基于异常检测与注意力机制的COVID-19识别模型

COVID-19 Recognition Model Based on Anomaly Detection and Attention Mechanism

旨在开发一种新的深度学习模型,以便从大量的胸部X光片中快速、可靠地筛查新冠肺炎患者。主要胸部X光片影像特征提出了一种融合异常检测技术以及Attention机制的COVID-19识别模型(Dev-SEDenseNet、Dev-SEResNet)。其中,Attention机制通过对CNN网络提取的特征分配注意力权重,能够在一定程度上排除无关信息的干扰。其次,将深度异常检测技术与卷积神经网络结合,能够更好地帮助模型学习异常数据特征。上述两个模型相较于原始DenseNet、ResNet算法性能指标有了较大提升,可以有效提高新冠肺炎预诊率。

This article aims to develop a new deep learning model to quickly and reliably screen patients with new coronary pneumonia from a lot of chest X-rays.Mainly based on the convolutional neural network algorithm to extract the image features of the chest X-ray film,and to solve the problem of insufficient correlation analysis and lack of interpretability in the CNN algorithm in solving the problem of chest X-ray film disease diagnosis.Dev-SEDenseNet and Dev-SEResNet model consist of anomaly detection technology and Attention mechanism.Among them,the Attention mechanism can eliminate the interference of irrelevant information to a certain extent by assigning attention weights to the features extracted by the CNN network.Secondly,the combination of deep anomaly detection technology and convolutional neural network can better help the model learn the characteristics of abnormal data.In order to evaluate the model,196 chest X-rays of patients with confirmed new coronary pneumonia were collected from the Github database,and 1457 chest X-rays of patients with other lung diseases were collected from the public data set Chest X-ray14.Compared with the original DenseNet and ResNet algorithm,the performance index of models have been greatly improved,which can effectively improve the pre-diagnosis rate of new coronary pneumonia.