In recent years, with constant innovation and development of computer science and medical imaging technology, medical imaging processing and analysis gradually become a new scientific area, and has been changing the traditional medical diagnosis and treatment. It also gives a great impetus to the development of modern medicine. However, human organs or tissues are of diversity, complexity, individual and pathological variability, which is a major challenge in the research field of medical image processing and analysis. On the other hand, there are still inadequacies in medical imaging equipments and technologies, which may result in uneven image signal, low image signal-to-noise ratio, the blurred edges of organs and tissues and so on. These factors make the existing object detection and segmentation algorithms difficult to remove background noise and accurately identify the structures of interest in the image. In addition, physicians and radiologists have to reconstruct three-dimensional shapes of lesions, organs or tissues in their brains by browsing the two-dimensional tomographic slices in the traditional clinical diagnosis, which depends on their professional experience. However, it is difficult to accurately obtain the internal structures of diseased regions, location, size, geometric shape and the spatial relationships between their surrounding organs or tissue. Therefore, the work of this paper concentrates on three aspects: 1. image filtering and object detection based on symmetry; 2. segmentation based on graph theory; 3. medical data visualization based on the segmentation. The contributions of this paper can be summarized as follows: (1) A new retinal blood vessel segmentation algorithm based on radial symmetry. The basic idea of the proposed method is to use the radial symmetry to detect blood vessels in retinal images and suppress noise generated by the non-vessel structures in the filtering process. We first use a Hessian matrix-based multi-scale filtering method to detect possible linear structures and estimate the scales and directions of blood vessels; since the two edges of blood vessels remained parallel and many non-vascular structures were unilateral, then the Canny edge detection algorithm is used to detect edges of gray images and compute the salience of radial symmetry for each pixel; finally the centerline of blood vessels and graph cuts are used to iteratively extract the entire network of blood vessels. The experimental resu...
修改评论