CT Images Analysis Of Novel Coronavirus Pneumonia By Enhanced Deep Learning

Until now,there is still clinical difficulty in the diagnosis of a catastrophic outbreak of novel coronavirus pneumonia in the world.CT examination is the main method of lung disease examination.Doctors can judge whether the lungs are diseased by observing CT images.However,the imaging features of some early novel coronavirus pneumonia lag behind the clinical manifestations,which makes the diagnosis of early novel coronavirus pneumonia difficult.It is of great practical and theoretical significance to use deep learning methods to explore the characteristics of early novel coronavirus pneumonia in CT images,and to construct a high-sensitivity early warning model for novel coronavirus pneumonia accordingly.In view of the difficulty in the diagnosis of early novel coronavirus pneumonia,recognition of early novel coronavirus pneumonia CT images based on enhanced deep learning analysis framework includes three aspects.First,we construct an interpretable lesion feature labeling strategy based on neural network.Based on the class activation maps,the model visually annotates important features in the image.At the same time,it is provided to the doctor as the interpretability basis of the model.On this basis,in order to reduce the labeling burden of doctors,the text feature discovery method based on self-clustering is used to extract the most critical lesion features from the CT image reports.Therefore,it can be semiautomatically associated with the visually marked features,and then find the CT image lesion features critical for the early identification of novel coronavirus pneumonia,and it is up to the doctor to judge.Secondly,an enhanced convolutional neural network fusion learning method is constructed.In the second round of iterative learning,the activation function is used to enhance the critical lesion features found in the first round.The enhanced image and the original image are analyzed using different convolutional neural networks to improve the model’s ability to extract features.The CT images and CT reports of 162 cases of early novel coronavirus pneumonia and 140 cases of common pneumonia in Wuhan Tongji Hospital were extracted for experimental verification.Combined with the doctor’s experience,the activation mapping was used to find out the characteristics of ground glass and light shadow of early novel coronavirus pneumonia.After extracting areas of interest,data augmentation,and images processing,the features of early new novel coronavirus pneumonia were extracted and annotated,which was recognized by the doctor.The enhanced iterative model constructed effectively improved the recognition rate of early novel coronavirus pneumonia.The Area Under Curve of the model for the classification of patients with novel coronavirus pneumonia reached 0.943,and the sensitivity was as high as 97%.Compared with the existing deep learning methods,the accuracy was improved by 3%.