Study Of DSA Medical Image Denoising And The Acceleration Of GPU

In recent years, the incidence of vascular disease is high, which harm has been more than infectious diseases, and it has become the largest threat for people health in China. Therefore, the diagnosis of these diseases is particularly important. The DSA medical image has an important position in the aspect of angiography. However, in the process of production, storage, processing and transmission, DSA medical images will inevitably introduce low quality, fuzzy of the image and the interference of noise. Furthermore, the information and features of collected images will fade, be covered, even dim. But due to the high resolution of the DSA medical images, it will be very time-consuming no matter what kind of denoising algorithm we use, and traditional serial calculation is difficult to meet the higher requirements for medical image denoising real-timely. Therefore, the doctor will be seriously affected for the diagnosis of the patient's condition.The main work of this paper includes the following sections:First, analyses of the DSA medical image noise, reduction of quality image and recovery model is established for the noise, then selects the experiment noise model. Build the evaluation system on DSA medical image. Time consuming to DSA image denoising of medicine is analyzed, and put forward the noise associated with DSA medical image denoising using GPU acceleration.Second, Put forward and realized for DSA medical image noise using KNN(K on his Neighbors, K Nearest neighbor) algorithm and denoising method for GPU acceleration, for the DSA medical image with noise parameter selection of KNN algorithm, and carries on the serialization implementation. Experiments show that, KNN algorithm on DSA medical image with Gaussian noise removal effect is the best. Through CUDA(Compute Unified Device Architecture, computing Unified Device Architecture) platform for KNN algorithm for single GPU system with many GPU parallel implementation of the system, and get the acceleration of 73.38 and 148.2 times than most.Thirdly, put forward and realized for DSA using medical image noise while NLM(Non-local Means, nonlocal average) algorithm and denoising method for GPU acceleration. Use for the DSA medical image with noise while NLM algorithm and the improved fast while NLM algorithm implementation and GPU acceleration, experiments show that the algorithm on DSA medical image with Gaussian noise removal effect is best, parallel implementation, respectively to get the acceleration of 158.8 and 72.9 times than the largest.Fourthly, comparative analysis of KNN algorithm, while NLM algorithm and its improved algorithm, aiming at DSA in medical image noise removing noise effect and two parallel time differences. Through comparison and analysis, while NLM algorithm of DSA medical image denoising ability slightly better than KNN algorithm and fast while NLM algorithm, KNN algorithm with fast while NLM algorithm basically the same; While NLM algorithm is the most time consuming, quick while NLM algorithm took slightly longer than KNN algorithm, KNN algorithm is implemented on GPU systems, high resolution advantages compared with single GPU is more obvious.