Research On Colonscopy Polyp Detection Method Based On Deep Learning

In recent years,the incidence and mortality of colon cancer have been increasing,which seriously threatens people’s lives and health.Colon polyps are a preliminary feature of colon cancer,and colonoscopy is the most effective method for the prevention of colon cancer.In recent years,deep learning has stood out in the field of computer vision.Applying deep learning technology to colonoscopy polyps detection and assisting doctors in clinical diagnosis can effectively improve the detection rate of polyps.Therefore,applied deep learning technology to quickly and accurately detect polyps has important theoretical significance and application value for the prevention and treatment of colon cancer.This article focuses on the detection of colonoscopy polyps based on deep learning.By analyzing the characteristics of colonoscopy images,and the problem of insufficient image data,the convolutional neural network structure was designed to achieve efficient and accurate detection of colonoscopy polyps.The main contents are as follows:First,analyzing the characteristics of colonoscopy images and the tissue morphplogy of common polyps.A colonoscopy image dataset for training and testing was established.Clustering algorithm is used to analyze data distrubution,and corresponding preprocessing methods are used for different distriuted data.In additon,data augmentation is used to enlarge the number of images.Secondly,the algorithm of colonoscopy image classification is studied to identify the polyps in the image.Through analysis the characteristics of classical convolutional neural network structures,a new network architecture which is suitable for medical image classification is designed.Aiming at the problem of less medical image data samples,transfer learning algorithm is applied in the classification system,and some optimized methods are used to improve the performance.The experimental results show that compared with the existing convolutional neural network,the network structure used in this paper can effectively extract image features and reduce the over-fitting phenomenon,and achieve accurate and efficient recognition of colon polyps.Finally,the detection algorithm of colon polyps is studied.The three semantic segmentation algorithms of FCN,U-Net and Deep Lab were analyzed in detail,and various experiments were carried out to test the colon polyp detection,and the results were analyzed.Aiming at the characteristics of small number of polyp images and simple semantic features,a semantic segmentation model based on encode-decoder structure is designed to improve the segmentation effect.The test set is used to test the model,and the experimental results are compared with the results of FCN and other segmentation algorithms.The results show that the segmentation model used in this paper has better segmentation result.In this paper,a colonoscopy polyp detection algorithm based on deep learning is designed,and a high-quality colonoscopy image dataset is established to achieve efficient and accurate detection of polyps.Aiming at the insufficient number of colonoscopy images and simple semantic features,a series of optimization algorithms have been adopted,which greatly improves the detection performance of the algorithm,and has great research significance and clinical applicability.