Research And Application Of Helicobacter Pylori Endoscopic Image Classification Based On Deep Learning

As an important factor causing gastric diseases,Helicobacter pylori(Hp)is very harmful to human body,so it is very important to detect and treat Helicobacter pylori effectively.The endoscopy is an important method for judging Hp infection.This method is non-invasive to the subject,but the inspection results depend on the experience and ability of the doctor,which is inefficient and costly.Using deep learning technology to extract important features of images,and then identify and classify them,can achieve high accuracy and efficiency.Therefore,on the basis of endoscopy,this paper uses deep learning technology to process and identify images to determine whether the subject is infected with Hp.The main work of the paper includes the following three aspects:(1)The classification algorithm of gastric anatomical parts based on attention mechanism is studied.The situation of Hp infection in different parts of the stomach is different,so the image classification of physiological and anatomical structures is the precondition for Hp detection.Aiming at the fact that the convolutional neural network has too many model parameters in the field of medical image classification and does not make full use of the interactive information between different channels of the image,this paper proposes a Multi-Channel Attention(MCA)module.MCA adds a branch based on the Efficient Channel Attention(ECA)module.The new branch uses maximum pooling and adaptive convolution kernel to extract the features of different channels of the image as a supplement to the original branch feature information.Applying MCA to the existing classification network,experiments show that the new network not only reduces the number of parameters,but also improves the classification accuracy.(2)The gastric Hp classification algorithm based on improved residual network is studied.Aiming at the disadvantage that the feature representation difference between Hp negative images and Hp positive images is small,sufficient discriminative features cannot be obtained,and it is not conducive to the accurate classification of a single image,this paper designs a new residual network module.This module introduces a variety of optimized structures on the basis of Res Net,and builds a single-view and multi-view Hp classification network respectively,and divides the endoscopic images into Hp negative and Hp positive.At the same time,this paper also uses Grad-CAM technology to visualize the features extracted by the classification network,which improves the interpretability of the model prediction results.(3)On the basis of algorithm research and improvement,an endoscopic image diagnosis and recognition system is realized.The system integrates a series of steps required for the diagnosis of Hp,and can obtain more accurate Hp infection results in a short period of time,which can assist endoscopists in making a diagnosis and improve the diagnosis efficiency.This paper analyzes the shortcomings of the deep learning model and makes improvements,applies the improved model to the classification of endoscopic images and builds a classification and diagnosis system,which provides an auxiliary diagnostic method for the detection of clinical Hp.