The Parallel Processing Based on CUDA for Convolution Filter FDK Reconstruction of CT

Computed tomography (CT) technology has been used in many fields. But the slow speed of CT image reconstruction is unbearable in some situation. The parallel processing based on graphic processing unit (GPU) is a great option to accelerate the speed of CT image reconstruction. But the general purpose GPU program model is difficult to use for the texture programming. Compute unified device architecture (CUDA) program model is much easier to use. A parallel processing based CUDA for CT image convolution filter FDK reconstruction was proposed in this article. The acceleration of FDK algorithm with convolution filtering on a single GPU had been realized. Under the condition that the precision was assured, the speed of CT image reconstruction has been accelerated about 80 times in the kernel algorithm and about 11 times in the overall performance according to the experiment result.