Study Of Segmentation Algorithms On Cardiac CT Sequence Images

Cardiovascular disease is one of the main causes of death in the world. As the main organ of the cardiovascular system, accurate segmentation of the heart is the key to diagnosis of cardiovascular disease. Domestic and overseas scholars have put forward many segmentation algorithms on the anatomical structures of the heart, such as the ventricular and atrial. However, the segmentation of the full heart is need in computer aided diagnosis. The segmentation of the full heart has some certain difficulty. First of all, the internal structure of the heart is very complicated, including ventricular atrial、 blood vessels、coronary artery; And it is difficult to locate the heart in the CT slices. Therefore, the segmentation of the full heart has always been a challenging problem in the field of image segmentation.Although complicated structure and difficult to locate, heart CT sequence images have highly correlation in profile between adjacent slices. Combining with traditional medical image segmentation algorithms and correlation between adjacent slices, this paper makes some analysis and study on the full heart and proposes two improved segmentation algorithms, which based on traditional region growing algorithm and edge detection algorithm, and compares the segmentation results of these two segmentation algorithms.The main contents and innovations are as follow:1. Display and processing of DICOM images. Introduce the specific content of DICOM format, and propose the implementation scheme of dividing DICOM format into the BMP format, which is in common use under the Windows platform for subsequent manual interaction. At the same time, combine with ITK and VTK toolkit to read and display the heart images of DICOM format, and in this way, we can adjust the window width and window level with the interaction provided by VTK toolkit.2. Implementation of automatic segmentation algorithms for heart sequence images. According to the correlation of adjacent slices in CT sequence images, using the segmentation result of the last slice as initial contour of the next slice, combining with Gaussian filter、morphological operation and other image processing operations, this paper improves the traditional region growing algorithm and traditional edge detection, applies them into the segmentation of the full heart CT sequence images, and makes analysis and comparison of the segmentation results with the gold standard (the manual segmentation results of the experts).3. Implementation of manual segmentation. Automatic segmentation can only be implemented in part of heart sequence images, and manual segmentation is need for the segmentation of subsequent images to complete the segmentation task. Draw the outline of the target area manually, and extract the pixels in the contour to achieve the manual segmentation of the target area.4. Surface rendering of the segmentation results to reconstruct the 3D structure.