum of square difference (SSD) of all pixels between the moving and reference images, during each iteration of registration. Elapsed registration duration (in CPU time) was recorded as well.
Results: In general, the Demons algorithm operated much quicker than the optical flow codes. The two-way algorithms were often better than the one-way algorithms, in terms of SSD. Two-way algorithms produced the best final SSD in 83% of registrations. Thus, our results suggest that the optimal choice is the two-way Demons algorithm. The two-way optical flow algorithm also performed reasonably well for small deformations, but took at least twice the time compared to the two-way Demons. In terms of the final registrations, all algorithms resulted in visually reasonable image registrations. Therefore, the differences were in the duration of registration and SSD.
Conclusion: The two-way Demons algorithm proved to be the best quantitative method for performing DIR on 4DCT of lung anatomy. All two-way algorithms were generally slower than the one-way algorithms, but their accuracy increased in terms of SSD. Therefore, we suggest using the two-way Demons for 4D planning of lung cancer radiotherapy.