基于轻量级神经网络的新冠肺炎CT新型识别技术

目的:为了满足临床新冠肺炎检测的实际需求，提出一种基于轻量级人工神经网络的新冠肺炎CT新型识别技术。方法:首先，选取目前公开的所有新冠肺炎CT图像数据集，经过图像亮度规范化和数据集清洗后作为训练数据，通过大样本提高深度学习的泛化能力；其次，采用GhostNet轻量级网络简化网络参数，使深度学习模型能够在医用计算机上运行，提高新冠肺炎CT诊断的效率；再次，在网络输入中加入肺部区域分割图像，进一步提高新冠肺炎CT诊断的准确性；最后，提出加权交叉熵损失函数减小漏诊率。结果:在本研究构建的数据集上进行测试，所提出方法的精确率、召回率、准确率和F1值分别为83%、96%、90%和88%，且在医用计算机上耗时为236 ms。结论:本研究提出方法的效率和准确性均优于其他对比算法，能较好地适应新冠肺炎诊断的需求。

COVID-19 recognition technology based on lightweight neural network

Abstract: Objective To propose a novel corona virus disease 2019 (COVID-19) recognition technology based on lightweight neural network for meeting the actual needs of COVID-19 detection in clinic. Methods All public COVID-19 CT image data set were selected and taken as training data after gray-level normalization and data cleaning. The generalization ability of deep learning was improved by large sample. Then the lightweight network GhostNet was adopted to simplify the network parameters, so that the deep learning model could run on medical computer and improved the efficiency of COVID-19 diagnosis based on CT. Subsequently, the diagnostic accuracy was further improve by adding lung image segmentation to network input. Finally, a weighted cross-entropy loss function was used to reduce the rate of missed diagnosis. Results The proposed method was tested on data set constructed in this study. The precision, recall rate, accuracy and F1 value of the proposed method were 83%, 96%, 90% and 88% respectively, and it took 236 ms to complete COVID-19 recognition on medical computer. Conclusion The proposed method is superior to other algorithms in efficiency and accuracy, and it can better meet the needs of COVID-19 diagnosis.