Table 3 Performance analysis of prediction models based on DL-based lesion subtyping, full lung radiomics or radiologist assessment for three outcomes (mortality, ICU admission and need of mechanical ventilation) studied using five-fold cross validation in a cohort of 103 subjects with RT-PCR positive COVID-19 pneumonia.

Mortality

DL-based lesion subtyping

0.874

[0.790,0.959]

1

[1]

0.775

[0.674,0.876]

0.5167 [0.196,0.838]

0.4833

[0.162,0.804]

Radiomics (full lung)

0.838

[0.749,0.927]

0.917

[0.791,1]

0.862

[0.78,0.944]

0.25 [−0.13,0.63]

0.5

[0.062,0.938]

Radiologist

0.725

[0.620,0.830]

0.875

[0.685,1]

0.7

[0.52,0.88]

0.25 [-0.13,0.63]

0.5

[0.062,0.938]

ICU admission

DL-based lesion subtyping

0.726

[0.582,0.871]

0.867

[0.743,0.991]

0.638

[0.475,0.801]

0.3462 [-0.0118,0.704]

0.4167

[0.0367,0.797]

Radiomics (full lung)

0.624

[0.446,0.802]

0.758

[0.61,0.906]

0.588

[0.37,0.806]

0.3235

[-0.0345,0.681]

0.6875

[0.329,1.05]

Radiologist

0.543

[0.394,0.691]

0.517

[0.248,0.78]

0.783

[0.63,0.936]

0.1111

[-0.0579,0.28]

0.6389

[0.279,0.999]

Mechanical ventilation

DL-based lesion subtyping

0.679

[0.496,0.862]

0.938

[0.843,1]

0.579

[0.453,0.705]

0.25

[-0.13,0.63]

0.75

[0.093,1.41]

Radiomics (full lung)

0.675

[0.494,0.857]

0.917

[0.791,1]

0.549

[0.359,0.739]

0.125

[-0.065,0.315]

0.875

[0.685,1.06]

Radiologist

0.302

[0.110,0.494]

0.917

[0.791,1]

0.192

[0.01,0.394]

0

[0,0]

1

[1]