Erdaw Yabsera, Tachbele Erdaw
Electrical and Mechanical Engineering, Addis Ababa Science & Technology University, Addis Ababa, Ethiopia.
Nursing & Midwifery, College of Health Sciences, Addis Ababa University, Addis Ababa, Ethiopia.
Int J Gen Med. 2021 Aug 28;14:4923-4931. doi: 10.2147/IJGM.S325609. eCollection 2021.
This research was designed to investigate the application of artificial intelligence (AI) in the rapid and accurate diagnosis of coronavirus disease 2019 (COVID-19) using digital chest X-ray images, and to develop a robust computer-aided application for the automatic classification of COVID-19 pneumonia from other pneumonia and normal images.
A total of 1100 chest X-ray images were randomly selected from three different open sources, containing 300 X-ray images of confirmed COVID-19 patients, 400 images of other pneumonia patients, and 400 normal X-ray images. In this study, a classical machine learning approach was employed. The model was built using the support vector machine (SVM) classifier algorithm. The SVM was trained by 630 features obtained from the HOG descriptor, which was quantized into 30 orientation bins in the range between 0 and 360. The model was validated using a 10-fold cross-validation method. The performance of the model was evaluated using appropriate classification metrics, including sensitivity, specificity, area under the curve, positive predictive value, negative predictive value, kappa, and accuracy.
The multi-level classification model was able to distinguish COVID-19 patients with a sensitivity of 97.92% and specificity of 98.91%, for the internal testing or cross-validation. For the independent external testing, the model showed sensitivity of 95% and specificity of 98.13%, for distinguishing COVID-19 from other pneumonia and no-findings. The binary classification model was able to distinguish COVID-19 patients with a sensitivity of 99.58% and specificity of 99.69%, for the internal testing. For the independent external testing, the model showed a sensitivity of 98.33% and specificity of 100%, for distinguishing COVID-19 from normal images.
The model can achieve the rapid and accurate identification of COVID-19 patients from chest X-rays with more than 97% accuracy. This high accuracy and very rapid computer-aided diagnostic approach would be very helpful to control the pandemic.
本研究旨在探讨人工智能（AI）在利用数字化胸部X线图像快速准确诊断2019冠状病毒病（COVID-19）中的应用，并开发一种强大的计算机辅助应用程序，用于从其他肺炎和正常图像中自动分类COVID-19肺炎。
从三个不同的公开来源中随机选取了1100张胸部X线图像，其中包括300张确诊COVID-19患者的X线图像、400张其他肺炎患者的图像和400张正常X线图像。在本研究中，采用了经典的机器学习方法。该模型使用支持向量机（SVM）分类算法构建。SVM通过从HOG描述符获得的630个特征进行训练，这些特征在0到360的范围内被量化为30个方向区间。该模型使用10折交叉验证方法进行验证。使用适当的分类指标评估模型的性能，包括敏感性、特异性、曲线下面积、阳性预测值、阴性预测值、kappa值和准确性。
对于内部测试或交叉验证，多级分类模型能够区分COVID-19患者，敏感性为97.92%，特异性为98.91%。对于独立的外部测试，该模型在区分COVID-19与其他肺炎及无异常发现时，敏感性为95%，特异性为98.13%。对于内部测试，二元分类模型能够区分COVID-19患者，敏感性为99.58%，特异性为99.69%。对于独立的外部测试，该模型在区分COVID-19与正常图像时，敏感性为98.33%，特异性为100%。
该模型能够以超过97%的准确率从胸部X线片中快速准确地识别COVID-19患者。这种高精度且非常快速的计算机辅助诊断方法将对控制疫情非常有帮助。
