Fig 1.
Graphical outline of the methods.
Purple rectangle shapes correspond to methods subsections, meaning they represent steps in the processing workflow, orange parallelogram shapes represent data, blue diamond shapes represent binary decisions on data, gray rectangle shapes represent excluded data, and green isosceles trapezoid shapes correspond to subcomponents of the algorithm.
Fig 2.
First example of the application of lexical targets and lexical modifiers.
Fig 3.
Second example of the application of lexical targets and lexical modifiers.
Table 1.
Radiology report characteristics.
Table 2.
Physician annotator comparison.
Fig 4.
Diagram depicting the overlap in annotations for two annotators (P1, P2) and tbiExtractor (ALG) in validation dataset. Dashed line indicates gold standard (i.e., where two annotators are in agreement).
Table 3.
tbiExtractor performance metrics.
Fig 5.
tbiExtractor output annotations.
tbiExtractor output annotations for the 27 selected lexical targets over the entire corpus (N = 417 radiology reports).
Table 4.
tbiExtractor lexical target errors.
Fig 6.
Divergent annotation examples.
Two examples of divergent annotations between annotators and tbiExtractor.