CUDA accelerated cone‐beam reconstruction

Cone-Beam Computed Tomography (CBCT) is an imaging method that reconstructs a 3D representation of the object from its 2D X-ray images. It is an important diagnostic tool in the medical field, especially dentistry. However, most 3D reconstruction algorithms are computationally intensive and time consuming; this limitation constrains the use of CBCT. In recent years, high-end graphics cards, such as the ones powered by NVIDIA graphics processing units (GPUs), are able to perform general purpose computation. Due to the highly parallel nature of the 3D reconstruction algorithms, it is possible to implement these algorithms on the GPU to reduce the processing time to the level that is practical. Two of the most popular 3D Cone-Beam reconstruction algorithms are the Feldkamp-Davis-Kress algorithm (FDK) and the Algebraic Reconstruction Technique (ART). FDK is fast to construct 3D images, but the quality of its images is lower than the quality of ART images. However, ART requires significantly more computation. Material ART is a recently developed algorithm that uses beam-hardening correction to eliminate artifacts. In this thesis, these three algorithms were implemented on the NVIDIA\u27s CUDA platform. These CUDA based algorithms were tested on three different graphics cards, using phantom and real data. The test results show significant speedup when compared to the CPU software implementation. The speedup is sufficient to allow a moderate cost personal computer with NVIDIA graphics card to process CBCT images in real-time