Artificial Intelligence vs Human opportunistic Detection of Vertebral Fractures in Routine CT Scans: results of a pilot study

Purpose Vertebral fractures (VFs) are the most common osteoporotic fractures and a hallmark of bone fragility, yet most morphometric VFs remain undiagnosed in routine imaging not focused on the spine. This pilot study evaluated the diagnostic performance of an artificial intelligence (AI) system for opportunistic vertebral fracture detection in chest and abdominal computer tomography (CT) scans compared with human readers in a real-world hospital setting. Methods Over two months, all thoracic and abdominal CTs performed for any indication in patients aged ≥ 50 years at a tertiary hospital were analyzed by the Bone Solution HealthOST AI platform (Nanox AI Ltd.). Routine radiology reports and targeted reviews by endocrinologists were compared to AI outputs, using a gold-standard adjudication by an expert panel. Sensitivity, specificity, predictive values, and odds ratios (ORs) were calculated and compared by χ² tests (p < 0.05). Results Among 427 eligible CT scans, AI detected vertebral fractures with significantly higher sensitivity (86.3%) than radiologists (50.0%) or endocrinologists (68.8%) (p ≤ 0.01 for both comparisons). Compared with radiologists, AI had 14.2-fold greater odds of correctly flagging true fractures (OR = 14.23; 95% CI 4.7–43.1; p < 0.000001) while against targeted review by endocrinologists, AI achieved a 4.8-fold advantage (OR = 4.77; 95% CI 1.6–13.6; p = 0.001). Extrapolated annually, AI integration could uncover around 150 new, otherwise undiagnosed, patients eligible for osteoporosis treatment. Conclusion Automated vertebral fracture detection in routine CT scans significantly enhances diagnostic yield versus human interpretation. Its integration into clinical workflows offers a cost-free, high-impact strategy to improve early osteoporosis diagnosis and treatment within the healthcare system.