Evaluation of Document-Level Identification of Pulmonary Nodules in Radiology Reports Using FLAIR Natural Language Processing Framework

Incidental radiographic findings are commonly found in radiology reports, can have significant clinical implications, yet frequently are missed or not followed-up on due to clinician error. This work evaluates a document-level classification task of radiology reports for a type of incidental finding, pulmonary nodules, utilizing the FLAIR NLP framework as a proof-of-concept for potential automation of identification of such findings for more consistent tracking.