Research And Verification Of Pneumothorax Diagnosis Method Based On Deep Learning

We live in the era of big data.Thanks to the continuous improvement of computing power,deep learning is increasingly widely applied.Pneumothorax is one of the most common pulmonary diseases.It is an acute disease in pulmonary department.In serious cases,it endangers the life and health of patients.A large number of practices have proved that the combination of deep learning and medical image processing for computer-aided diagnosis can not only reduce the workload of doctors,but also reduce the misdiagnosis rate of diagnosis.At present,there are few studies on pneumothorax diagnosis in deep learning.Most of the current studies focus on the multi-disease diagnosis of X-ray film,and there are few papers specializing in the diagnosis of pneumothorax,especially those on X-ray film with multiple diseases at the same time.The data sets used in this paper are Chest X-ray14 and Che Xpert.Firstly,the data of these two data sets are analyzed to verify the reliability of the data.A six-layer convolutional neural network is designed to clean the data,remove the bad data,and obtain the standard data capable of deep learning training.There are two major tasks in pneumothorax diagnostic analysis.One of them is the diagnosis of pneumothorax,that is,judging whether it is pneumothorax or not.The other is the extraction of the focal area of pneumothorax,which determines the severity of pneumothorax.In the pneumothorax diagnosis task,the problem of unbalanced data distribution was solved,a network model based on Dense Net was designed,the test set accuracy was 84.9416 and the AUC was 86.0732.In the task of extracting pneumothorax lesions,three network models were designed and compared,the network model using u-net ++ structure had the best effect in the encode process,and the Dice coefficient of test set was 0.8997.