Skip to main content
. Author manuscript; available in PMC: 2022 Apr 30.
Published in final edited form as: Proc Mach Learn Res. 2021 Dec;158:196–208.

Table 9:

Detailed F1-evaluation of RadBERT-CL variant (Algorithm 3) for the classification tasks of positive extraction, negation extraction, uncertainty extraction, and blank for each of our 14 observations. Note that for ”Blank”, we have f1-scores related to positive extraction and blank, while the other two are set to zero.

Category Positive F1 Negation F1 Uncertain F1 Blank F1
Enlarged Cardiomediastinum 0.579 0.786 0.831 0.965
Cardiomegaly 0.870 0.862 0.433 0.978
Lung Opacity 0.820 0.200 0.512 0.910
Lung Lesion 0.777 0.571 0.211 0.983
Edema 0.913 0.901 0.745 0.993
Consolidation 0.909 0.824 0.876 0.997
Pneumonia 0.786 0.916 0.807 0.991
Atelectasis 0.962 0.444 0.874 0.999
Pneumothorax 0.850 0.971 0.526 0.996
Pleural Effuison 0.938 0.957 0.596 0.985
Pleural Other 0.623 0.234 0.114 0.981
Fracture 0.894 0.333 0.667 0.993
Supported Devices 0.902 0.100 0.000 0.942
No Finding 0.592 0.000 0.000 0.978