Acceleration of filtered back-projection algorithm for 3D cone-beam CT reconstruction using parallel computation

Filtered back-projection (FBP) is the most commonly used reconstruction method for computed tomographic image reconstruction due to its practicality in comparison with other computationally demanding methods. However, the FBP is still a very computationally expensive algorithm to implement, given that real world data obtained from CT machine are relatively large. Memory management and arithmetic computation of the FBP algorithm on conventional PC-based sequential executions still pose challenges in terms of computation time required for demanding diagnostic requirements. In this work, we propose and develop an approach for GPU-based implementation of the FBP algorithm that fully utilizes the parallel computing power of the graphic processing unit (GPU). We confirmed that the computation performance is significantly improved by applying our proposed approach with actual data obtained from our in-house dental CBCT machine called DentiiScan. The experiments show that we are able to achieve up to 80 times of computation time reduction.