Parallelizing strain estimation

Estimating deformation of biological tissue from ultrasound imaging using Normalized Cross Correlation (NCC) is very time-consuming. It would therefore benefit tremendously by a speedup. A major time-reduction could be achieved by parallelizing the separate computations of NCC. On simulated and in vivo data the OpenMP interface on a multi-CPU system has been investigated for parallelization of the NCC computations. The performance of the OpenMP-approach was assessed using 1, 4 and 8 threads. The results were compared with the MATLAB implementation of the same algorithm. The 8 threaded OpenMP approach achieved a maximum reduction factor in computation time of 130 for small templates and a maximum performance of 39 GFLOPS (24.4% of the system´s single precision peak performance) for the largest template. Parallelization of NCC computations using OpenMP significantly reduces the computation time required for estimating deformation in biological tissue from hours to minutes.