The Research And Application On Unsupervised Redundant Image Deletion For Wireless Capsule Endoscopy Examination

Wireless capsule endoscopy (WCE) is a new type of noninvasive tool for small bowel disease diagnosis. Examination including swallowing WCE, checked the entire gastrointestinal tract with gastrointestinal motility after taking photos, and image data were taken by the receiver that the patient carried around, and ultimately were downloaded to PC for doctor to observe and make a diagnosis. This process takes about eight hours and involves about50,000color images of the gastrointestinal tract. The higher rate of misdiagnosis with excessive work time and low application efficiency of this approach are serious impact on the development and popularization of the capsule endoscope. Therefore, it is very necessary to develop a new approach to reduce the intensity of diagnostic work and to ensure a higher diagnostic rate through computer technology.This research is carried out under the auspices of the Guangdong Province Science and Technology Project (NO.2007B031302008, NO.2009B010800019) and Guangdong Province, the Ministry of Education Combination Project (NO.2008B090500200, NO.2010B090400543) and focused on the use of normalized mutual information and the normalized cross-correlation coefficient as the similarity coefficient of images to delete the redundant images, so that the optimal way of deleting unsupervised redundant image for wireless capsule endoscopy examination will be obtained. Meanwhile, this study also includes a feature extraction method to get the feature of the similarity coefficient with capsule endoscope image. Then the capsule endoscope image-assisted analysis system is designed on the basis of this study, to achieve threshold deletion by using the normal distribution characteristics and the iterative algorithm in the image deletion. In addition, backtracking method is used to optimize the automatic deletion method instead of one by one deletion rules, so as to reduce quantity of the WCE images, improving the purpose of speed. The following is the content of this article:1.The analysis of capsule endoscopy image characteristics and extraction of color images feature(chapter3)Based on the data analysis of several cases, the capsule endoscopy image presents three main characteristics:large amount of image data, high redundancy with invalid and similar image, lower proportion of the diagnostic images. At the same time, this article realizes the color conversion of capsule endoscopy color images from RGB color space to HSV color model, and uses non-equigap quantification to abstract one-dimensional vector as image feature according to the characteristics of the capsule endoscopy image, which is used in the process of automatic redundancy deletion.2.Proposed a method based on the normalized mutual information redundancy images automatic deletion(chapter4)First of all, make the normalized mutual information and cross-correlation coefficient as the image similarity coefficient, and do normality analysis, then carry out redundant image deletion by the distribution characteristics combined with the affine transformation. Meanwhile, according to clinical needs, iterative method is used to achieve the fixed image deletion and performance analysis. This paper first proposed the evaluation criteria of image accidental deletion rate of retention rate as the algorithm through experimental data analysis to determine the normalized mutual information for optimal image similarity coefficient.3.The design of capsule endoscopy image aided analysis system(chapter5)In this chapter, capsule endoscopy image aided analysis system is built up. Firstly, the basic functions of the basic information editing and mass data import are available after setting up the system database. Then, the batch image browsing and smart player had come true, which can tag a key image during process of images automatically play with the diagnostic significance of the image selected by the user to record. The key images are saved to the image tag bar, in order to be read at any time. The third step is to achieve the level deletion and fixed deletion of redundant images, so that display the deletion results to assist clinical diagnosis. Furthermore, the function of displaying original image data is provided to make sure the reliability of the deletion results. Finally, according to the clinical needs, the system realized case report editing functions4.Combining optimization method redundant back images automatic deletion process(chapter6)The redundant images automatic deletion method is optimized by the basic theory of backtracking method, so as to reduce the number of image comparison effectively, to improve the efficiency of the algorithm on the basis of insurance of the lesion retention.