The present invention discloses a GPU parallel acceleration dual spectrum
CT reconstruction method based on a
CUDA architecture. Combined with the characteristic of
dual energy CT reconstruction, the method of carrying out dual-energy CT rapid analysis reconstruction and obtaining four reconstruction images at the same by using the parallel ability of a GPU based on the
CUDA architecture is provided, and the images comprises a post-
processing and reconstructed high and low energy line
attenuation coefficient image and pre-
processing and reconstructed equivalent
atomic number and
electronic density images. The method has the advantages that the speed is accelerated, the efficiency is promoted, compared with an existing single CT acceleration
reconstruction method, according to the dual-energy
CT reconstruction method, under the premise of registration, the repeated weight calculation of an original projection and a decomposed projection in
a weighting step is avoided, the time of repeatedly calculating the back-projection addresses of the four reconstruction images in a back-projection step is saved, the
data access time of a calculation middle process can be reduced. Multiple real parts and imaginary parts can be fully utilized to carry out
dual energy synchronization filtering in a filtering step, and when a reconstructed image is increased, no more back-projection time consumption is increased.