Research And Application Of Helicobacter Pylori Infection Detection In Endoscopy Based On Deep Mixing Neural Network

Helicobacter pylori(Hp)is a common chronic pathogen,which can cause a series of stomach related diseases,which seriously threaten human health,so the detection of Hp infection and the later diagnosis and treatment are extremely important.For the detection of Hp infection,a relatively common method in clinical practice is to determine whether patients are infected with Hp through the empirical judgment of endoscopists.However,this method mainly relies on the personal qualifications of endoscopists,which is inefficient and easy to misdiagnose.Therefore,the deep learning method was adopted in this paper,and the deep mixed neural network model was constructed to realize the classification of Hp infection in endoscopic images,so as to help clinicians complete a more efficient Hp diagnosis and treatment process.The main work content of this paper includes the following aspects:(1)In view of the problems of redundant information and excessive noise of endoscopic images,the original image data set was preprocessed to screen out the image data that had a great influence on Hp classification.Subsequently,the size of all images was unified through extraction of areas of interest,thresholding and size filling,which was convenient for the subsequent training and testing of Hp classification model.(2)In view of the insufficient interpretability of Hp classification results of a single endoscopic image,a deep hybrid neural network model FP-gMLP was proposed for Hp classification of multiple endoscopic images.Based on the excellent feature map representation ability of the MLP under the big data sample,it can effectively improve the spatial information loss of the feature map.Firstly,gMLP was used to complete the feature extraction of a single endoscopic image,and then the weight of each image was aggregated by Self Attention.The multi-scale feature fusion was completed by FPN,which effectively combined the superficial and deep feature information of the image to further improve the Hp classification performance of the model for multiple endoscopic images and increase the interpretability of its classification results.(3)A deep hybrid neural network model Swin-MLP is proposed to solve the problem of local information loss in feature maps.Local information in endoscopic images will play a key role in the final diagnosis of Hp.By introducing Swin Mlp Block,the acquisition ability of local information in images is improved,and the balance between local information and global information is fully considered,so as to enhance the semantic information fusion from the local to the whole feature map,and achieve the high efficiency of the Hp classification model.Experiments show that the accuracy of this model in Hp classification task is 93.2%,and compared with other MLP models,it runs faster on the test set,which not only reflects the real-time performance of Hp classification,but also ensures a high accuracy of classification.(4)In order to realize the practical application value of the Hp classification algorithm for endoscopic images proposed in this study,an Hp auxiliary diagnosis system was constructed in this paper.The system mainly includes user login,image input,image detection,result display and other modules.By inputting a set of endoscopic images,it can quickly diagnose whether patients are infected with helicobacter pylori and provide reference for clinicians to make follow-up treatment plans.It can play a role in the auxiliary diagnosis and treatment of Hp infection,and has certain application value and practical significance.