Skip to main content

Table 2 Cross-testing results between multiple datasets where S denotes source dataset and T denotes testing dataset

From: A study of generalization and compatibility performance of 3D U-Net segmentation on multiple heterogeneous liver CT datasets

S

T

LiTS

Anatomy CT

Zhujiang

App-0

App-1

App-2

App-3

PB0

PB1

PD0

PD1

LiTS

–

0.907

0.958

0.943

0.937

0.909

0.889

0.795

0.732

0.860

0.832

AnatomyCT

0.661

–

0.630

0.858

0.857

0.882

0.861

0.797

0.783

0.815

0.781

Zhujiang

0.955

0.633

–

0.925

0.908

0.892

0.805

0.650

0.555

0.827

0.728

App_0

0.926

0.881

0.905

–

0.954

0.935

0.928

0.824

0.767

0.903

0.880

App_1

0.793

0.866

0.740

0.954

–

0.943

0.949

0.768

0.835

0.866

0.855

App_2

0.795

0.821

0.764

0.938

0.950

–

0.937

0.820

0.813

0.892

0.881

App_3

0.662

0.880

0.591

0.844

0.957

0.922

–

0.754

0.828

0.831

0.843

PB0

0.789

0.788

0.776

0.847

0.842

0.878

0.847

–

0.830

0.886

0.849

PB1

0.677

0.714

0.644

0.733

0.853

0.830

0.854

0.879

–

0.872

0.888

PD0

0.830

0.711

0.816

0.884

0.873

0.883

0.851

0.722

0.636

–

0.932

PD1

0.581

0.787

0.566

0.730

0.848

0.853

0.837

0.457

0.454

0.921

–

  1. The bold denotes the best segmentation result of T from S