Skip to main content

Table 3 Performance of seven DL models based on ECA-ResNet50t, Inception-Resnet-V2, EfficientNet-B3, and EfficientNet-B3 trained on the original image, Deeplab seg image, extended 5 image, extended 10 image, extended 20 image, extended 40 image, and extended 60 image

From: Predicting prognosis of nasopharyngeal carcinoma based on deep learning: peritumoral region should be valued

Dataset

Neural network

Sensitivity

Specificity

F1 score

Precision

AUC

95%CI

aAUC (SD)

Original image

ECA-ResNet50t

0.810

0.724

0.803

0.796

0.774

0.730–0.818

0.717 (0.043)

Inception-Resnet-V2

0.738

0.695

0.750

0.763

0.722

0.685–0.759

EfficientNet-B3

0.656

0.695

0.696

0.741

0.676

0.635–0.721

EfficientNet-B0

0.659

0.743

0.712

0.773

0.695

0.644–0.746

Deeplab seg image

ECA-ResNet50t

0.781

0.695

0.777

0.773

0.745

0.704–0.786

0.739 (0.016)

Inception-Resnet-V2

0.774

0.729

0.783

0.791

0.759

0.719–0.798

EfficientNet-B3

0.706

0.752

0.746

0.791

0.727

0.684–0.770

EfficientNet-B0

0.688

0.771

0.740

0.800

0.725

0.678–0.773

Expand 5 image

ECA-ResNet50t

0.803

0.719

0.797

0.792

0.768

0.728–0.809

0.760(0.010)

Inception-Resnet-V2

0.778

0.733

0.786

0.795

0.761

0.718–0.804

EfficientNet-B3

0.746

0.790

0.783

0.825

0.765

0.718–0.812

EfficientNet-B0

0.710

0.790

0.760

0.818

0.745

0.703–0.787

Expand 10 image

ECA-ResNet50t

0.821

0.738

0.813

0.806

0.786

0.739–0.832

0.768 (0.018)

Inception-Resnet-V2

0.789

0.738

0.794

0.800

0.766

0.730–0.802

EfficientNet-B3

0.756

0.800

0.793

0.834

0.777

0.732–0.822

EfficientNet-B0

0.706

0.790

0.758

0.817

0.744

0.707–0.781

Expand 20 image

ECA-ResNet50t

0.849

0.767

0.839

0.829

0.817

0.766–0.868

0.802 (0.013)

Inception-Resnet-V2

0.817

0.776

0.823

0.829

0.801

0.757–0.845

EfficientNet-B3

0.789

0.829

0.822

0.859

0.805

0.754–0.856

EfficientNet-B0

0.749

0.833

0.799

0.857

0.785

0.732–0.837

Expand 40 image

ECA-ResNet50t

0.875

0.790

0.861

0.847

0.84

0.783–0.897

0.782 (0.039)

Inception-Resnet-V2

0.774

0.738

0.785

0.797

0.758

0.713–0.803

EfficientNet-B3

0.742

0.786

0.780

0.821

0.761

0.721–0.802

EfficientNet-B0

0.735

0.819

0.785

0.844

0.77

0.725–0.815

Expand 60 image

ECA-ResNet50t

0.806

0.724

0.801

0.795

0.773

0.732–0.814

0.753 (0.014)

Inception-Resnet-V2

0.767

0.724

0.777

0.787

0.75

0.709–0.791

EfficientNet-B3

0.728

0.771

0.766

0.809

0.747

0.708–0.786

EfficientNet-B0

0.703

0.786

0.754

0.813

0.74

0.699–0.781

  1. AUC Area under the curve, aAUC Average area under the curve (AUC) of the four DL models based on ECA-ResNet50t, Inception-Resnet-V2, EfficientNet-B3, and EfficientNet-B0, SD Standard deviation