Research And Application Of Chest X-ray Classification And Pneumonia Detection Based On Deep Learning

Facing the rapid growth of medical imaging data,but the serious shortage of radiologists,seeking for fast and accurate intelligent diagnostic methods has been the direction of unremitting efforts of domestic and foreign researchers.At the same time,the deep learning algorithm based on convolution neural network has exceeded the traditional algorithms in image classification,detection and segmentation,and has been widely used in autonomous driving,robot vision and other fields.Therefore,introducing deep learning technology into the medical field to study an efficient and accurate intelligent way to interpret medical images can not only assist doctors in diagnosis,reduce missed diagnosis and misdiagnosis,but also improve the efficiency of disease diagnosis and make up for the gap in medical resources,which has theoretical research value and clear application prospects.In this paper,based on deep learning algorithm to study the classification and detection of pneumonia on chest X-ray of medical imaging.The main work is as follows:1.The idea of model-based transfer learning is used to improve the classification effect of deep learning models on chest X-rays.In order to alleviate the shortage of chest X-ray samples and the imbalance of categories,a model pre-training on the Image Net data set is used to initialize the weights of the convolutional neural network,and the data set is expanded through data enhancement and oversampling is used to balance the positive and negative samples.A large number of experiments show that,in the case of insufficient samples,the pre-training model parameters of natural images can help to improve the classification effect of the network on chest X-rays.2.A chest X-ray pneumonia detection algorithm based on improved FSSD is proposed.This algorithm uses Rest Net50 as the feature extraction network,and then uses the learnable transposed convolution to sample the feature map.Compared with the manual bilinear interpolation method,it can obtain more accurate feature information.Finally,concatenate channels are used for feature fusion.In addition,during the training process,the pre-training model Rest Net50~*of transfer learning is used to initialize the network weights.Experimental results show that the improved algorithm effectively improves the accuracy of pneumonia detection.3.A chest X-ray pneumonia detection algorithm based on the improved Center Net is proposed.This algorithm uses Dense Net121 to extract object feature information,and at the same time draws on the idea of FPN algorithm,combines shallow and deformable convolutional up-sampling deep features to obtain richer position and detailed information of the object.In addition,during the training process,the pre-training model Dense Net121~*of transfer learning is used to initialize the network weights.Experimental results show that the improved algorithm effectively improves the accuracy of pneumonia detection.In summary,this paper proposes to use transfer learning to alleviate the shortage of chest X-ray samples and improve the effect of medical imaging classification model.Two improved one-stage object detection algorithms are proposed to achieve and verify the feasibility of chest X-ray pneumonia detection and improve detection accuracy.