Research Of Dose Evaluation And3D Reconstruction In Radiation Treatment Planning System Based On CUDA

With the development of radiotherapy and its related technologies, medical image datasize of patient is becoming bigger and bigger, the amount of data which Radiation TreatmentPlanning System (RTPS) dealt with has been greatly increased. In Clinical, doctors are oftenforced waiting for the end of RTPS calculation, which usually costs a very long waiting time.So it is necessary to improve and accelerate RTPS. In recent years the development of theCPU has met the bottleneck. CPU performance ascension now is usually performed bymulti-core architecture. Correspond to the development of CPU, in recent years GPU goforward very fast, better than multicore CPU system both in speedup performance and price.NVIDIA Company launched CUDA (Compute Unified Device Architecture)architecture in2007, which made people do not need to consider too much hardware detailswhen developing the GPU applications. Aiming to solve the problem of slow speed in RTPSdose assessment and3D reconstruction, this paper uses CUDA to research and implement theacceleration. Main works of the paper are:(1) Analyzed traditional DVH figure statistical algorithms according to theheterogeneous programming model of CUDA. Through the analysis of the storage model ofCUDA, optimized the data storage of DVH. Later, due to the serious bank conflicts of thevoting process in DVH figure statistics and the characteristics of the heavy ion dosedistribution, puts forward a voting algorithm to avoid bank conflicts. Using CUDA accelerateDVH figure statistical algorithm for the parallel acceleration in RTPS dose evaluation module.(2) Analyzed the transfer function in the process of volume rendering. In view of theparameter gradient amplitude which transfer function is often used, this paper uses thethree-dimensional sobel operator for precise calculation. At the same time, in order to solvethe problem of large amount of the body data gradient calculation, use CUDA parallelcomputing gradient amplitude.(3) Analysis and research the volume rendering algorithm ofmedical image. Analysis and implement light casting algorithm on CUDA. Comparison withthe experimental data, achieved the ideal rendering speed. According to the research results,show that the algorithms of this paper achieved a good acceleration rate in DVH figurestatistics and volume rendering reconstruction.