Research On Lung X-Ray Image Classification Model Based On Deep Learning

In lung X-ray images,the features of tissues and lesions are not significant,the types are diverse,and the morphological scales are changeable,which become a difficult problem for image classification.The traditional lung X-ray image classification relies too much on the diagnostic experience of doctors,is highly subjective and has low repeatability.It is obviously unable to meet the needs of clinical diagnosis when the number of medical images increases rapidly.However,due to the defects of structural design,some existing lung X-ray image classification models do not fully extract the deep features in the images,and the small amount of data available for model training and the imbalance of categories limit the improvement of classification accuracy.In view of this,this thesis combined with deep learning related technologies to study the classification model of lung X-ray images,as follows:(1)This thesis constructs a lung X-ray image preprocessing model based on Res Unet.Based on U-Net architecture,residual learning unit is introduced to transfer low-level information,skip connection are used to enhance the relationship between layers,and image features are fully learned through lower convolution and upper sampling to achieve the purpose of target region extraction,and image scaling and enhancement are combined to improve the pre-processing effect.The experimental results show that the proposed model completes the preprocessing task well and performs better than the comparison model in the target region extraction.(2)This thesis proposes a binary classification model for pulmonary tuberculosis X-ray images based on VGG-16 and SE-Block.After the preprocessing model was used to process the lung X-ray images,the VGG-16 network combined with SE-Block was used to extract the features of the images.Finally,SVM was used to classify the normal X-ray and pulmonary X-ray.The experimental results show that the proposed model can effectively alleviate the problems of model training difficulty and low classification accuracy caused by insufficient lung X-ray images,and accurately classify pulmonary tuberculosis X-ray images.(3)This thesis proposes a fine-grained classification model for lung diseases based on multi-mode bilinear pooling.Disease features were extracted using Efficient Net B0 and improved Mobile Net V2.The extracted features were fused based on multi-mode bilinear pooling,and the classification of imbalanced data was improved by Focal Loss function.The experimental results show that the proposed model still has good multiclassification performance when the categories of lung X-ray data are imbalanced,which is helpful to assist the diagnosis of lung diseases.