Research On Fast Volume CT Reconstruction Algorithms

Comparing with the traditional CT, volume CT has distinctive competitive advantages in the clinic medical field, which results in the rapid increase of usage in hospitals, especially in Oral Medicine and Interventional Therapy fields. However, volume CT image reconstruction is time-consuming problem which has become a barrier and influents the clinic efficiency hardly. Additionally, the projections scanned by volume CT usually is not in accordance with their angles collected by capture card, which may cause artifacts in the reconstructed images. Therefore, it is very important for developing and optimizing the volume CT system to handle the two problems introduced above.Using the great properties of parallel and Floating-Point power of GPU, a fast volume CT reconstruction algorithm is proposed based on the idea of"pixel level parallel". Then, to reduce the repetitive operation, the geometry computation and pixel computation are separated,which improves the computation efficiency again. At last, based on the background of medical engineering, the parallel of volume CT scan and reconstruction is implemented. The results demonstrated that the time of volume CT reconstruction was only 1/70 of that in common PC.According to the problems appeared in the project of"Real Time 3D Imaging System", the cases of"all projections have the same angel error"and"all projections have different angel errors"are deeply analyzed. Based on the back projection formula of FDK, the result shows that the first case has no any affection on causing artifacts and the second case has effect on the artifacts depend on the variance of the angel errors. And, the result is verified by some experiments.The idea of"pixel level parallel"is also expanded to other two image processing fields. Firstly, the algorithm structure of back projection computation is transplanted to dose calculation of radioactive particle in TPS. The time of dose calculation is reduced to 1/35 of that before in CPU. Secondly, the idea of"pixel level parallel"is expanded to the field of image enhancement in spatial domain. With structured design method, computing model, data and algorithm resource which are indispensability in GPU computation are encapsulated. This architecture removes the configuration details of GPU computation and reduces the repetitive work. In addition, new algorithms of enhancement in spatial domain could be added conveniently in the architecture. The time of executing algorithms could reduce to 1/136 of that before in CPU, which meets the demand of real time system.