Research On Real-time Detection Algorithm Of Colorectal Polyp Based On Electronic Colonoscopy Image

Colorectal cancer is one of the main causes of death in recent years.The incidence rate and mortality rate of colorectal cancer are third and second respectively in all cancers.Timely detection of polyps can effectively prevent colorectal cancer.Colonoscopy is the main tool to detect polyps.Clinically,doctors find polyps by observing the intestinal images taken by colonoscopy.This manual method has high requirements for doctors’ professional knowledge and experience,and has a large workload.It is easy to cause misdiagnosis and missed diagnosis due to subjective factors such as fatigue.It is urgent to use a computer-aided diagnosis system to help doctors improve work efficiency.At present,the colorectal polyp detection algorithm based on deep learning can automatically find polyps through supervised learning,but its performance is affected by the quality of data labeling,and the false positive rate is too high.Therefore,this paper studies the real-time detection algorithm of colorectal polyp based on electronic colonoscopy image.The main research contents are as follows:1.Aiming at the problem that it is difficult to collect and label lesion images and a large number of normal images can not be used effectively,taking polyp images as an example,an automatic polyp detection algorithm based on semi-supervised learning is proposed.Firstly,the algorithm designs the maximum likelihood loss function based on the multi-scale sampling strategy to learn the suspected region from the normal image without lesions to reduce false positives.At the same time,the cosine similarity loss function is used to improve the ability of the model to identify real polyps.Then,the multi-scale spatial attention mechanism is adopted to make the model pay more attention to the polyp lesion region.Finally,the cross-stage local connection mechanism is used to improve the detection efficiency of the model.Experiments show that the accuracy of automatic polyp detection is effectively improved through semi-supervised learning.2.To meet the clinical needs of high-precision and efficient automatic polyp detection,polyps are detected based on video images.Aiming at the low detection accuracy caused by complex environments such as lens jitter,intestinal peristalsis and intestinal fecal water occlusion.A polyp video detection algorithm based on post-processing method is proposed.The time consistency between adjacent frames is used to distinguish noise and polyp targets.Firstly,the Io U(Intersection over Union)between the current frame and the adjacent frame target frame is calculated.When the Io U between the target frame of more than half of the adjacent frames and the current frame is lower than the threshold,it is determined as false positivity.When missing detection occurs in the current frame,considering different situations,the detection frame in adjacent frames is used to recover the target of the current frame.Experiments show that the designed algorithm reduces the problems of missed detection and false detection in video targets,and the detection speed is 31.25fps(frame per second),which meets the requirements of real-time detection.3.To further meet the clinical needs of artificial intelligence assisted surgery,the polyp area can be measured automatically.Four representative segmentation models are studied in this paper.Through experiments,their performance on private and public polyp data sets is analyzed,and the performance differences of these four models in colorectal polyp image detection are discussed.Finally,nn U-Net(no-new-Net)with better generalization performance is selected as the basic framework of polyp segmentation in the future.The semi-supervised method proposed in this paper has a precision rate of 92.0%and a recall rate of 88.8% in the private polyp image data set,and the video detection algorithm has a precision rate of 96.1% and a recall rate of 71.8% in the video data set CVC-Clinic Video DB.The proposed method can better solve the above problems.In addition,this paper uses representative segmentation methods to carry out polyp segmentation experiments on private datasets and public data sets,and discusses the feasibility of accurate detection of polyps.