Skip to main content

Table 2 Accuracy stratified by prediction probability

From: Deep learning to convert unstructured CT pulmonary angiography reports into structured reports

Prediction probability threshold

Number of statements

Accuracy by strict criteria

Accuracy by modified criteria

0.1

4,157

3806/4157 (91.6%)

3986/4157 (95.9%)

0.15

4,155

3806/4155 (91.6%)

3986/4155 (95.9%)

0.2

4,154

3806/4154 (91.6%)

3986/4154 (96.0%)

0.25

4,154

3806/4154 (91.6%)

3986/4154 (96.0%)

0.3

4,151

3805/4151 (91.7%)

3985/4151 (96.0%)

0.35

4,143

3802/4143, (91.8%)

3982/4143 (96.1%)

0.4

4,127

3794/4127 (91.9%)

3974/4127 (96.3%)

0.45

4,112

3786/4112 (92.1%)

3966/4112 (96.4%)

0.5

4,094

3779/4094 (92.3%)

3958/4094 (96.7%)

0.55

4,069

3771/4069 (92.7%)

3945/4069 (97.0%)

0.6

4,034

3761/4034 (93.2%)

3934/4034 (97.5%)

0.65

4,009

3747/4009 (93.5%)

3920/4009 (97.8%)

0.7

3,961

3711/3961 (93.7%)

3881/3961 (98.0%)

0.75

3,925

3693/3925 (94.1%)

3854/3925 (98.2%)

0.8

3,886

3667/3886 (94.4%)

3824/3886 (98.4%)

0.85

3,858

3651/3858 (94.6%)

3806/3858 (98.7%)

0.9

3,810

3623/3810 (95.1%)

3773/3810 (99.0%)

0.95

3,715

3546/3715 (95.5%)

3689/3715 (99.3%)

1.0

3,262

3187/3262 (97.8%)

3259/3262 (99.9%)