Graph Theory Based Medical X-ray Image Segmentation

Medical image segmentation is a hotspot and difficulty in image processing. Exact segmentation result is an important guarantee for subsequent processing and clinical diagnosis. However, due to the complexity and diversity of medical image, including low gray contrast, high similarity between different regions, fuzzy boundaries, etc, there is not better solution for medical image segmentation. Graph theory based image segmentation aims at graph and implements segmentation in the field of graph theory, which is regarded as a new and promising solution. This approach has its own characteristics while still has problems needed to be solved. Based on this background, graph theory based segmentation of medical X-ray image is deeply researched in order to assist clinical diagnosis and treatment.Main contributions and innovation are:(1) Research the corresponding relation between the characteristics of image and graph. Summarize the basic framework of graph theory based medical image segmentation; Then, analyze typical characteristics of medical image from gray contrast, smoothness, consistency and gray histogram. Finally, aiming at graph cut, analyze problems and limitations like large amount of calculation, poor distinction to regions with high consistency, etc.(2) According to low gray contrast and fuzzy boundaries of medical X-ray image, two approaches are proposed for segmentation of medical X-ray image:Multiple Normalized Cut and pseudo color based Normalized Cut. Based on global segmentation, Multiple Normalized Cut obtains accurate segmentation result under the control of local threshold. The advantage of this approach is consistent and accurate result, avoiding under-segmentation of image with low contrast. Through adding pseudo color to X-ray image, color information is added to image and thus increases the contrast between object and background. Based on this increased contrast, complete segmentation is obtained under the criteria of Normalized Cut. This approach is suitable to X-ray image with fairly low contrast and artifacts. These two approaches are tested on the application of dental caries of X-ray images. Testing results show the improved approaches are feasible and effective.(3) According to the disadvantage of great data and slow speed of graph cut, Kruskal algorithm is employed to span Minimum Spanning Tree (MST) to implement segmentation of X-ray image. Similarity matrix and cut criteria are designed and segmentation implementation is researched. Then dental caries X-ray images are used to verify the effectiveness of the MST based segmentation algorithm. The result show that MST based segmentation can acquire global information of X-ray image and segmentation result is accurate, stable and robust. Typical advantages of MST based segmentation are high speed and simple data structure which are suitable to real-time image processing with large volume of data.