Skip to main content

Table 4 T-stage errors by category

From: T-staging pulmonary oncology from radiological reports using natural language processing: translating into a multi-language setting

Error group Error type Description Training (n = 200) Validation (n = 225)
Data selection Sectionizer Detects information in wrong subheadings 1 3
Missing blacklist synonyms Falsely matched/falsely not excluded 0 5
Context Context missing Context not matched because of missing modifier 1 0
Context mismatch Context mismatch, wrong modifier detected 1 3
Concept matching Measurement extractor e.g., using abbreviations (e.g., (AP) × (TVR) × (SI)) 1 2
Complexity T4 multiple lobes 2 1
Ambiguity Confusion between node and mass (specific site: hilar) 4 7
Nonspecific 4 9
Missing concepts synonyms Lobulated 1 0
Cystic 2 0
Pleural thickening 1 0
Spinal metastasis 1 0
Costal involvement 0 1
Supraclavicular extension 0 1
Reporter Wrong input Different sizes for the same tumor, no unit (mm/cm) present, size for tumor and atelectasis 7 2
Satellite node 1 1
Total errors    27 35
  1. T-stage errors by category for training and validation sets