Research On Lung Cancer Auxiliary Diagnosis And Treatment System Based On Deep Learning

Radiotherapy is a common method of lung cancer treatment,and the key to the success of radiotherapy depends on whether the organ at risk can be accurately segmented.In recent years,deep learning has achieved certain results in the application of organ segmentation,but there are also some challenges.Because areas such as the edge of the organ at risk are often highly similar to the background,it is difficult to distinguish the boundary,resulting in insufficient segmentation accuracy;and the complexity of the deep learning algorithm is high,and more GPU memory and time are required to train the network model.In view of the above problems,this paper studies the lung automatic segmentation technology of the organ at risk,and combines the segmentation algorithm to develop a lung cancer auxiliary diagnosis and treatment system.The main research work is as follows:(1)To accurately segment the organs at risk in medical images,a relatively lightweight algorithm model is designed.Based on the U-Net algorithm,the DAM-UNet segmentation algorithm is proposed by integrating the dual attention mechanism DAM module.The convolution group structure in the encoding and decoding process of U-Net is modified,and the network is trained by a combination of BCE and DICE loss functions.The DAM module enhances the dependence between image feature information from the two dimensions of space and channel,improves the segmentation accuracy;and modifies the lightweight algorithm model of the convolution group structure.In this paper,several sets of comparative experiments are carried out on the public lung data set and evaluated under multiple evaluation indicators.The results show that the segmentation effect of DAM-UNet is better than other networks in the experiment,and the model calculation amount and parameter amount are small.(2)In order to further improve the accuracy of model segmentation,the RALP-Net segmentation algorithm is proposed based on the optimization of the DAM-UNet algorithm.The encoder-decoder structure is used to complete the endto-end organ segmentation task,and the residual pyramid pooling RSPP module and the asymmetric large convolution kernel ALCK module are added.The RSPP module is responsible for fusing multi-scale feature information in patient images,and the ALCK module is responsible for obtaining larger receptive field feature information.In order to verify the performance of the RALP-Net algorithm,several sets of comparative experiments were carried out on the lung dataset,and the same evaluation indicators as the DAM-UNet experiment were used for evaluation.The results show that the segmentation performance of the RALP-Net algorithm is better than that of the DAM-UNet algorithm and more stable.(3)In order to assist doctors in their daily work,an auxiliary lung cancer diagnosis and treatment system is designed and implemented.This system uses Spring Boot to build the business layer and uses the Flask framework to encapsulate the RALP-Net algorithm written by Pytorch into the algorithm layer.The functional modules include user login and registration module,registration module,auxiliary diagnosis and treatment module,pricing and charging module,medicine taking the module,user center module,and system maintenance module.After the patient image is uploaded to the system,the RALP-Net algorithm is invoked to automatically segment and return to the front end of the system for display to assist relevant experts in subsequent lung cancer diagnosis and treatment research.