Table 2 Radiologist and Deep Learning System Performance for Chest Radiographs and Tuberculosis.
Human Readers | Deep Learning Systems | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
CAD4TB (v6) | Lunit (v4.7.2) | qXR (v2) | ||||||||||
Accuracy | Sensitivity (95%CI) | Specificity (95% CI) | Accuracy | Sensitivity (95% CI) | Specificity (95% CI) | Accuracy | Sensitivity (95% CI) | Specificity (95%CI) | Accuracy | Sensitivity (95% CI) | Specificity (95% CI) | |
Nepal | ||||||||||||
Senior Radiologist | 0.57 | 0.96 | 0.48 | 0.74 | 0.96 | 0.69 | 0.67 | 0.96 | 0.6 | 0.7 | 0.97* | 0.65 |
(0.89–0.99) | (0.43–0.53) | (0.89–0.99) | (0.64–0.73) | (0.89–0.99) | (0.55–0.65) | (0.91–0.99) | (0.6–0.69)) | |||||
Junior Radiologist & Residents | 0.72 | 0.87 | 0.69 | 0.77 | 0.87 | 0.75 | 0.85 | 0.87 | 0.78 | 0.69 | 0.87 | 0.81 |
(0.79–0.93) | (0.64–0.73) | (0.79–0.93) | (0.71–0.79) | (0.79–0.93) | (0.73–0.82) | (0.79–0.93) | (0.76–0.84) | |||||
Cameroon | ||||||||||||
Radiologist | 0.74 | 0.8 | 0.74 | 0.9 | 0.8 | 0.9 | 0.94 | 0.8 | 0.94 | 0.94 | 0.8 | 0.95 |
(0.52–0.96) | (0.71–0.78) | (0.52–0.96) | (0.87–0.92) | (0.52–0.96) | (0.92–0.96) | (0.52–0.96) | (0.93–0.96) | |||||
Teleradiology Company | 0.74 | 0.8 | 0.74 | 0.9 | 0.8 | 0.9 | 0.94 | 0.8 | 0.94 | 0.94 | 0.8 | 0.95 |
(0.52–0.96) | (0.71–0.77) | (0.52–0.96) | (0.87–0.92) | (0.52–0.96) | (0.92–0.96) | (0.52–0.96) | (0.93–0.96) | |||||
