Skip to main content

Table 5 Comparison of classification methods for thoracic diseases. The classification methods, measurements, and best results in the review are shown in each column, respectively

From: Computer-aided detection in chest radiography based on artificial intelligence: a survey

  Methods Measurements Best results
Traditional machine learning methods Maharanobis distance [70] AUC Lung nodules: 85%
KNN [67] Sensitivity Lung nodules 4FP/image: 67%
ANN [71] Sensitivity Lung nodules 5.05FP/image: 70.1%
SVM [30, 65, 89, 104] Sensitivity, accuracy
Specificity, AUC
Lung nodules: sensitivity 5FP/image: 83.3%
Tuberculosis: accuracy 82.8%, specificity 86.8%, sensitivity 78.8%
Cardiomegaly: accuracy 76.5%, sensitivity
77.1%, AUC 79.2%
Pleural effusion: AUC 80%
Septum enlargement: AUC 88.2%
Fisher linear discriminant [72] Sensitivity Lung nodules 4FP/image: 78.1%
Minimum distance [80] Accuracy Tuberculosis: 95.7%
Decision tree [81] Accuracy Tuberculosis: 94.9%
Bayesian classifier [88] Sensitivity Tuberculosis: 0.237 FP/image: 82.35%
Traditional machine learning methods + CNN CNN transfer learning + SVM [75] AUC Right pleural effusion: 93%
Cardiomegaly: 89%
AlEXNET transfer learning + random forests [77] Sensitivity, specificity Lung nodules: 1.19FP/image: sensitivity 69.27%, specificity 96.02%
Deep learning methods RESNET transfer learning [76] Sensitivity
Specificity
Lung nodules: sensitivity 92%, specificity 86%
CNN transfer learning [90] AUC, accuracy Tuberculosis: 96.4%, 90.3%
CNN [101] Sensitivity, accuracy, AUC, specificity Cardiomegaly: 93%, 97%, 94%, 92%
GoogleNet CNN [106] AUC Cardiomegaly: 87.5%, pneumothorax: 86.1%, pleural effusion: 96.2%, pulmonary edema: 86.8%