AI-Based Opportunistic CT Risk Assessment Using TotalSegmentator in Osteoporotic Vertebral Fractures.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology and Nuclear Medicine, University Hospital Basel and University of Basel, Petersgraben 4, 4031, Basel, Switzerland.
- Department of Endocrinology, Diabetology and Metabolism, University Hospital Basel, Basel, Switzerland.
- Department of Radiology and Nuclear Medicine, University Hospital Basel and University of Basel, Petersgraben 4, 4031, Basel, Switzerland. [email protected].
Abstract
Osteoporotic vertebral fractures impair quality of life and increase both morbidity and mortality, yet they are largely preventable. Routine CT examinations offer an opportunity for early risk stratification. This study aimed to evaluate whether fully automated analysis of routine lumbar CT examinations integrating vertebral fracture detection with vertebral attenuation and paraspinal muscle metrics improves identification of patients with osteoporotic vertebral fractures. This retrospective single-center study included lumbar spine-covering CT examinations acquired between January 2020 and May 2025. Of 3299 examinations identified in the PACS, contrast-enhanced studies and examinations with bone metastasis, haematological malignancy, severe artefacts or incompatible imaging parameters were excluded using a rule-based NLP pipeline, a large language model and manual review, resulting in 1209 examinations for analysis. Automated segmentation of vertebrae and paraspinal muscle groups was performed using TotalSegmentator to extract volumes and mean attenuation values. An in-house height-based algorithm automatically detected vertebral fractures, which were reviewed by a radiology resident under supervision of a board-certified radiologist to establish the reference standard (Genant grade ≥ 1). Group comparisons used nonparametric tests. Discriminatory performance was assessed using ROC analysis and an L1-penalized logistic regression model combining bone and muscle features. At least one lumbar vertebral fracture was present in 678 of 1209 examinations (56.1%). Automated fracture detection achieved an accuracy of 90.9% at the examination level. Mean attenuation of non-fractured lumbar vertebrae was significantly lower in examinations with fractures than in those without fractures (85.8 HU vs 133.8 HU, p < .001). Attenuation alone yielded an AUC of 0.73 for fracture discrimination. A multiparametric model integrating vertebral attenuation with paraspinal muscle volume and attenuation improved discrimination significantly (AUC 0.83, sensitivity 58%, specificity of 95%). In conclusion, automated opportunistic analysis of routine CT scans integrating vertebral fracture detection with bone and muscle biomarkers improves identification of patients with osteoporotic vertebral fractures beyond attenuation-based assessment alone.