Fig. 6

Decision curve analysis for the radiomics nomogram. The x-axis represents the threshold probability and the y-axis measures the net benefit. The blue line represents the model combines the clinical features with radiomics scores and the pink line represents the radiomics nomogram. The green line assumes that all patients are severe COVID-19. Thin black line hypothesizes that all patients are non-severe COVID-19. The calculation of net benefit was performed by subtracting the proportion of false positive from proportion of true positive in all patients, weighting with the relative harm of giving up treatment compared with the negative consequence of an unnecessary treatment [24]. The calculation of relative harm was performed by (Pt/(1Pt)), Pt is the threshold probability, where the expected benefit of treatment is equal to the expected benefit of avoiding treatment. According to the decision curve, using the method which combines radiomics with clinic information to predict the probability of severe COVID-19 always adds more benefit than other three methods (only including radiomics, or the teat-all-patients scheme, or treat-none scheme). For example, if the personal threshold probability of a patient is 80% (the patient is willing to receive treatment if his probability of severe COVID-19 is > 80%), then the benefit is about 0.6 when using the method that combines radiomics with clinic to decide whether to undergo treatment, which exhibits the best benefit compared with other three methods