Table 6 Gain in class prediction performance after using CIRCA compared to COVID-Net in the hold-out and independent test sets.
From: CIRCA: comprehensible online system in support of chest X-rays-based screening by COVID-19 example
Dataset | Index | Normal [%] | Pneumonia [%] | COVID-19 [%} |
|---|---|---|---|---|
Hold-out POLCOVID, COVIDx and AIforCOVID test sets | Sensitivity | 22.00 | 28.00 | − 10.89 |
Specificity | − 3.07 | − 7.47 | 30.00 | |
NPV | 7.28 | 8.99 | − 3.37 | |
PPV | − 4.16 | − 9.33 | 24.52 | |
Balanced Accuracy | 9.47 | 10.27 | 9.56 | |
F1 | 11.55 | 17.63 | 3.65 | |
BIMCV independent test set | Sensitivity | 24.96 | 30.67 | − 11.35 |
Specificity | − 7.58 | − 11.07 | 37.36 | |
NPV | 14.07 | 1.07 | − 0.19 | |
PPV | − 1.03 | − 7.29 | 25.20 | |
Balanced Accuracy | 8.69 | 9.80 | 13.00 | |
F1 | 18.07 | 9.73 | 23.04 |
