FIG. 3.

Parallel computation for 4D image registration. (a) Procedures of using parallel computation to register different Z-movies. The initial step was to determine the cardiac cycle. Each CPU core was assigned to initialize the calculation process for one movie, and the GPU was utilized to match a targeted image in red with others in yellow that correspond to the same cardiac phase in subsequent cardiac cycles within the same Z-movie. The matched frame was marked with an additional red box. The next step involved aligning Z-movies in different depths to start from the same cardiac phase. Each CPU core was assigned to initialize the calculation process between two consecutive movies, and the GPU was used to match the targeted image in red with others in yellow that correspond to the same cardiac phase in the next Z-movie. (b) Quantitative comparison of the processing time between parallel computing and sequential computing. The registration time for different numbers of frames (on the left) and movies (on the right) improved by over tenfold in both the overall process (at the top) and in individual steps (at the bottom).