Table 4.
Results for diagnostic imaging models – part 1.
| Study, Country, Outcome | No. of CPP* | AI methods | Predictors | Val. methods | Performance (AUC, Accuracy (Acc%), Sensitivity (SEN%), Specificity (SPE%), PPV/NPV (%), (95% CI)) | Risk of Bias**: Participants/Predictors/Outcome/Analysis/Overall | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| Chen et al. [92], China, Diagnosis | 51 | DL | Imaging features | TTS | Acc 95.24, SEN 100, SPE 93.55, PPV 84.62, NPV 100 | H | U | L | H | H |
| Rahimzadeh et al. [33], Diagnosis | 118 | DNN, Nadam optimizer | Imaging features | TTS | Acc 99.50 | L | U | L | H | H |
| Roy et al. [72], Italy, Diagnosis | 17 | DL | Imaging biomarkers | 5-FCV | F1-score 65.9 | H | U | Η | Η | Η |
| Zhou et al. [73], China, Diagnosis | 35 | DL | Imaging features | Ext. val. | AUC > 0.93 | H | U | U | H | H |
| Ter-Sarkisov et al. [93], China, Diagnosis | 150 | DL, CNN | Imaging features | TTS | Acc 91.66, SEN 90.80 | H | U | U | H | H |
| Qjidaa et al. [19], Early detection | unclear | DL, CNN | Imaging features | Int.- ext. val. | Acc 92.5, SEN 92 | U | U | U | H | H |
| Babukarthik et al. [74], Early diagnosis | 102 | GDCNN | Imaging features | unclear | Acc 98.84, SEN 100, SPE 97.0 | H | H | H | H | H |
| Minaee et al. [20], Diagnosis | unclear | CNN | Raw images without feature extraction | TTS | SEN 98, SPE 92 | U | U | U | H | H |
| Yan et al. [94], China, Diagnosis | 206 | CNN | Imaging features | TTS | SEN 99.5 (95%CI: 99.3–99.7), SPE 95.6 (95%CI: 94.9–96.2) | L | H | H | H | H |
| Lokwani et al. [95], India, Diagnosis | 55 | NN | Imaging features | TTS | SEN 96.4 (95% CI: 88–100), SPE 88.4 (95% CI: 82–94) | U | U | H | H | H |
| Jin et al. [96], China, Screening (early detection) | 751 | DL, DNN | Imaging features | TTS | AUC 0.97, SEN 90.19, SPE 95.76 | H | U | U | H | H |
| Ko et al. [29], South Korea, Diagnosis | 20 | 2D DL | Imaging features | TTS, ext.val. | Acc 99.87, SEN 99.58, SPE 100.00 | U | U | U | H | H |
| Ezzat et al. [34], Diagnostic imaging | 99 | Hybrid CNN | Not applicable | TTS | Acc 98 | H | U | H | L | H |
| Ouchicha et al. [40], Diagnosis | 43 | DCNN | Imaging features | 5-FCV | Acc 97.20 | U | U | L | H | H |
| Xiong et al. [75], China, Diagnosis | 521 | DL, CNN | Imaging features | TTS, ext. val. | AUC 0.95, Acc 96 (95% CI: 90–98), SEN 95 (95% CI: 83–100), SPE 96 (95% CI: 88–99) | H | U | H | H | H |
| Li et al. [2], China, Diagnosis | 468 | DL, CNN | Imaging features | TTS | AUC 0.96, SEN 90 (95% CI: 83–94), SPE 96 (95% CI: 93–98) | L | U | H | H | H |
| Mahmud et al. [25], China, Diagnosis | unclear | DCNN | Imaging features | 5- FCV | Acc 97.4 | U | U | U | H | H |
| Li et al. [97], China, Diagnosis | 305 | NN | Imaging features | TTS | Precision 93% | U | U | L | H | H |
| Sun et al. [98], China, Diagnosis | 1495 | LR, SVM, RF, NN | 30 Imaging features: Volume features, Infected lesion number, Histogram distribution, Surface area, Radiomics features | 5-FCV | ACC 91.79, SEN 93.05, SPE 89.95 | L | U | H | H | H |
| Results for diagnostic imaging models – part 2 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Study, Country, Outcome | No. of CPP* | AI methods | Predictors | Val. methods | Performance (AUC, Accuracy (Acc%), Sensitivity (SEN%), Specificity (SPE%), PPV/NPV (%), (95% CI)) | Risk of Bias**: Participants/Predictors/Outcome/Analysis/Overall | ||||
| Ozturk et al. [35], Diagnosis (automatic COVID-19 detection) | 125 | DNN | Imaging features (1. Ground-glass opacities (GGO) (bilateral, multifocal, subpleural, peripheral, posterior, medial and basal) 2. A crazy paving appearance (GGOs and inter-/intra-lobular septal thickening) 3. Air space consolidation. 4. Broncho vascular thickening (in the lesion). 5. Traction bronchiectasis | 5- FCV | Acc 98.08, SEN 95.13, SPE 95.03, F1-score 96.51 | H | U | H | H | H |
| Zhang et al. [7], United States, Diagnosis | 2060 | DNN (CV19-Net), NLP | Imaging features | TTS | AUC 0.92 (95% CI: 0.91–0.93), SEN 88% (95% CI: 87–89), SPE 79% (95% CI: 77–80) | H | U | H | H | H |
| Borkowski et al. [37], Diagnosis | 103 | Automated ML platform Microsoft CustomVision | Imaging features | Ext. val. with US Dep. of Vet. Affairs (VA) PACS | Acc 97, SEN 100, SPE 95, PPV 91, NPV 100 | H | H | H | H | H |
| Xu et al. [6], China, Diagnosis | 432 | DL, CNN, FL | Imaging features | 5-FCV | SEN 77.2, SPE 91.9 | L | U | U | H | H |
| Wang et al [99], Diagnosis | 266 | DCNN | Imaging features | Int. val. | Acc 93.3, SEN 91.0, PPV 98.9 | L | U | U | H | H |
| Elaziz et al. [21], Diagnosis | unclear | CNN | Imaging features | Ext. val. | Acc 98.09 | H | U | U | L | H |
| Gozes et al. [26], United States, China, Diagnostic imaging | unclear | DCNN | Imaging features | TTS | AUC 0.996, (95%CI: 0.989–1.00), SEN 98.2, SPE 92.2 | U | U | H | H | H |
| Salman et al. [22], Diagnosis | unclear | DL, CNN | Imaging features | TTS | Acc 100, SEN 100, SPE 100, PPV 100, NPV 100 | U | U | U | H | H |
| Liu et al. [24], Diagnosis | unclear | FL (Federated Learning) | Imaging features | Comparison with and without FL | ResNet18 (highest accuracy): Acc 91,26, SEN 96.15, SPE 91.26 | U | U | L | L | H |
| Castiglioni et al. [76], Italy, Diagnosis | 250 | CNN | unclear | 10- | ||||||
| FCV | AUC 0.89 (95%CI 0.86–0.91), SEN 78 (95% CI 74–81), SPE 82 (95%CI: 78–85) | H | U | H | H | H | ||||
| Padma et al. [23], Diagnosis | unclear | ML, CNN | Imaging features | TTS | Acc 99 | U | U | U | H | H |
| Zhang et al. [100], China, Diagnosis | 752 | Light GBM, CoxPH | Imaging features | 5-FCV | AUC 0.9797 (95% CI: 0.966–0.9904), Acc 92.49, SEN 94.93, SPE 91.13 | U | U | U | H | H |
| Wang et al. [77], China, Diagnosis | 79 | DL, CNN | Imaging features | Int. – ext. | ||||||
| val. | Acc 89.5, SEN 87, SPE 88 | H | U | U | H | H | ||||
*CPP = COVID-19 Positive Patients, Abbreviations of medical terms included in this Table are provided in the Appendix.
**L: Low, H: High, U: Unclear