Table 1. Characteristics of CAD systems in chest radiographs interpretation reported in the literature.
| Disease | Reference | Database | Commercial | Performance |
|---|---|---|---|---|
| Lung nodules | 5 | 45 patients with lung nodules and 45 healthy patients | Yes | Area under the curve was improved from 0.924 without the CAD to 0.986 with the commercialized CAD |
| 7 | 34 patients | Yes | Overlooked cancer in 12 (35%) patients was detected but with 5.9 FP/image | |
| 8 | 114 chest radiographs (89 patients) | Yes | Overlooked nodules were detected on 47% of radiographs and in more than 50% of patients with 3.9 FP per radiograph | |
| 9 | 113 patients | Yes | The sensitivity of inexperienced readers for the detection of small nodules was improved (39 vs. 45%) and remained unchanged for experienced readers (50 vs. 51%) | |
| 47 | 25 normal and 25 abnormal | No | The value of the AUC has improved from 0.926 to 0.962 | |
| 48 | Training set: 172 TP nodules, 44 TP + 377 FP for the test set. For the template 3,077 FP + 236 TP | No | 44.3% of the FP nodules were removed with reduction of TP (2.3%) | |
| 49 | 924 chest radiographs | No | 70.1% of sensitivity with 5 FP/image | |
| 50 | 106 pairs of PA and lateral views | No | Sensitivity of 70.5%, with 4.9 FP/image for the PA view, sensitivity of 86.9% (6.6 FP/2 images) for PA + lateral view | |
| 51 | Public database: 247 chest radiographs | No | A sensitivity of 0.71 and 0.78 with 1.5 and 2.5 FP/image, respectively | |
| Interstitial lung disease | 52 | 20 radiographs with interstitial infiltrates and 20 normal chest radiographs | No | The radiologists' detection accuracy was improved from an Az of 0.948 to 0.970 |
| Pneumoconiosis | 54 | 300 normal and 125 pneumoconiosis cases | No | Sensitivity of 91.2% at a specificity of 86.3% |
| 55 | 85 normal chest radiographs and 40 with pneumoconiosis | No | Sensitivity, specificity, and accuracy were 0.974 ± 0.018, 0.957 ± 0.021, 0.873 ± 0.024, and 0.929 ± 0.018, respectively | |
| TB | 10 | 161 subjects: 120 had abnormal chest radiographs | Yes | The AUCs for CAD and the clinical officers were, respectively, 0.91 and 0.89–0.94 |
| 59 | 147 images with TB and 241 normal images. 100 normal images and 100 abnormal images with interstitial disease | No | For the diffuse lung disease database, a sensitivity of 0.97 at a specificity of 0.90. For the TB database, a sensitivity of 0.86 at a specificity of 0.50 | |
| 60 | 365 chest radiographs | No | The AUC (Az) is 85% | |
| 61 | The first: 138 chest radiographs (80 normal, 58 TB). The second set: 340 normal, 275 with TB | No | Accuracy of 78.3 for the first set and 84% for the second set | |
| 62 | 50 normal and 45 with TB | No | Sensitivity of 91% at a specificity of 95.4%. | |
| ARDS | 63 | 120 chest radiographs A training database:321 patches Test database: 90 radiographs |
No | Sensitivity of 90.6% and a specificity of 86.5% |
Abbreviations: ARDS, acute respiratory distress syndrome; AUC, area under the ROC curve; CAD, computer-aided diagnosis; FP, false positive; TB, tuberculosis; TP, true positive.