The derivative back-projection filtered algorithm for a helical cone-beam CT is a newly developed exact reconstruction method.Due to its large computational complexity

the reconstruction is rather slow for practical use.General purpose graphic processing unit(GPGPU) is an SIMD paralleled hardware architecture with powerful float-point operation capacity.In this paper

we propose a new method for PI-line choice and sampling grid

and a paralleled PI-line reconstruction algorithm implemented on NVIDIA’s Compute Unified Device Architecture(CUDA).Numerical simulation studies are carried out to validate our method.Compared with conventional CPU implementation

the CUDA accelerated method provides images of the same quality with a speedup factor of 318.Optimization strategies for the GPU acceleration are presented.Finally

influence of the parameters of the PI-line samples on the reconstruction speed and image quality is discussed.