Research And Application Of Bleeding Detection Algorithm Based On Wireless Capsule Endoscope Images

Wireless capsule endoscopy as a detection tool for digestive tract disease,no trauma,no pain,no blind spots,safe and hygienic,easy to operate,no effect to the daily life of patients,overcoming that the traditional plug-in endoscopy is not comprehensive and bring pain to patients.Wireless capsule endoscopy has become a hot spot in medical equipment research.Wireless capsule endoscope presents gastrointestinal internal situation through color photos,each patient will produce 4-6 million pictures,we need professional medical staff to check these pictures to determine whether the digestive tract has lesions.This is a hug test to the medical staff’s attention and time.The problem also limits the widely use of capsule endoscopy.So we need computer intelligence technology to make a preliminary prediction of the lesion range.This article is mainly for the gastrointestinal bleeding.Gastrointestinal bleeding is a sign of many malignant diseases,so we need to find it as early as possible and take measures.The core research of this paper is image preprocessing and color-based endoscopic image bleeding detection.A lot of regions of wireless capsule endoscopy images will be useless because of lack of light,digestive juice shelter,food residue shelter,bubble shelter and other reasons.It will have effect on subsequent bleeding detection.We find that the hue H and saturation S of the invalid area are different from the valid area in the hsv color space.By limiting the three color channel range of hsv color space can delete the dark invalid area,the area where the cloudy digestion juice and the food residue are blocked.Most bleeding area of obvious bleeding images is less than 1/3 of the normal area,Most bleeding area of not obvious bleeding images is less than 1/5 of the normal area,and color characteristics are mostly statistical features,a large normal areas will reduce the sensitivity of bleeding detection.Super-pixel segmentation can effectively extract the bleeding area,but it increases the amount of computation while increasing the accuracy.It is found that the saturation of the bleeding area is higher than that of the normal region,and the ratio of the bleeding area can be effectively improved for the obvious bleeding images.Therefore,this paper improves the relative area of the bleeding region by extracting the high saturation region and the super pixel segmentation.The loss of effective area in the process of removing the invalid region and extracting the high saturation region is analyzed.At the stage of image bleeding detection,the color histogram features,color moment features and red purity features of the bleeding area and non-bleeding area are analyzed,and the features are quantified according to the different range of the two area.In the pretreatment stage,a part of the effective region was lost.We analyze the effect of the relative area of the bleeding region on bleeding detection by experiment.The applicable scenarios of the naive Bayesian classifier,the random forest classifier and the support vector machine classifier were analyzed.And we select the appropriate features and classifiers for the obvious bleeding and no obvious bleeding images at the picture level and the super pixel level.The algorithm is applied to the patient’s data set and the real-time performance of the algorithm is analyzed.Results Optimize is mainly based on the time sequence of pictures and the continuity of gastrointestinal bleeding.The secondary processing is based on the results of the previous image classification,The idea of a secondary processing is that if the picture around a normal picture is mostly bleeding,then the picture may be labeled wrongly,then its label is changed to bleeding;In turn,if the picture around a bleeding picture is mostly normal,then the picture may be misjudged,then its label is changed to normal.