Research And Implementation Of Deep Learning Segmentation Technology For Esophageal Cancer

As a common cancer in life,esophageal cancer has ranked high in its incidence and mortality.The current diagnostic images for esophageal cancer mainly include electronic gastroscopy,ultrasound endoscopy,computed tomography,and nuclear magnetic resonance.doctors rely on experience and professional skills when analyzing medical images,and problems such as time-consuming,labor-intensive,and inaccurate diagnosis are prone to occur in the diagnosis.In order to improve the diagnosis rate of esophageal cancer and protect the life and health of patients,there is an urgent need for an efficient and fast way to assist doctors in diagnosis.Deep learning has achieved good results in the field of image processing.It can effectively identify,classify and extract target regions.Using deep learning to analyze medical images can bring new opportunities to the clinical diagnosis of esophageal cancer.The purpose of this thesis is the research and realization of esophageal cancer segmentation technology based on deep learning.The main research contents of this thesis are as follows:(1)Aiming at the problems of traditional artificial segmentation and extraction of esophageal cancer,this thesis uses deep learning neural network to segment the lesion area.The network uses the encoder-decoder structure as the backbone.In the model,we use a recurrent structure,which can extract feature information multiple times without increasing the model parameters.In order to speed up the convergence of the model and increase the generalization ability of the model,this thesis uses batch normalization layers.We also use the residual structure to avoid network performance degradation.In order to focus on the characteristic information of the esophageal cancer area,this thesis uses the attention module,which can simultaneously focus on channel and spatial information.We also use multi-scale modules to extract feature information of models at different scales.(2)When using neural networks for image segmentation,data processing has a significant impact on the final result.This thesis needs to extract the image information and annotation information from the original medical image file when conducting the experiment.For the obtained image information,window technology is used to increase the contrast between the esophageal tumor area and other tissues.After preparing the experimental data,this thesis uses the proposed network model to conduct experiments,and also compares some classic network models.The experimental results show that compared with U-Net network,the network model used in this thesis respectively improves by 5% and 9% under the Iou and Dice evaluation indicators.(3)This thesis implements an auxiliary diagnosis system for esophageal cancer based on deep learning.The system can use multiple neural network models to segment and extract esophageal cancer.The system is implemented using the Spring Boot framework,which can assist doctors in diagnosis and improve diagnosis efficiency.