https://hgpu.org/?p=4465
CUDA-based acceleration and algorithm refinement for volume image registration