Skip to main content
. 2022 Nov 30;36(2):603–616. doi: 10.1007/s10278-022-00734-4

Table 3.

Summary results reported for each predicted outcomes and kind of validation

Outcomes AI model Validation ACC SENS SPEC PPV NPV ROC-AUC MCC F1-score
Mortality CT Internal 68.3% 28% 89.7% 70.5% 59.3% 77% 0.23 40%
External 80.4% 28.7% 90.4% 86.8% 38% 70% 0.22 43%
CT + CF Internal 91.7% 90.5% 92.4% 94.8% 86.2% 95% 0.82 92.6%
External 71.7% 55.6% 74.8% 90% 29.8% 72% 0.25 68.7%
CT + CF + LD Internal 92.7% 90.5% 93.7% 95% 88.7% 96% 0.84 92.7
External 70.2% 66% 71% 91.2% 31.6% 74% 0.3 76.6%
CF + LD Internal 66% 53.7% 72.5% 75% 50.8% 71% 0.26 62.6%
External 76.8% 11.5% 89% 84% 16% 45% 0.005 20.2%
Intubation CT Internal 70.3% 75% 41.7% 22.4% 88.6% 63% 0.14 34.5%
External 71.9% 75.4% 30% 9.3% 92.8% 53% 0.03 16.6%
CT + CF Internal 91.3% 91.5% 89.8% 64.3% 98.3% 95% 0.71 75.5%
External 72.6% 74.7% 45.7% 12.4% 94.7% 64% 0.12 21.3%
CT + CF + LD Internal 90% 90% 90.3% 60% 98.3% 95% 0.7 72%
External 70.7% 72.3% 50% 12.6% 95% 66% 0.3 21.5%
CF + LD Internal 64% 62.6% 71.3% 28.4% 91.8% 68% 0.26 40%
External 68.7% 71.3% 34.3% 8.8% 93.4% 51% 0.03 16%
ICU admission CT Internal 75.3% 84% 37% 38% 86.8% 73% 0.21 52.3%
External 80% 84% 30% 15.5% 93.8% 63% 0.11 26%
CT + CF Internal 89.6% 90% 86.5% 65.6% 97% 94% 0.69 76%
External 74.7% 77% 46% 13.4% 95% 70% 0.14 23%
CT + CF + LD Internal 89% 89% 86.7% 60.7% 97.7% 94% 0.66 72%
External 73% 74.6% 52% 14% 95.2% 69% 0.16 24%
CF + LD Internal 60% 57.5% 70% 20.7% 92.4% 73% 0.19 30.4%
External 76% 79% 36% 11.7% 94% 57% 0.09 20.4

ACC accuracy, SENS sensitivity, SPEC specificity, PPV positive predictive value, NPV negative predictive value, ROC-AUC receiving operator characteristic – area under the curve, MCC Matthew correlation coefficient, CT computerized tomography features, CT + CF computerized tomography features and clinical features except laboratory data, AI artificial intelligence, CT + CF + LD computerized tomography features, clinical features, and laboratory data, CF + LD clinical features and laboratory data