Auxiliary Diagnosis System For Pneumonia Disease Based On Chest X-ray Images

Pneumonia disease seriously damages the respiratory function of the human body.Especially in recent years,the outbreak of Corona Virus Disease has led to a rapid increase in the number of confirmed cases.Pneumonia-related diseases are highly valued by all mankind.The diagnosis of pneumonia-related diseases is predominantly manual,but the large number of pathological images poses a huge challenge to medical staff,and the over-intensive workload can easily lead to misdiagnosis.Deep learning technology is now widely used in the medical field,and artificial intelligence based aided diagnosis is recognized by the industry.In this dissertation,the Res Unet segmentation model and the Dense Net121 classification model are investigated separately using X-ray images as the research object,and an auxiliary diagnosis system for pneumonia disease based on chest X-ray images is designed to improve the ability to classify pneumonia disease with an average experimental accuracy of 94.4%,efficiently screening new coronary pneumonia cases and providing a practical and easy-to-operate auxiliary The system is highly efficient in screening Corona Virus Disease cases,providing a practical and easy-to-use aid.The research in this dissertation covers the following points.(1)An image pre-processing method based on lung segmentation is given for improving the feature extraction ability of classification network models.Chest X-ray images often contain some human tissues other than the lungs,and the Res Unet segmentation network is first investigated in order to exclude the interference of this part on the classification results.By adding the Dense Atrous Convolution module to the Res Unet network model,the network model feature fusion capability is enhanced and the segmentation effect is improved.The segmentation network was then combined with an image cutting algorithm to process the lung region in the X-ray image to create the conditions for training the classification network model.(2)An attention mechanism-based classification network model for pneumonia is given for improving the classification accuracy of the classification network model.Some improvements were made to the Dense Net121 network model structure by adding the attention Coordinate Attention module to the network model structure.The attention mechanism module enhanced the information extraction ability of the classification network model,and the improved Dense Net121-CA4 classification network was 1.5%more accurate than the original network model in predicting category The average accuracy of the predicted categories was 94.4%,and the values of other evaluation metrics such as precision,recall and specificity were 95.5%,94.8% and 96.5% respectively.(3)Designed and implemented an auxiliary diagnosis system for pneumonia disease based on chest X-ray images.The system utilizes Py Qt5 tool for the front-end interface design of the system,and combines the improved classification network model to complete each functional module of the system,and finally the system can display the disease classification prediction results visually.Finally,the system meets the design requirements after process commissioning and performance testing.The system can help medical personnel to complete the auxiliary diagnosis of diseases,which has certain practical significance and value.