Research On Segmentation Of Pulmonary Embolism Image Based On Hybrid Domain Attention Mechanism

Pulmonary embolism is a common lung disease in today's life.It is caused by emboli in the pulmonary arteries that block blood vessels,which causes blood circulation disorders in the lungs.This disease usually causes chest tightness and chest pain in some cases,and it is life-threatening in some cases.In recent years,the number of people with pulmonary embolism has been increasing year by year worldwide,and the rate of mortality for this disease is high.The reason for the occurrence partly because early pulmonary embolism is not easy to detect and treat in time,and partly because the disease has a relatively high misdiagnosis rate.The main diagnosis method of pulmonary embolism is pulmonary arterial CT angiography,which can clearly image the pulmonary arteries in a computer and help doctors diagnose the disease.In the current background of high incidence of pulmonary embolism diseases,doctors need to spend a lot of time and energy to diagnose.In order to alleviate this situation,this article hopes to study a feasible method to assist doctors in diagnosis of pulmonary embolism.In recent years,the use of deep learning to assist doctors' diagnosis has gradually become a trend,especially the application of convolutional neural networks in deep learning has become more widespread.Convolutional neural networks can effectively extract features from images,and quickly and accurately predict information such as the type and location of targets.In the field of medical images,image segmentation is very necessary,because in the process of disease diagnosis,it is not only necessary to judge its type,but also to quantify the area of the disease,and perform pixel-by-pixel segmentation and calculation.There are many classic network models for image segmentation in convolutional neural networks.Such network models often extract the abstract features of the target first,then use the combination of deep information and shallow information to gradually recover the features,and finally output the model prediction as segmentation results.At present,many works have applied some segmentation models to a variety of medical datasets.After experiments,many models can reach the diagnostic level of doctors.Therefore,using convolutional neural network to segment pulmonary embolism is a very meaningful research work.In this paper,an in-depth study of the pulmonary embolism disease and the image segmentation model in the convolutional neural network is proposed.Based on the open source pulmonary embolism dataset,a hybrid domain attention mechanism of pulmonary embolism segmentation method is proposed.The mixed domain attention mechanism refers to adding attention mechanisms to the spatial domain and the channel domain respectively.This mechanism makes the model pay more attention to the important information in the space and the channel during the learning process,and ignore and suppress irrelevant information.The realization of this attention mechanism is to use the scSE module to continuously adjust the weight of the input features.In order to verify the effectiveness of scSE in this experiment,U-Net was selected as the basic network,and scSE was added to improve U-Net.The experimental results showed that the improved network achieved a Dice value of 0.837 on the pulmonary embolism dataset,compared with the result of base Network is improved by 3.4%,which proves that the mixed domain attention mechanism can effectively improve the accuracy of the basic network for segmentation of pulmonary embolism.In order to verify the rationality of the improved model studied in this paper,a series of additional comparative experiments were added to the article,and a variety of network variants with attention mechanism based on UNet were designed: cSE-U-Net,sSE-U-net,scSE-U-Net(encoding path)and scSE-U-Net(decoding path),these networks have improved the basic network to varying degrees,but the results are still lower than the model studied in this paper,further verifying the scientific and effective of our proposed model for segmentation of pulmonary embolism.