Classification Network Of Benign And Malignant Gastric Ulcers Based On Deep Learning

The early diagnosis and treatment of malignant gastric ulcer under common gastrointestinal endoscopy and the accurate judgment of benign gastric ulcer are directly related to the safety of patients’ lives.However,there are many similarities between the clinical manifestations and the appearance of common digestive tract endoscopes,which can easily cause misdiagnosis and missed diagnosis.This article focuses on the research of common digestive endoscope images,and uses digital image processing methods such as Sobel operator and HSI color space conversion to address the problem of poor imaging quality.This research proposes a network that combines improved Non-local attention mechanism and Dense module and deep separation and convolution of benign and malignant classification of gastric ulcers,and conducts in-depth research on the feature extraction and learning capabilities of the model.In addition,this paper also addresses the problem of uneven distribution of the number of benign and malignant gastric ulcer samples and the problem of excessive network parameters and calculations after the introduction of the Non-local attention module.The research content of this article mainly includes the following points:(1)Investigate and summarize the current classification methods and research status of gastric ulcer lesions under endoscopy,focusing on the application of deep convolutional neural network in the classification of benign and malignant gastric ulcers under general gastrointestinal endoscopy,and introducing its principle and basic structure.(2)This paper uses Dense Net as the basic model,and analyzes the impact of data set sampling strategies,preprocessing methods and data set enhancement methods on the performance of convolutional neural networks,and through ablation experiments analysis screens out a combination of preprocessing methods with better data sets.This article also conducts in-depth research on the problem of uneven distribution between sample categories of the gastric ulcer dataset under the general digestive endoscope.(3)Summarizes the recent development and main research ideas of convolutional neural networks based on attention mechanisms in the field of computer vision in deep learning,and selects the non-local attention mechanism designed for global long-range context-dependent semantic information and the Dense Net network was combined to improve its effectiveness in the classification of benign and malignant gastric ulcers.(4)This paper proposes a classification network for benign and malignant gastric ulcers combining deep separable convolution,Non-local attention module and Dense module.In the actual design process of the model,some simplified network structures were applied to accelerate the network training strategy to optimize the network model,and achieved good results in the classification of benign and malignant gastric ulcers.The index of the model is F1 score is 94.50%,accuracy rate is 96.57%,sensitivity is 92.51%,specificity is 90.89%.The classification network of benign and malignant gastric ulcers proposed in this paper can be used to classify benign and malignant gastric ulcer lesions based on common gastrointestinal endoscopy images,which provides new ideas for the computer-aided diagnosis of common gastrointestinal endoscopy images,and has important theoretical research significance and clinical application.value.