Table 2 Testing results for the hold-out sets per subtype and dataset.
From: CIRCA: comprehensible online system in support of chest X-rays-based screening by COVID-19 example
Sub-type | Sensitivity [%] | Specificity [%] | NPV [%] | PPV [%] | Balanced accuracy [%] | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
POLCOVID | COVIDx | AIforCOVID | POLCOVID | COVIDx | AIforCOVID | POLCOVID | COVIDx | AIforCOVID | POLCOVID | COVIDx | AIforCOVID | POLCOVID | COVIDx | AIforCOVID | |
N1 | 96.00 | 100.0 | – | 100.0 | 100.0 | – | 99.50 | 100.0 | – | 100.0 | 100.0 | – | 98.00 | 100.0 | – |
N2 | 74.00 | 92.00 | – | 97.75 | 99.75 | – | 96.78 | 99.01 | – | 80.43 | 97.87 | – | 85.88 | 95.88 | – |
N3 | 62.00 | 88.00 | – | 95.25 | 97.75 | – | 95.25 | 98.49 | – | 62.00 | 83.02 | – | 78.63 | 92.88 | – |
P1 | 77.78 | 100.0 | – | 99.28 | 100.0 | – | 98.09 | 100.0 | – | 90.32 | 100.0 | – | 88.53 | 100.0 | – |
P2 | 74.51 | 98.00 | – | 99.75 | 99.75 | – | 96.84 | 99.75 | – | 97.44 | 98.00 | – | 87.13 | 98.88 | – |
P3 | 73.02 | 94.00 | – | 89.66 | 92.00 | – | 95.33 | 99.19 | – | 53.49 | 59.49 | – | 81.34 | 93.00 | – |
C1 | 84.00 | 84.48 | 94.34 | 98.75 | 100.0 | – | 98.01 | 97.76 | – | 89.36 | 100.0 | – | 91.38 | 92.24 | – |
C2 | 88.00 | 83.82 | 95.59 | 96.25 | 99.74 | – | 98.47 | 97.19 | – | 74.58 | 98.28 | – | 92.13 | 91.78 | – |
C3 | 42.00 | 50.00 | 37.93 | 94.25 | 99.53 | – | 92.86 | 97.25 | – | 47.73 | 85.71 | – | 68.13 | 74.77 | – |
Subtype-balanced class estimates | |||||||||||||||
N | 77.33 | 93.33 | – | 97.67 | 99.17 | – | 97.18 | 99.17 | – | 80.81 | 93.63 | – | 87.50 | 96.25 | – |
P | 75.10 | 97.33 | – | 96.23 | 97.25 | – | 96.75 | 99.65 | – | 80.42 | 85.83 | – | 85.67 | 97.29 | – |
C | 71.33 | 72.77 | 75.95 | 96.42 | 99.76 | – | 96.45 | 97.40 | – | 70.56 | 94.66 | – | 83.88 | 86.26 | – |
Class-balanced overall estimates | |||||||||||||||
Global | 74.59 | 87.81 | 75.95 | 96.77 | 98.73 | – | 96.79 | 98.74 | – | 77.26 | 91.37 | – | 85.68 | 93.27 | – |
