Fig. 4.

a Bootstrapped ROCs for discriminating COVID-19 from ILD, other pneumonia, and no pathology control by the models proposed in this study. The models M1, M2, and M3 and CO-RADS scoring [16] were evaluated with 100 COVID-19 positive, 33 ILD, 33 other pneumonia, and 34 healthy without pathologies on CTs. The 95% confidence intervals (shown as a band) are computed by bootstrapping over 1000 samples with replacement from the predicted scores. b Bootstrapped ROCs for our 3D DL classifier (M3), the model proposed by Li et al [10], and the model proposed by Harmon et al [15]. For the model proposed by Li et al, we trained and tested on our dataset using the code provided by the authors. The 95% confidence intervals (shown as a band) are computed by bootstrapping over 1000 samples with replacement from the predicted scores