Study On Interactive Medical X-ray Image Segmentation Based On Graph Cuts Algorithm

Medical Image segmentation is a hot spot in digital image processing and analysis. Accurate segmentation of medical X-ray images is the guarantee for modern medicine in several fields such as clinical diagnosis and optimal option of treatment. In recent years, with the development of medical imaging technology, medical images are provided with higher resolution and higher complexity in details. Due to the uneven distribution of gray levels, poor gray contrast and complex textural features, automated image segmentation technology has its limitations. However, the interactive image segmentation based on Graph Cuts can exploit users’input information and come out a segmentation of ROI fast yet accurately. Previous work shows that Graph Cuts based segmenting methods have prone to some drawbacks when applied to medical images.With the research in related work about medical image segmentation, this paper discusses the graph theory based image segmentation methods, introduces a framework of interactive image segmentation based on Graph Cuts. Main contributions are:Firstly, this paper presents the state-of-art research about graph theory based image segmentation, introduces the framework of Graph Cuts based interactive image segmentation. This paper also discusses the drawbacks of the medical image segmentation due to the low gray contrast and fuzzy boundaries. These shortcomings include limitations in robustness and poor distinction to regions with high consistency.Secondly, according to the low contrast and shadows in tissue boundaries, this paper designs two approaches for segmenting medical X-ray images:Introducing shape prior into Graph Cuts theory and pseudo color based Graph Cuts method. With the shape prior and user’s interactive inputs, the first approach is able to solve the insufficient segmentation of tissue boundaries. And by introducing pseudo color technology and K-means Clustering method, the segmentation based on Graph Cuts results more accurately. Experiments and analysis are conducted with mammography images of Peking University Peoples’ Hospital, and the results demonstrate that the designed approaches are effective and feasible.Thirdly, this paper implements a breast X-ray image processing system. With the analysis of user’s demand, the system is implemented under the Windows7operating system and the platform of QT5, besides that this system is implemented with the OpenCV library. The system includes functions of image pre-processing, image segmentation and patient’s file management.This paper designed two refined approaches about the classic graph cuts based segmenting method. Experiments on mammography images demonstrate that the segmenting results can provide doctors with a good reference for further diagnosis.