Skip to main content

Table 2 Evaluation results for the classification of normal and abnormal findings extracted from radiological reports

From: Semantic representation of reported measurements in radiology

  Oncologic data set Mixed data set
All cases 420 393
True normal 115 86
True abnormal 290 279
False normal 2 2
False abnormal 2 1
Unclassified 11 25
Accuracy 96.4 % (405/420) 92.9 % (365/393)
  1. The oncologic data set contains reports from lymphoma patients; the mixed data set contains reports from diverse medical and surgical patients. ‘True normal’ means that the algorithm correctly classified a finding as normal (accordingly for ‘true abnormal’). ‘False normal’ means that the algorithm falsely classified a finding as normal while the radiologist classified it as abnormal (accordingly for ‘false normal’). The term ‘unclassified’ means that the algorithm was not able to classify the finding