Research On Lung Image Segmentation Algorithm Based On Deep Learning

The lung is one of the most important organs in the human body,but it is also one of the most prone to disease.Pneumonia,lung cancer,emphysema,and chronic obstructive pulmonary disease(COPD)are all common lung diseases.When clinicians examine and diagnose lung diseases,they usually need to combine medical images of the lungs,especially CT images of the lungs generated by computed tomography(CT)technology.However,doctors usually only pay attention to the lung parenchyma in lung images,so lung parenchyma segmentation is almost the first step in medical image processing of all lung-related diseases.Accurate lung parenchyma segmentation can help doctors analyze and diagnose lung-related diseases more accurately,which is of great significance.Therefore,this paper mainly conducts research and innovation around lung parenchyma segmentation.The specific work content is as follows:(1)Aiming at the problems of high complexity,large amount of calculation and many parameters in the lung parenchyma segmentation method based on deep learning,a lung parenchyma segmentation method based on LUNet network is proposed for automatic segmentation of lung parenchyma images.Among them,the LUNet network is improved and innovated on the basis of the UNet network.First of all,by changing the number of layers and channels of the original UNet network,it is used to improve the operation speed of the network and reduce the number of parameters of the network;secondly,the pre-activated residual module is introduced to deepen the network without causing gradient dispersion or gradient explosion.The depth of the network improves the performance of network segmentation;again,a multi-scale context module is introduced to obtain rich context information and strengthen the connection between information.Finally,the model is trained in combination with the designed hybrid loss function.The experimental results show that the segmentation performance of the LUNet network is better than other segmentation networks,and the amount of parameters and calculations is small.(2)Aiming at the problems of insufficient sample data and inability to share data among medical institutions during deep learning model training,a lung parenchyma segmentation method FL-LUNet based on horizontal federated learning was proposed.First of all,FL-LUNet adopts the method of horizontal federated learning,which can jointly train deep learning models with different medical institutions without revealing user privacy.Secondly,localized differential privacy technology is introduced to perturb the parameters before the medical institutions upload the local model parameters to protect the model parameters and user attributes from being stolen by malicious attackers.The experimental results show that although FL-LUNet will lose certain accuracy,it can jointly train the deep learning model with multiple medical institutions without sharing data,which not only protects the privacy of users,but also promotes the integration and utilization of medical data.