Research On The Suppression Method Of Blood Speckle Noise From Intravascular Ultrasound Image Sequences Based On CUDA

With intravascular ultrasound images in the diagnosis and therapy for cardiovascular disease is gradually widely used, using digital image processing technology to carry on the analysis of the IVUS images of the vascular wall, vascular lumen shape, can assist doctors to understand the disease, interventional therapy and surgery to develop. Combining with clinical cardiovascular medicine application, based on the IVUS image denoising effect and the problem of processing time, the thesis, aimed at the suppression of blood speckle noise from intravascular ultrasound image sequences and improvement of the denoising efficiency,has certain theoretical meaning and engineering value.Firstly, the original IVUS image sequences are pre-processed. For IVUS image by the curvature change of blood vessel and heart beating from image sequence caused dislocation,the heart beat cycle time rules, resampling new image to the image in the same phase of the image data, ensuring the consistency of image sequence space and time, and to the subsequent image to the accuracy of noise to provide a guarantee.Secondly, this paper analyzes and studies the noise suppression of IVUS image sequence. On the basis of the spatio-temporal correlation denoising algorithm, the algor:ithm for the correction of the gray value of blood flow speckle is proposed. According to the characteristics of image morphology, vascular wall tissues are relatively static in fixed position in the image sequences, while blood speckle will shift, which is the phenomenon that the blood speckle noise region is not consistent in gray values. On the basis of the characteristics mentioned above,the process of space and time neighborhood data of pixels extracted from the image sequences involves two-dimensional Fourier transform,from which the energy ratio of high frequency and low frequency can be obtained. Thus, it can be determined that whether signals to be processed is noises or not by combining the energy ratio and a given threshold. When judged as noises, the gray values of the signal and its time neighborhood data are calculated for the similarity weight,which is normalized as a result of the final blood speckle correction value according to the algorithm proposed in this thesis,.Finally,this paper is aimed at the problem that the computation complexity of the improved algorithm is complex and the processing time is slow, the research combines the CUDA parallel programming, which is used to improve the denoising processing time and achieve the purpose of real-time. In addition, the CPU+GPU model is used to design the complex parallel algorithm and process data in large scale. the design of CUDA parallel program involves data transmission and storage, task allocation, design of grid and thread block and optimization of parallel algorithms.