Research On Endoscope Assisted Diagnosis Algorithm Based On Deep Learning

The capsule endoscope is an important medical diagnostic tool.Actually,the endoscope is a mini camera,and the doctor diagnoses disease by the images taken.However,there will be some problems when doctor inspections.After patient swallows the capsule,they can’t leave immediately,because the capsule may stay in the stomach,so the doctor should ensure the capsule enter intestine,then the patient can leave,this time may be as long as two hours.At the same time,the doctor needs to monitor the position of capsule and record the time when capsule enters the intestine.Obviously,this is boring work.In addition,the capsule will take tens of thousands of images,and the doctor needs to read each image one by one for diagnosis.The workload is huge,and it is prone to fatigue,which will increase the probability of missed diagnosis and misdiagnosis.Therefore,the realization of the computer to automatically monitor the position of the capsule and the automatic diagnosis of the disease can greatly reduce the workload of the doctor,and is also of positive significance for the timely detection and treatment of the disease.In this paper,we have studied the above issues.Firstly,we propose a capsule location monitoring algorithm which consists of classification neural network and demarcation point determination algorithm.The neural network will classify stomach and intestine in the digestive tract image.The demarcation point determination algorithm is based on the network recognition results,and find the location of the capsule according to the recognition results.At the same time,our capsule position detection algorithm has achieved good experimental results and indicators on the complete capsule examination data of 115 clinical patients.Second,for intestinal hookworm diseases,we propose a hookworm detection algorithm based on YOLOv3.The algorithm consists of a YOLOv3 detection network,a hookworm-interference reclassification network and an error correction algorithm.Our hookworm detection algorithm gets a good experimental result in complete clinical data.Third,we conducted an experimental study on the automatic identification of intestinal bleeding diseases.Similar to the capsule position monitoring algorithm,we followed the same classification network structure and designed a bleeding point determination algorithm.Finally,we achieved good experiment results on clinical data.