Research And Application Of Helicobacter Pylori Recognition And Classification Based On Convolutional Neural Network

Helicobacter Pylori(HP)infection can cause a variety of gastric diseases.Its incidence is very high,and the number of infected people is increasing year by year.If HP can be killed and eradicated in time,it can effectively prevent gastric mucosal inflammation and gastric atrophy,and better prevent and control the occurrence of gastric cancer.In clinical practice,helicobacter pylori infection can easily recur,and it is very difficult to treat.Traditional detection methods have the limitations of many external influences,complex detection methods,and low patient compliance.Endoscopists can judge HP infection through gastric endoscopy images based on rich clinical experience.However,it is inefficient to rely solely on doctors to judge the image category,and it also requires rich experience of doctors.In order to solve the above problems,this paper mainly analyzes gastroscopic images,and conducts research on the recognition and classification algorithm of helicobacter pylori based on convolutional neural network.A series of convolutional neural network models based on multi-scale features are constructed.The main research contents of the paper include the following aspects:Firstly,a HP recognition and classification model based on multi-scale features is proposed.The problems such as uneven illumination and specular reflection in the gastroscopic image dataset are analyzed and processed.Then,a multi-scale feature classification network is constructed with three Inception modules and compared with the common convolutional network.Experiments show that the HP recognition and classification model based on multi-scale features performs better.The indicators of the model are F1-Score 83.57%,precision 85.14%,sensitivity 82.05%,specificity88.68%,AUC 0.916.Secondly,a HP recognition and classification model based on multi-scale features and attention is proposed.Aiming at the problem of insufficient feature extraction in multi-scale feature classification network,dilated convolution is introduced to extract multi-scale features of images.The CBAM attention mechanism is added to enhance the model’s ability to learn important features.Through comparative experiments,it is proved that the model can further improve the effect of helicobacter pylori recognition and classification.The indicators of the model are F1-Score 84.77%,precision 84.55%,sensitivity 85%,specificity 88.32%,AUC 0.9281.Finally,this paper adds residual and depthwise separable convolutions based on multi-scale features and attention model.This method effectively solves the problem of model gradient disappearance or gradient explosion,and reduces the amount of parameters and calculation of the model.The effectiveness of the method is proved by experiments.The hyperparameters of the network model are also determined and applied to the helicobacter pylori diagnosis system,which proves the practical application effect of the model.The indicators of the model are F1-Score 87.34%,precision 89.31%,sensitivity 85.45%,specificity 91.91%,AUC 0.9463.The helicobacter pylori recognition and classification network proposed in this paper can recognize and classify helicobacter pylori based on gastroscopic images,which provides a new method for computer-aided diagnosis of common digestive tract endoscopic images.