AI may help clinicians identify vertebral fractures on CT scans with accuracy comparable to experienced specialists, according to new research.
Why Detecting Vertebral Fractures Matters
Vertebral fractures are a common and serious complication of osteoporosis, a condition in which bone density lowers, making bones weaker and more fragile. As the structure of the bone deteriorates, fractures can occur more easily, particularly in the spine. However, these fractures are frequently missed on CT scans, partly because it can be difficult to distinguish true osteoporotic fractures from other forms of vertebral height loss.
Identifying vertebral fractures is important because they can signal underlying osteoporosis and an increased risk of future fractures. Yet diagnostic accuracy can vary depending on the experience of the reader interpreting the scan.
To investigate whether deep learning algorithms could improve detection and grading, researchers compared eight human raters with four deep learning models and one commercial AI-based software tool. The human raters differed in expertise, including three medical students, three radiology residents and two attending clinicians.
Comparing AI with Clinicians
The retrospective analysis used two publicly available Vertebral Segmentation datasets. In total, 3,548 thoracic and lumbar vertebrae from 331 patients were evaluated.
Fractures were graded using the Genant semiquantitative scale, which classifies vertebral deformities from grade 0 (normal) to grade 3 (severe). Among the vertebrae assessed, 190 (5.4%) were fractured and 139 (3.9%) were classified as moderate or severe fractures, considered the most clinically relevant due these usually being associated with greater structural compromise, risk of disability and future fractures.
Deep learning models performed similarly to radiology residents in detecting moderate or severe fractures in the thoracic and lumbar spine. Diagnostic accuracy for these fractures was 0.988 for the AI models and 0.991 for residents, with no statistically significant difference.
Are AI Tools Moving Closer to Clinical Practice?
The findings suggest that AI-assisted CT analysis could help improve the identification and grading of vertebral fractures, particularly when algorithms are specifically trained to detect this condition on large, diverse datasets. It is also unknown whether the public CT datasets truly reflect the range of real-world imaging conditions, such as variable image quality, patient movement or unusual anatomy.
More consistent detection could support earlier recognition of osteoporosis and allow clinicians to intervene sooner to reduce the risk of future fractures.
While further research in clinical settings will be needed, the results indicate that deep learning tools are moving closer to potential integration into routine imaging workflows, where they could support clinicians in detecting vertebral fractures that might otherwise be overlooked.
Reference
E O Riedel et al. Diagnostic accuracy of deep learning vs. human raters for detecting osteoporotic vertebral compression fractures in routine CT scans. Eur Radiol. 2026;DOI:10.1007/s00330-026-12393-y.
Featured Image: Samunella on Adobe Stock
