Research On Image Processing Algorithms For Wireless Capsule Endoscopy

Diseases of the Gastrointestinal (GI) tract, such as bleeding, tumor and ulcer, are great threats to human health. The traditional imaging modalities, such as X-ray, barium, CT and double-balloon enteroscopy are invasive to human body. Ultrasound modality, though no invasiveness to body, but suffer from low resolution in images. The gastroscopy and colonoscopy, which make use of fiber optics for light and video transmission, are not only do they cause discomfort, but also these devices fail to diagnose diseases in the small intestine since this region is out of reach by these devices. Double-balloon enteroscopy allows for visualization of the entire small intestine to the terminal ileum, but this procedure require that patients be admitted to hospital for exceed three hours.Wireless Capsule Endoscopy (WCE) is a state-of-the art technology that en-ables imaging of the entire human gastrointestinal tract without invasiveness. It is a pill-shaped device which consists of some miniature components. After swallowed by the patient, it takes pictures of the GI tract and transmits images out of the human body wirelessly. Compared to the traditional methods such as gastroscopy and CT, this new technique can view the entire small intestine without pain, sedation or X-ray radiation. The development of WCE has opened a new chapter in small intestine examination. More than one million PillCam video capsules have helped clinicians evaluate patients for GI disorders since Given Imaging Ltd. produced WCE. Based on WCE of the GI tract, clinicians are now able to detect severe diseases include tumor in early development states. It is recognized as the gold standard method for examining small intestine and make early tumor diagnosis possible.Despite of many significant breakthroughs, there exist one major problem con-cerning the WCE images. The problem is that the viewing process of the video data for the physicians per examination is very time consuming because of the low con-trast image quality and great amount of the video data. This task can only be carried out by two trained clinicians. The duration of this assessment typically varies from45minutes to two hours. If we could use computerized methods to help the physi-cians detect some abnormal regions in the WCE images, it will of course reduce the burden of the physicians. Concentrated on this goal, this dissertation mainly studies some main challenging problems in the development of computer aided diagnosis system for WCE images. To aid clinicians and improve the accuracy of endoscopy capsule endoscopy video diagnosis, we research on WCE images processing methods from three differ-ent levels of digital image processing technology:image processing, image analysis and image recognition, explore the feasibility of a variety of image processing tech-niques used in the WCE computer-aided diagnosis system.For image processing, a novel approach for image enhancement is presented in this thesis. The enhancement result is achieved by combination the means of contrast enhancement, image smoothing and image sharpening techniques into one partial differential equation (PDE) framework. In the image evolution process, three kinds of operation execute at the same time, using the regularization parameters to adjust the proportion of each operation in the evolution process, ultimately achieve the desired results. The simulations for both gray, color and WCE images are given in the experiments. The experimental results show that the proposed framework can simultaneously perform contrast enhancement, image sharpening and smoothing op-eration. Compared with the traditional sequential processing methods, the proposed PDE based framework can achieve better visual effect so as to assist both the inspec-tion and the computer aided detection.Afterwards, we investigate the image analysis methods for the WCE images. First, a mechanism that allows the clinicians to gain certain evaluation of a video without watching the whole video is designed. A shot detection based method is presented for automatically establishing the WCE video static storyboard, and then moving storyboard is extracted based on the selected representative frames under the supervision of clinicians. Experimental results show that most of the representative frames containing relevant features can be extracted from the original WCE video. The proposed method can significantly and safely reduce the number of frames that need to be examined by clinicians and thus speed up the diagnosis procedures.For image recognition, the computer aided diagnosis (CAD) system is designed. In section5, we propose a new method which can detect bleeding regions from WCE video more effectively and efficiently. Since edge pixels and bleeding pixels share similar hue, traditional algorithms often mistake edge pixels for bleeding pixels. We first detect the edge pixels, and then use the morphological dilation to locate and remove the edge regions. Instead of processing each pixel or dividing the image uniformly, we group pixels adaptively based on color and location through super-pixel segmentation. Thus each image can be represented by hundreds of superpix-els and the computational complexity is also reasonable. For each superpixel, the feature is defined using the red ratio in RGB color space. Finally, support vector machine (SVM) is performed to classify the bleeding and non-bleeding superpixels. Experimental results show that the proposed method has low computational com-plexity and maintains high performance as pixel based method. In section6, a novel computer-aided method for tumor detection in WCE video is proposed. In order to improve the detection accuracy, the specular reflectance regions are first removed from WCE image, and mean shift is employed for the initial segmentation of the image. Since tumor may be bleeding or non-bleeding, two schemes are proposed to detect them. After initial segmentation, a red ratio based bleeding detection method is first employed to find the bleeding tumors. Then, an adaptive thresholding algo-rithm is proposed to find the suspected non-bleeding tumor segments. Finally, the true non-bleeding tumor region is picked from them based on geometric analysis. Comparative experiments show that the proposed algorithm is superior to the exist-ing methods in terms of sensitivity and specificity.In conclusion, this thesis investigates some main challenging problems such as WCE images enhancement, video abstraction and computer aided diagnosis, and some novel methods are proposed to solve those problems successfully. Comparative experiments confirm the effectiveness of our proposed methods.