Design And Implementation Of ME-NBI Early Gastric Cancer Image Recognition System Based On Deep Learning

Gastric cancer is one of the most common malignant tumors in the world,and it is the fourth largest cancer in the world,and the total number of deaths has increased year by year.The timely detection and treatment of early gastric cancer is an important measure to reduce the rate of deterioration and mortality of gastric cancer.Therefore,it is urgent to recognize and diagnose early gastric canser lesion timely and accurately.As the main tool for diagnosing early gastric cancer,ME-NBI has been widely used in clinical practice.However,its learning curve is long,the diagnosis is subjective and lacks an objective diagnostic system.Therefore,this project designed and implemented a computer-aided diagnosis system for early gastric cancer recognition and localization of early gastric cancer lesions in ME-NBI stomach image,helping doctors to improve the diagnosis efficiency and accuracy of early gastric cancer,providing a second diagnosis basis for doctors,reducing the false positive rate and missed diagnosis rate of subjective diagnosis,and helping patients get medical treatment and rehabilitation earlier.At present,although the automatic diagnosis of early gastric cancer is important and urgent,there are not many computer-aided diagnosis systems based on ME-NBI images of early gastric cancer,and there are few studies on ME-NBI stomach images.This thesis mainly studies the recognition and computer-aided diagnosis system based on ME-NBI early gastric cancer images.It is mainly divided into four parts: 1)Four pre-training models of Vgg19,ResNet50,Xception and DenseNet121 were selected,and ME-NBI images were used for fine-tuning to achieve the classification of ME-NBI early gastric cancer images,and compared with the traditional manual texture feature extraction methods(LBP and Gabor).The experiment proves that the deep learning method used in this thesis is superior to the traditional texture feature extraction method,and the ResNet50 effect is optimal;2)Using Faster R-CNN,YOLO,SSD depth target detection model to detect lesions in ME-NBI early gastric cancer images,Faster R-CNN is better than the other two methods;3)Using Vanilla Grad-CAM,Deconv Grad-CAM and Guided Grad-CAM deep learning technology realizes the segmentation of ME-NBI early gastric cancer image lesions,and Guided Grad-CAM is better than the other two methods;4)The optimal algorithm of the above three parts were integrated into the early gastric cancer image recognition system,and a computer-aided diagnosis system based on ME-NBI early gastric cancer image recognition was developed and implemented.The system carried out the requirements analysis,system architecture design,sub-function block design,database construction,fusion of stomach image classification algorithm,cancer target detection algorithm and cancer target segmentation algorithm according to the standardized software design flow.Computer-aided diagnosis will help doctors to find lesion images and their lesions faster and better,and improve the efficiency and accuracy of diagnosis of gastric lesions.At the same time,the various functions of the system can also deepen the doctor’s understanding of the ME-NBI stomach image,and help inexperienced doctors to learn the diagnosis based on the ME-NBI stomach image.The computer-aided diagnosis system for early gastric cancer designed by this theme can also add,query and arrange the basic information and pathological information of the patient,so that the doctor can view and call the patient information in real time and make the final diagnosis in time.The computer-aided diagnosis system for early gastric cancer mainly uses the web interface to provide the doctor with the interface,and has been gradually used in hospitals.