Research On Algorithms Of Segmentation And Extraction Based On Cerebral Hemorrhage CT Images

Image segmentation is the most basic and the most important technology in image processing, computer vision and target recognition; and it is a difficult and challenging tasks in image analysis field as well. As an indispensable branch of image segmentation field, medical image segmentation has played important roles in computer aided diagnosis, the shape statistical analysis and visualization of medical image. The purpose of the segmentation is to change the description of an image of the suspected lesions that are interested in, and make it’s easier and more meaningful to analyze. Unlike the general image, medical image often accompanied by the emergence of the phenomenon of weak boundaries, low contrast and strong noise. It is the medical image’s diversity and particularity that caused the complexity of the segmentation.In recent years, with the rapid development of medical imaging and computer technology, especially the computer tomography technology (Computer Tomography, referred to as CT) have made a significant progress, it demands higher on the accuracy and the speed of image segmentation; and the present algorithms of the image segmentation cannot meet the needs of the current complex segmentation applications any more. The algorithms of image segmentation are developing into a high degree of automation, a better stability and robustness. At the same time, the digital of medical imaging equipment has become a reality. DICOM standard has been used widely, which makes the traditional medical image data from the localized to the networked, and realized the sharing of medical imaging data, which provides the reliable technical support of the establishment of the mechanism of off-site expert consultation and telemedicine.This paper mainly studied on the segmentation algorithm of cerebral hemorrhage CT images and the extraction algorithm of suspected hematoma region. Firstly, it introduces the background and the significance of the cerebral hemorrhage and discusses the advantages of CT images in the diagnosis of cerebral hemorrhage. Secondly, it selectively analyzed the contents of the generation, the content and the information model of the DICOM3.0standard and how it displays in the Windows system. Finally, in the segmentation stage, this paper mainly divides four steps to realize accurately and effectively extract the suspected hematoma region and calculate the volume of the target:(1) In the coarse segmentation stage, this paper uses horizontal direction scanning algorithm, which based on the threshold method, to extract intracranial structure area and remove the extracranial part. Therefore, it can reduce, as much as possible, the interference of the extracranial part of the diagnoses of the doctors; and it largely upgrading the speed and accuracy of the segmentation.(2) In the fine segmentation stage, this paper first analyzes the standard algorithms of FCM, FCM_S, FCM_S1, FCM_EN. Aimed at the disadvantages of those algorithms, we provide modified FCM algorithm after re-designing the filter. We treat the regional boundary points, the regional interior points and noise points differently. We can get rid of the regional boundary points and noise points by setting the filter coefficients. For the regional interior points, according to the different spatial distances between neighborhood pixels and the center pixel, we set different weighting coefficients for the influences of the neighborhood pixels to the center pixel; and we consider these weighting coefficients as the weights and do the weighted sum with the gray value of the neighborhood corresponding pixel, so as to change the gray value of the center pixel. Simulation results show the proposed algorithm has a better segmentation results for the salt and pepper noise and Gaussian noise; meanwhile outstanding the interesting suspected lesions of regional information in the image.(3) Accomplishing the positioning of the suspected lesion region by combining the level set method and the active contour model. Due to the disadvantages of the traditional PC model that not using the image gradient information and the need to periodically re-initialize the level set function, the paper puts forward the modified PC model. On the one hand, we add the energy functional which based on regional gradient information; on the other hand, it integrated with the penalty term of the Li model. The modified PC model improves the accuracy of segmentation, but also increased the speed of segmentation as well.(4) It is quite important to estimate, rapidly and accurately, the volume for many medical diagnoses, treatments and assessments. This paper selects cerebral hemorrhage CT image sequences images, with their admission, which contain suspected hematoma region; and it uses the above three steps to position and recognize the suspected hematoma region in each image; and through the scanning algorithm to extract the suspected hematoma area, and then counts the regional pixels hematoma in each image. The area (S) of the suspected hematoma area is the area of the pixel multiplied by the number of pixels in the hematoma area; and then the volume of layer of images is gained after the area of the suspected hematoma area of each image multiplying the thickness layer of each image. Finally, we stack all the CT layer of images which contain suspected hematoma areas to calculate the volume of the hematoma area. Compared with the traditional manual method, the accuracy of this method has been greatly improved.