Study On Assistant Diagnosis Of Early Gastric Cancer Based On Gastroscope Image

Cancer has long been a major public health problem worldwide.China is a high incidence area of gastric cancer,an average of 20.7 people per 200,000 people suffer from gastric cancer,which as a relatively common type of cancer,because the direct cause of the disease is unknown,and the degree of malignancy is high,coupled with its early symptoms are not obvious,in many cases it is easy to cause misdiagnosis and missed diagnosis,seriously threatening the life safety of residents.In recent years,with the rapid development of artificial intelligence technology,its application in the field of clinical medicine has also made significant progress.Especially in the detection of early gastric cancer,artificial intelligence technology has great research value and social significance.Based on deep learning technology,this paper studies the auxiliary diagnosis method of early gastric cancer,and innovatively proposes an auxiliary diagnosis technology route for early gastric cancer combined with object detection and image semantic segmentation technology,which has important clinical application value for early gastric cancer screening.This paper summarizes the characteristics of object detection and semantic segmentation machine learning models,and discusses their application in gastroscopic image processing to realize auxiliary diagnostic techniques for lesion area detection and semantic segmentation.The experimental results show that the detection and segmentation model proposed in this paper has good results in the auxiliary diagnosis of early gastric cancer,and has high clinical application significance for reducing the workload of doctors and improving the accuracy of detection.Based on deep learning technology combined with gastroscopic images,this paper mainly proposes the following work content:(1)Research on the detection method of early gastric cancer based on the target detection algorithm YOLOv5.This study developed a detection network based on YOLOv5 specifically targeting early gastric cancer lesion regions.First,the attention mechanism is added in the feature extraction stage,which improves the expression ability of early gastric cancer features.Using Bidirectional Feature Pyramid Network(Bi FPN)at the bottleneck of the network can effectively fuse feature information of different scales.And achieve fast processing,and finally replace the C3 in the original neck network with the C3 Ghost module.The experimental results show that the accuracy of the method is 94.8%,the recall rate and m AP value are also better than the original network and other comparison networks.(2)Explored a U-Net network-based early gastric cancer lesion segmentation method,which is based on the results of gastric cancer detection for further lesion segmentation.Inspired by the encoding and decoding structure of the U-Net network,this research work designed an improved U-Net network for segmenting early gastric cancer lesion regions.First,introduce circular convolution to replace the original U-Net network 3×3 convolutional layer,so that the problem of loss of low-level image features can be avoided to a certain extent,and then the residual module is introduced in each layer after circular convolution,and finally The SE channel attention module is introduced,which can adaptively weight the extracted image features to highlight useful features.In the task of using the improved U-Net algorithm to segment early gastric cancer lesions,the Dice similarity coefficient was 72.09%,and other performances were also better than the comparison algorithm,and the segmentation results were more accurate.