Cone-beam computed tomography reconstruction accelerated with CUDA

For years, Cone-beam Computed Tomography (CBCT) represents a state-of-art CT technique and catch many researchers and doctors' attention for its great efficiency. However, the 3D reconstruction requires billions of calculation blocking CBCT to have real-time reconstruction. With the help of novel Graphic Process Unit (GPU) parallel computing structure, Compute Unified Device Architecture, known as CUDA, this paper proposes a practical CBCT reconstruction method derived from original FDK method, which changes the whole reconstruction process from serial computation into parallel one. This method accelerates the whole reconstruction process by 150 times faster than CPU method, making it possible for real-time volume reconstruction. To deal with highly noise contaminated projection data, this paper also discusses a modified filter to gain better reconstructed image without adding extra computation.