Automatic Extraction of Major Osteoporotic Fractures from Radiology Reports using Natural Language Processing

Abstract

In this study, we developed a rule-based natural language processing (NLP) algorithm for automatic extraction of six major osteoporotic fractures from radiology reports. We validated the NLP algorithm using a dataset of radiology reports from Mayo Clinic with the gold standard constructed by medical experts. The micro-averaged sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score of the proposed NLP algorithm are 0.796, 0.978, 0.972, 0.831, 0.874, respectively. The highest F1-score was achieved at 0.958 for the extraction of proximal femur fracture while the lowest was 0.821 for the hand and finger/wrists fracture. The experimental results verified the effectiveness of the proposed rule-based NLP algorithm in the automatic extraction of major osteoporotic fractures from radiology reports.