Table 3 Performance comparison of different prediction models.

From: Noncontrast CT-based deep learning for predicting intracerebral hemorrhage expansion incorporating growth of intraventricular hemorrhage

Model

 

AUC (95%CI)

Accuracy

Sensitivity

Specificity

PPV

NPV

F1-Score

BRAIN score

Training

0.729(0.681–0.772)

0.712

0.254

0.911

0.556

0.737

0.349

 

Internal-testing

0.655(0.577–0.727)

0.695

0.294

0.871

0.500

0.737

0.370

 

External-testing

0.658(0.591–0.720)

0.685

0.350

0.878

0.622

0.701

0.448

Clinical-radiologic

Training

0.782(0.737–0.822)

0.756

0.407

0.908

0.658

0.778

0.503

 

Internal-testing

0.688(0.612–0.757)

0.689

0.451

0.793

0.489

0.767

0.469

 

External-testing

0.690(0.625–0.751)

0.662

0.463

0.777

0.544

0.715

0.500

Combined-logistic

Training

0.788(0.744–0.828)

0.761

0.407

0.915

0.676

0.780

0.508

 

Internal-testing

0.682(0.605–0.752)

0.689

0.530

0.759

0.491

0.786

0.509

 

External-testing

0.694(0.629–0.755)

0.630

0.488

0.712

0.494

0.707

0.490

Combined-SVM

Training

0.855(0.846–0.912)

0.784

0.398

0.952

0.783

0.784

0.528

 

Internal-testing

0.692(0.616–0.761)

0.701

0.412

0.827

0.512

0.762

0.457

 

External-testing

0.703(0.638–0.763)

0.671

0.338

0.863

0.587

0.694

0.429

2D-ResNet-101

Training

0.882(0.843–0.910)

0.802

0.746

0.827

0.652

0.882

0.696

 

Internal-testing

0.782(0.712–0.842)

0.766

0.667

0.810

0.607

0.847

0.636

 

External-testing

0.777(0.716–0.830)

0.767

0.637

0.842

0.699

0.801

0.667

2D-ResNet-152

Training

0.879(0.843–0.910)

0.802

0.661

0.863

0.678

0.854

0.670

 

Internal-testing

0.795(0.726–0.854)

0.731

0.667

0.759

0.548

0.838

0.602

 

External-testing

0.759(0.696–0.814)

0.717

0.700

0.727

0.596

0.808

0.644

2D-DenseNet-121

Training

0.868(0.830–0.900)

0.830

0.627

0.923

0.779

0.850

0.695

 

Internal-testing

0.721(0.646–0.787)

0.683

0.627

0.707

0.485

0.812

0.547

 

External-testing

0.735(0.671–0.792)

0.689

0.650

0.712

0.565

0.780

0.604

2D-DenseNet-201

Training

0.902(0.868–0.929)

0.823

0.763

0.849

0.687

0.891

0.723

 

Internal-testing

0.759(0.687–0.822)

0.737

0.647

0.776

0.559

0.833

0.600

 

External-testing

0.740(0.677–0.797)

0.703

0.662

0.727

0.582

0.789

0.620

3D-ResNet-101

Training

0.945(0.902–0.955)

0.882

0.771

0.930

0.827

0.903

0.798

 

Internal-testing

0.644(0.567–0.717)

0.605

0.608

0.603

0.403

0.778

0.484

 

External-testing

0.615(0.547–0.680)

0.616

0.525

0.669

0.477

0.710

0.500

3D-ResNet-152

Training

0.932(0.902–0.955)

0.851

0.831

0.860

0.721

0.921

0.772

 

Internal-testing

0.644(0.566–0.716)

0.623

0.647

0.612

0.423

0.798

0.512

 

External-testing

0.577(0.509–0.643)

0.589

0.500

0.641

0.444

0.690

0.471

3D-DenseNet-121

Training

0.952(0.926–0.971)

0.913

0.881

0.926

0.839

0.947

0.860

 

Internal-testing

0.645(0.567–0.717)

0.641

0.510

0.698

0.426

0.764

0.464

 

External-testing

0.619(0.551–0.683)

0.635

0.537

0.691

0.500

0.722

0.518

3D-DenseNet-201

Training

0.938(0.909–0.959)

0.869

0.822

0.889

0.764

0.920

0.792

 

Internal-testing

0.665(0.588–0.736)

0.611

0.608

0.612

0.408

0.780

0.488

 

External-testing

0.510(0.442–0.578)

0.562

0.325

0.697

0.382

0.642

0.351

  1. PPV, positive predictive value; NPV, negative predictive value; SVM, support vector machine.
Back to article page