Research On Optimization Algorithm And Platform Of Digestive Endoscopy Image Classification Based On CNN

In recent years,digestive endoscopy has been widely used in the diagnosis of gastrointestinal and biliary,and pancreatic diseases as a micro-invasive examination method.As a kind of digestive endoscopy,white light gastroscope is often used in the diagnosis and treatment of the upper gastrointestinal tract.Due to the uneven distribution of medical resources in my country,ordinary white light gastroscope is still used in the majority of hospitals in my country.However,the huge number of patients,insufficient data of endoscopists,and uneven diagnosis levels have led to the development of digestive tract endoscopy in my country that emphasizes quantity rather than quality.In order to improve the quality of digestive endoscopy,the artificial intelligence-assisted diagnosis technology based on deep learning has attracted the attention of doctors and scientific researchers.Due to the large differences in white light gastroscopy images,many influencing factors,and low image quality,there are relatively few studies at home and abroad.At present,the convolutional neural network has achieved good results in the field of digestive endoscopy diagnosis.Aiming at the above problems,this paper is based on the algorithm research and system prototype design of the white-light gastroscopy image for the classification and diagnosis of digestive tract diseases based on the convolutional neural network.The specific work is as follows:1.In view of the large differences in white light gastroscopy images,the existence of a large amount of text,intubation,etc.,In this paper,the YOLOv3 algorithm is used to realize the detection of noise,combined with the median filtering method and the Sym4 wavelet transform algorithm to denoise the image,and at the same time,the image is preprocessed by data enhancement,cropping,and other algorithms to improve the quality of the sample.2.Aiming at the characteristics of white light gastroscopy images,the Res Net network is optimized and a new algorithm Res Net+ model is proposed for classification and recognition of white light gastroscopy images.By moving the Res Net maximum pooling layer,replacing the activation function and other methods,the accuracy,generalization ability,and training speed of the Res Net+ model are improved.By comparing the test results of common gastrointestinal disease classification models,it is found that the overall accuracy of the optimized Res Net model should be increased by 2-5 percentage points.At the same time,most of the convolutional neural networks have achieved good experimental results.Among them,the optimized Res Net has the best experimental effect,and the F1 value of the classification and recognition model is up to 94.87%.3.Based on the research results of this article,a digestive endoscopy image-assisted diagnosis system was designed and developed.The system realizes real-time synchronization and real-time auxiliary diagnosis of digestive endoscopy images by integrating deep learning technology and big data technology.Through real-time auxiliary diagnosis of digestive endoscopy images,it can guide and assist doctors in gastroscopy examinations,and improve doctors’ work efficiency and medical quality.The research results of the thesis verify the feasibility of the convolutional neural network in the classification model of digestive endoscopy.Most of the classification model experiments have achieved good results,providing technical support for the classification and recognition of common digestive tract diseases.Through the design and development of a digestive endoscopy imaging auxiliary diagnosis system,real-time auxiliary diagnosis of digestive endoscopy for patients has been realized,and the clinical value of the model has been fully utilized.