Table 2 Performance of 3D and hybrid 3D classification models for two experimental conditions.

From: Artificial intelligence for the detection of COVID-19 pneumonia on chest CT using multinational datasets

Design

Model

Validation accuracy

Test summary stats

ACC

SENS

SPEC

PPV

NPV

AUC

Original training schema

3D

0.917

0.908

0.840

0.930

0.794

0.948

0.949

Hybrid 3D

0.924

0.889

0.853

0.901

0.735

0.950

0.947

Independent testing population

3D

0.939

0.896

0.845

0.916

0.793

0.939

0.941

Hybrid 3D

0.905

0.895

0.751

0.951

0.853

0.909

0.938

  1. Original training design included 1337 patients in testing cohort (of which, n = 326 patients with COVID-19 positivity). Independent testing population design included 1397 patients in testing cohort (entire patient cohort from Tokyo, Japan excluded from training/validation), with a total of n = 386 patients with COVID-19 positivity.
  2. ACC accuracy, SENS sensitivity, SPEC specificity, PPV positive predictive value, NPV negative predictive value, AUC area under the curve.
Back to article page