Artificial intelligence was of limited use in the quality control of primary care chest radiographs

Background Chest radiograph is a commonly used imaging examination, but it is challenging to interpret. In primary care, a quality-controlling artificial intelligence (AI) could help detect missed pulmonary nodules that require further investigation. The aim of this study was to assess the usefulness of AI in this task, and to determine whether its use could reduce the number of missed potential early-stage lung cancers.

Methods In this study, a deep learning algorithm analyzed all chest radiographs and corresponding radiologists’ reports taken during the year 2021 in the primary care of the city of Oulu, Finland. The algorithm assessed if any pulmonary nodules had been missed by the reporting radiologist. With the help of an external radiologist and our own radiologist, we evaluated the clinical significance of the AI findings and the need for further investigation.

Results Out of nearly 10 000 images, AI helped to identify nine clinically significant findings. According to the AI analysis, 410 images contained a previously unreported nodule. Of these, 395 (96%) were clinically insignificant.

Conclusions AI detected more lung cancer-suspect findings than radiologists but due to its low specificity, substantial human input remained necessary for diagnosis.

Ilmari Alanampa, Tommi Keski-Filppula, Osmo Tervonen