Table 3 The diagnostic performance of the different parameters of CXR analysis including the AI, combined visual/AI, and the visual models
Sensitivity | Specificity | PPV | NPV | Accuracy | |
|---|---|---|---|---|---|
Per CXR | |||||
AI | 87.9% (87/99) [80–94] | 90.2% (46/51) [79–97] | 94.6% (87/92) [88–88] | 79.3% (46/58) [69–87] | 88.7% (133/150) [82–93] |
Visual /AI | 86.9% (86/99) [79–93] | 86.3% (44/51) [74–94] | 92.5% (86/93) [86–96] | 77.2% (44/57) [67–85] | 86.7% (130/150) [80–92] |
Visual | 82.8% (82/99) [74–90] | 80.4% (41/51) [67–90] | 89.1% (82/92) [82–94] | 70.7% (41/58) [61–79] | 82% (123/150) [75–88] |
Per nodule* | |||||
AI | 84.7% (183/216) [79–89] | 90.2% (46/51) [79–97] | 97.3% (183/188) [94–99] | 58.2% (46/79) [50–66] | 85.8% (229/267) [81–90] |
Visual /AI | 80.6% (174/216) [75–86] | 86.3% (44/51) [74–94] | 96.1% (174/181) [93–98] | 51.2% (44/86) [44–58] | 81.7% (218/267) [76–86] |
Visual | 74.1% (160/216) [68–80] | 80.4% (41/51) [67–90] | 94.1% (160/170) [90–97] | 42.3% (41/97) [36–49] | 75.3% (201/267) [70–80] |