Research On Detection Technology Of Early Gastric Cancer Based On Deep Learning

Gastric cancer is the third leading cause of cancer death in the world.The symptoms of early gastric cancer are not obvious,which is similar to the focus shape of gastric polyps,which is easy to cause misdiagnosis and delay the condition,resulting in the deterioration of cancer.The early detection effect of gastric cancer will play a vital role in the later treatment of patients with gastric cancer.With the development of artificial intelligence,machine learning model in the field of computer vision can be used to assist in the detection of early gastric cancer,Early detection of gastric cancer can reduce the late treatment cost of patients with gastric cancer.Based on the research status of machine learning assisted diagnosis of early gastric cancer,this paper introduces the clinical diagnosis methods of early gastric cancer,puts forward the diagnosis technical route of computer-aided diagnosis of this kind of disease,summarizes the machine learning model for early gastric cancer detection,and studies the classification and segmentation technology of gastric cancer images under gastroscope.In this paper,the deep learning technology and target detection algorithm are applied to the gastroscope image classification and recognition of gastric cancer.The convolution neural network is used to classify the gastroscope image,and the target detection algorithm is used to realize the semantic segmentation of gastric cancer lesions.Experiments show that the model proposed in this paper can achieve good results,can help doctors improve the accuracy of gastric cancer judgment,and has strong practical significance.The main work of this project includes the following two aspects:（1）Based on deep learning training convolutional neural network,the automatic classification of early gastric cancer images is realized.Firstly,vgg-16 and googlenet are used for model training.Then a new model combining vgg-16 and googlenet is proposed.The characteristics of the two models are fused by designing the parallel convolutional neural network of the new model.The experimental training and result analysis are carried out to draw a conclusion.The model further improves the classification performance of gastroscopic images of gastric cancer.（2）Based on the target detection algorithm,the semantic segmentation of gastroscope image of gastric cancer is realized.Firstly,the feature extraction method of Mask R-cnn feature pyramid is used,and then the expansion convolution is combined with cascade RCNN to design a multi-scale feature fusion model.Experiments show that the semantic segmentation effect of this model is better than other target detection algorithms.