Table 1 Performance in detecting PE on the RSPECT private test set

From: High performance with fewer labels using semi-weakly supervised learning for pulmonary embolism diagnosis

Slice Labels (%)

TP

FN

TN

FP

AUC

Acc

SEN

SEC

PPV

NPV

F1

0

351

87

437

569

0.682

0.546

0.801

0.434

0.382

0.834

0.517

2.5

313

125

871

135

0.858

0.820

0.715

0.866

0.699

0.874

0.707

5

305

133

912

94

0.878

0.843

0.696

0.907

0.764

0.873

0.729

10

344

94

860

146

0.900

0.834

0.785

0.855

0.702

0.901

0.741

20

327

111

955

51

0.916

0.888

0.747

0.949

0.865

0.896

0.801

27.5

344

94

954

52

0.928

0.899

0.785

0.948

0.869

0.910

0.825

35

351

87

942

64

0.927

0.895

0.801

0.936

0.846

0.915

0.823

42.5

364

74

915

91

0.935

0.886

0.831

0.910

0.800

0.925

0.815

50

356

82

942

64

0.928

0.899

0.813

0.936

0.848

0.920

0.830

75

367

71

921

85

0.934

0.892

0.838

0.916

0.812

0.928

0.825

100

350

88

939

67

0.932

0.893

0.799

0.933

0.839

0.914

0.819

  1. TP true positive, FN false negative, TN true negative, FP false positive, AUC area under the receiver operating curve, Acc accuracy, SEN sensitivity, SPEC specificity, PPV positive predictive value, NPV negative predictive value.