Research On Computer-aided Diagnosis Algorithm Of Gastrointestinal Diseases Based On Deep Learning

According to the World Health Organization,the number of deaths due to gastrointestinal diseases accounts for about 10.9%of the total number of deaths in the world each year.Due to the large number of types of gastrointestinal diseases and the similar appearance characteristics among disease severity grades,it is a laborious and error-prone task for endoscopists to accurately identify the types and severity of gastrointestinal diseases from a large number of endoscopic images.Although many studies have proposed computer-aided diagnosis(CAD)models to assist endoscopists in the diagnosis of gastrointestinal diseases,most of these systems are usually limited to a single lesion or a specific organ.Moreover,there is a lack of studies that rely on a small amount of labeled data to achieve grading diagnosis of disease severity.Therefore,in view of the above problems,this paper carried out the following research:(1)A CAD model of various gastrointestinal diseases based on multi-scale feature fusion is proposed.The global feature extraction path and local feature extraction path are constructed based on transformer and convolutional neural network structure,respectively.In the global feature extraction path,the convolution operation is used to achieve downsampling to obtain multi-scale global features,which further enrichments the feature representation while reducing the computational cost and memory consumption of the model.In addition,a channel and spatial attention module with a small number of parameters are proposed in the model,which enhances the model’s attention to the target region and reduces the loss of important information during the change of spatial dimensions.Finally,in the feature fusion module of the model,the attention module is used to effectively fuse the extracted multi-scale global features and local features.Through extensive experiments,the performance of the proposed model is not only better than other advanced models,but also better than experienced endoscopists.It can effectively assist endoscopists to achieve accurate diagnosis of various gastrointestinal diseases.(2)A CAD model of disease severity grading based on few-shot fine-grained learning is proposed.The model is composed of an embedding module and a metric module.The embedding module gradually focuses on the input target region,extracts rich multi-scale features,and fuses them to generate richer feature representations,which are used as the input of the metric module.In the metric module,the bi-similarity metric module is used to map the feature information into a more compact metric space,so that the model can pay attention to more minor feature differences on the basis of a small amount of labeled data.Extensive experiments on five few-shot fine-grained datasets show that the proposed model has excellent generalization and strong adaptability.It can achieve fine-grained grading diagnosis of disease severity with only a small amount of labeled data and no additional manual annotation.It has a broad development prospect in the field of computer-aided grading diagnosis of the severity of gastrointestinal diseases.(3)A CAD system of gastrointestinal diseases based on PyQt5 for classification and severity grading of gastrointestinal diseases is designed and implemented.With the joint efforts of this system and three experienced endoscopists,a gastrointestinal disease dataset with more categories and labeled samples is constructed,which can promote the development of CAD technology in the field of gastrointestinal diseases to a certain extent.