Skip to main content

Table 1 Comparison of ROC curves for different models

From: Preoperative diagnosis of meningioma sinus invasion based on MRI radiomics and deep learning: a multicenter study

 

AUC

ACC

SN

SP

PPV

NPV

Train set

Radiomics

0.791

0.695

0.774

0.660

0.500

0.870

ResNet

0.775

0.708

0.774

0.679

0.514

0.873

DenseNet

0.827

0.689

0.914

0.590

0.494

0.940

VGG

0.709

0.649

0.688

0.632

0.451

0.822

Combined

0.818

0.725

0.753

0.712

0.534

0.868

Internal validation set

Radiomics

0.722

0.643

0.795

0.578

0.449

0.867

ResNet

0.808

0.690

0.795

0.644

0.492

0.879

DenseNet

0.762

0.574

0.872

0.444

0.405

0.889

VGG

0.694

0.636

0.692

0.611

0.435

0.821

Combined

0.814

0.744

0.821

0.711

0.552

0.901

External validation set

Radiomics

0.704

0.635

0.722

0.475

0.716

0.483

ResNet

0.747

0.749

0.787

0.678

0.817

0.635

DenseNet

0.684

0.671

0.861

0.322

0.699

0.559

VGG

0.645

0.617

0.630

0.593

0.739

0.467

Combined

0.769

0.701

0.778

0.559

0.764

0.579

  1. ACC balanced accuracy, AUC area under receiver operating characteristic curve, SN sensitivity, SP specificity, PPV positive predictive value, NPV negative predictive value. Combined, including the radiomics features and ResNet features
Back to article page

Cancer Imaging

ISSN: 1470-7330

Contact us