Algorithm Research Of Bleeding Lesions Detection Based On The Capsule Endoscope Image

Nowadays, gastrointestinal diseases have become a major threat to human health. Mechanical endoscope as a traditional test method of gastrointestinal diseases, not only the operation is inconvenient, but also brings about physical pain to the patient. With the development of semiconductor technology, wireless communication technology and integrated circuit technology, the capsule endoscope appeared, and soon has made a unique advantage in the detection of gastrointestinal disease. However, the number of the digestive tract images generated by the capsule endoscopy is up to tens of thousands, if the medical staff to view one by one, which not only brings a heavy burden to the medical staff, but also increases the misdiagnosis rate.In order to solve these problems,in this study, on the basis of the introduction of present situation of the capsule endoscopy and the bleeding detection technology based on the capsule endoscopy image,i conduct in-depth research on technologies of the bleeding lesion detection based on the capsule endoscopy image. The main work includes:image preprocessing,extraction of interested area based on color feature, classification of images based on color similarity and connected domain area.Implemented the algorithm of bleeding lesions detection based on the capsule endoscope image.The bleeding detection algorithm based on the capsule endoscope image mainly carries out with two traditional methods. One is the image area is divided into fixed size, which will destroy the mechanical division boundary information contained in the image itself, and the resulting accuracy is not high. The other one carries out template operation on the whole image. This algorithm can maximally reflect the information of the original image. However, due to the large amount of data, resulting in detection algorithm is too slow. In summary, taking the original boundary information of the image and the speed of detection algorithms into account, this study, firstly, Extract the interested area and reduces the redundant information of image by using the color boundary box in RGB vector space.Then, the regions of interest are classified based on the color similarity and connected domain area, which can improve the detection speed in the premise of ensuring the accuracy of detection. The results showed that the sensitivity of the algorithm reached 91% and the specificity reached 88%, which basically achieved the automatic detection of bleeding lesions in the capsule endoscopy image and this can be applied to the practice of treatment.