Table 4 Radiologists’ specificity and sensitivity results for each threshold
From: A comparison of Covid-19 early detection between convolutional neural networks and radiologists
Reader | Threshold | Specificity | Sensitivity | Ensemble4Covid sensitivity |
|---|---|---|---|---|
Reader 1 | High | 98% (96–99%) | 20% (14–25%) | 31% (21–41%) |
Low | 86% (82–90%) | 35% (28–41%) | 66% (58–76%) | |
Reader 2 | High | 96% (94–98%) | 19% (14–24%) | 41% (33–48%) |
Low | 87% (83–91%) | 32% (27–39%) | 65% (55–73%) | |
Reader 3 | High | 98% (96–99%) | 15% (10–20%) | 31% (21–41%) |
Low | 88% (84–92%) | 25% (19–31%) | 62% (53–71%) | |
Reader 4 | High | 86% (81–90%) | 51% (44–58%) | 66% (58–76%) |
Low | 70% (64–75%) | 61% (55–68%) | 85% (79–89%) | |
Reader 5 | High | 82% (77–86%) | 56% (50–62%) | 76% (67–82%) |
Low | 67% (61–73%) | 69% (63–75%) | 85% (81–90%) |