Ultrasound Image Filtering And DSA Motion Artifact Elimination

Compared with general image processing, medical image processing has distinct features. Much data should be processed, detail highly demanded, and algorithm real-time. Besides, evaluation of image quality needs doctors'experience. One the one hand, we only use existing empirical models or mathematical distributions to analyze whether the point in an image is noise or not. On the other hand, the operation between the images, such as the DSA image sequence subtracting, often introduce new noise and result in motion artifacts.This paper is to address how to eliminate noise in single image and the artifact between image sequence. Compared with other medical imaging modality, speckle noise in ultrasound images is very common and difficult to eliminate. Speckle filtering is of great importance because the dynamic characteristics of ultrasound imaging is too excellent to be replaced by other imaging. There are two ways for speckle noise elimination: image averaging and image filtering. [05] The former is simple but loses lots of spatial information and reduces the spatial resolution. In this paper, we start from an algorithm based on anisotropic diffusion equation, summarize a variety of improvements for this algorithm and at last put forward an improved algorithm. In addition, we use GPU's FBO (Frame Buffer Object) for parallel processing to solve the huge computation caused by many templates'traversal when discreting the diffusion equation.The second theme of this article is to eliminate the motion artifacts in DSA. For the dynamic DSA sequence (can be extended to other image sequence) subtraction, motion artifacts will easily occur in some edges of tumor because there is no adequate alignment. New improved three-step method in motion estimation is used to calculate the offset between the images. Moreover, we combines absolute offset and relative offset, which has different weights in different intervals, to gain a better performance to eliminate the motion artifacts. Throughout the algorithm implementation, the number of comparisons is greatly reduced by the introduction of reference image. Ideally the number of comparisons between images is just one more than that of mask images. In order to make the result more accurate, hash table mapping is done on contrast image, which result in a slight increase in the number of comparisons. One extra comparison for the relative offset between target image and the first image in Slot has to be done. This algorithm is also designed combining statistical data and image features. Motion displacement algorithm just focus on the statistical data. Diaphragm is a very obvious feature in our chest images especially its height, so we use this feature to confirm our result. Apart from the above algorithm designed to reduce computation, we also selected a region centered in diaphragmatic as our target object for each DSA image. In the ways above, the data to be processed is greatly reduced.The method about elimination of noise and artifacts can be used in the clinical condition to aid doctors in diagnosing some disease.