Automatic Lesion Detection Of Small Bowel In Wireless Capsule Endoscope With Deep Learning

In recent years,with the development of science and technology,wireless capsule endoscope(Wireless capsule Endoscopy,WCE)has been widely used in clinic,and it has gradually become the mainstream technique for doctors to detect intestinal lesions.However,each WCE check produces more than 50,000 images of the data.In order to detect pathological images,doctors need to spend a lot of time viewing image of the work,because of visual fatigue and false detection.Therefore,the study of a set of wireless capsule endoscope lesion Image detection method is a key problem to be solved urgently.At present,in the field of automatic detection of small intestine lesions in wireless capsule endoscope,many researchers adopt traditional machine learning methods,while traditional machine learning methods often require manual extraction,and their methods are tedious and difficult to achieve the desired results.Compared with traditional machine learning,deep learning is widely used in many image processing fields because of its advantages of automatically extracting image features,and it can achieve the desired effect in the final processing result,therefore,based on the depth learning idea,this paper puts forward a method using Convolution Neural Network(CNN)The research on detection and identification of the common changes of small intestine image in wireless capsule endoscope,its main contents are as follows:(1)Location and identification of small intestine starting and ending pointThe aim of the small intestine is to help doctors find the small intestine area accurately and quickly.The paper realizes the extraction and classification of three parts of the small intestine,the anterior segment of the small intestine and the posterior segment of the small intestine by constructing the CNN model.Finally,using the new image data to identify the training classification model,the results show that the average deviation of the starting point and stop point of the small intestine is 0.172%,2.29%,and the identification rate of the locating model is up to 96.84%.(2)Detection and identification of common intestinal lesionsThe research focus of this paper is to realize the detection and identification of four kinds of intestinal diseases such as hemorrhage,erosion,ulcer and polyp.In the study of lesion detection,the author divides it into two kinds of cases: 1 single lesion detection and 2 multiple pathological detection.In the detection of single lesion,four kinds of detection models of hemorrhage,erosion,ulcer and polyp were established respectively,then the corresponding data were validated and identified by the four models respectively,and the corresponding ROC(Receiver Operating Characteristic)curve was drawn and calculates the AUC(Area Under the Curve)value.In the detection of multiple lesions,a five classifications identification model comprising the four lesions and normal images was established,then the model was used for verification and identification,and the final recognition rate was calculated.In this paper,based on convolution neural network method in deep learning,the detection of small intestine lesions was studied and a different model of pathological changes is established.The experiments show that the method based on deep learning has higher detection and recognition ability than traditional machine learning method in WCE of small intestinal lesions,and its recognition rate can reach 98.39%,95.34%,95.16% and 99.82% respectively in the model of bleeding,erosion,ulcer and polyp lesion.It can be seen that in the image detection of WCE small intestine lesions,a better detection and recognition result can be obtained by using the method based on deep learning.In practice,for doctors,the detection of WCE small intestinal lesions based on the method of deep learning is of great value and significance to the diagnosis of patients' condition.