Breaking the silence: AI’s contribution to detecting vertebral fractures in opportunistic CT scans in the elderly—a validation study

Abstract

Vertebral fractures frequently go undetected in clinical practice. AI-assisted detection on CT scans demonstrates considerable promise, with a sensitivity of 86% and a specificity of 99%. The performance varied based on sex, and CT kernel, showing superior results in females and in scans using non-bone kernel protocols. PURPOSE: Vertebral fractures (VFs) are highly underdiagnosed, necessitating the development of new identification methods for opportunistic screening in computed tomography (CT) scans. This study validated an AI algorithm (ImageBiopsy Lab [IBL], FLAMINGO) for detecting VFs in a geriatric cohort, with various subgroup analyses including different CT protocols. METHODS: The performance of the AI in detecting VFs was compared to assessments by two experienced radiologists. A total of 246 thoracic or abdominal CT scans, primarily conducted for purposes other than skeletal examination, were included in the study. RESULTS: The patients had a mean age of 84 years (range 62 to 103), with 42% being female. The AI demonstrated high accuracy (0.93), sensitivity (0.86), and specificity (0.99) in detecting moderate to severe VFs. Subgroup analysis revealed accuracy ranging from 0.88 to 0.96, with higher accuracy in females compared to males (0.96 vs. 0.89, p = 0.03) and in scans performed with non-bone kernel versus bone kernel protocols (0.96 vs. 0.88, p = 0.02). No significant differences were found for age, contrast phase, or spinal region. CONCLUSION: The results indicate that the AI algorithm exhibits high performance in a geriatric setting. If effectively integrated with a fracture liaison service, this could enhance VF detection considerable in the future.

Ähnliche Arbeiten