The Research On Segmentation Method Of Polyp Image In Colonoscopy

Colon cancer is a common malignant tumor of the digestive tract.Studies have shown that most colon cancers have evolved from adenomatous polyps.Therefore,it is of clinical significance for early detection of colon polyps.Researchers have found that colonoscopy is the most effective way to detect colon polyps,but some small polyps are missed during the test.We use a computer as a diagnostic aid to segment images of colon polyps to help diagnose.In recent years,with the continuous development of medical assisted diagnosis technology,a variety of image segmentation methods of colonic polyps in colonoscopy have been produced clinically,including traditional colon polyp image segmentation method and image segmentation of colon polyps derived from deep learning method.At the same time,the traditional colon polyp image segmentation method requires clinical extraction of specific features of the patient’s colon,and the accuracy of image segmentation needs to be improved.The current colon polyp image segmentation method based on deep learning technology can specifically solve the problems existing in the traditional colon polyp image segmentation method.However,the method of segmentation based on deep learning also has the problem of large amount of parameters.The overall research goal of this article is that to study the problem of large model parameters in existing deep learning method of colon polyp image segmentation.First,this paper proposed a fully convolutional densenets method for the problem of large model parameters of existing deep learning-based colon polyp image segmentation methods.We accurately segmented the patient’s colon polyp image.Then by studying the methods of full convolutional networks and dense block strategy,and combining the advantages of the two methods.This paper shows that through the fully convolutional densenets method,the end-toend structure of the full convolutional networks can accurately segment the image,and the short connection of the dense block can effectively reduce the colon model parameters,while using the dense block instead of the ordinary as the basic block of full convolutional networks.Through this experiment,this paper compared the amount of model parameters and the accuracy of image segmentation of the fully convolutional densenets method and the deep learning-based colon polyp image segmentation method,which clearly proved that the fully convolutional densenets method can not only guarantee accuracy of image segmentation accuracy,and can also accurately and effectively solve the problem of large amount of model parameters.Secondly,in order to solve the problem of too many connections and large number of calculations based on the fully convolutional densenets,this paper proposed a method of grouping for the fully convolutional densenets to segment colon polyp images.The grouping for fully convolutional densenets method is a further optimization method of the fully convolutional densenets.This paper have studied and learned the method of grouping convolutions and grouping for the fully convolutional densenets,pruning the basic layer of dense blocks in the fully convolutional densenets,and removing the redundant connections of the layers in the dense blocks.Through experiments,this paper compared the method of grouping for fully convolutional densenets and their calculations and segmentation accuracy with fully convolutional densenets,and proved that the method of grouping for fully convolutional densenets not only guarantees the accuracy of segmentation of colon polyp images.Moreover,it also significantly reduces the calculation amount of the model and reduces the complexity of model.In this way,the fully convolutional densenets method is optimized,and a more optimized colon polyp image segmentation model is established.