Generative AI tested for lumbar osteoporosis screening
Researchers evaluated three multimodal AI models and 92 physicians classifying 90 lumbar scans against dual-energy X-ray absorptiometry T-score thresholds.
Research Square
In human imaging scans, researchers tested whether multimodal generative artificial intelligence models could screen for osteoporosis using routine spinal images. The preprint evaluated 45 lateral lumbar radiographs and 45 sagittal computed tomography images classified using dual-energy X-ray absorptiometry T-score thresholds of −1.0 and −2.5 as reference standards. Diagnostic performance was compared across 42 board-certified orthopedic surgeons, 50 non-specialist physicians, and three generative artificial intelligence models: ChatGPT, Gemini 3.1 Pro, and Claude 4 Opus. Each model evaluated every image five times under zero-shot and few-shot conditions using 20 labeled reference cases. The authors also tested combined specialist and model classifications using OR and AND rules. At the −1.0 cutoff, orthopedic specialists outperformed non-specialists on radiographs.
Why it matters
Osteoporosis is a common age-related skeletal condition that frequently remains undiagnosed until a fracture occurs. Opportunistic screening using lumbar images obtained for other indications could improve early detection in aging populations.
Caveats
This work is a preprint that has not yet undergone peer review. The evaluation relied on a small sample of 45 radiographs and 45 CT images.
The paper
SHOWA Medical University
Research Square · 29 Sep 2026 · Preprint, not peer-reviewed