Table 2 The performance of predictive models and readers.

From: CT-based radiomics and deep learning models for predicting thyroid cartilage invasion and patient prognosis in laryngeal carcinoma

Cohort

AUC (95%CI)

Accuracy

Sensitivity

Specificity

PPV

NPV

Training cohort

Radiomics model

0.867 [0.819–0.916]

0.810

0.798

0.816

0.710

0.878

2D DL model

0.846 [0.797–0.895]

0.794

0.640

0.880

0.750

0.813

3D DL model

0.959 [0.934–0.984]

0.923

0.888

0.943

0.898

0.937

Nomogram

0.873 [0.831–0.915]

0.749

0.933

0.646

0.597

0.944

Internal validation cohort

Reader 1

0.742 [0.644–0.841]

0.727

0.558

0.836

0.686

0.747

Reader 2

0.727 [0.630–0824]

0.700

0.512

0.821

0.647

0.724

Radiomics model

0.727 [0.621–0.823]

0.682

0.605

0.731

0.591

0.742

2D DL model

0.835 [0.758–0.911]

0.791

0.744

0.821

0.727

0.833

3D DL model

0.732 [0.638–0.827]

0.682

0.674

0.687

0.580

0.774

Nomogram

0.867 [0.799–0.936]

0.836

0.953

0.761

0.719

0.962

External validation cohort

Reader 1

0.726 [0.598–0.854]

0.689

0.565

0.763

0.591

0.843

Reader 2

0.715 [0.790–0.944]

0.689

0.478

0.816

0.591

0.744

Radiomics model

0.705 [0.567–0.843]

0.721

0.391

0.921

0.750

0.714

2D DL model

0.804 [0.696–0.913]

0.705

0.957

0.553

0.564

0.955

3D DL model

0.698 [0.569–0.836]

0.607

0.870

0.447

0.488

0.850

Nomogram

0.823 [0.714–0.931]

0.754

0.826

0.711

0.633

0.871

  1. DL deep learning, AUC area under the curve, CI confidence interval, PPV positive predictive value, NPV negative predictive value.
Back to article page