To evaluate the effectiveness of a machine-learning radiomics model derived from lumbar spine CT in detecting osteoporosis compared to traditional osteoporosis detection methods.
Key Findings:
The radiomics model outperformed vertebral bone quality scoring and Hounsfield unit measurements in detecting osteoporosis.
The XGBoost model achieved an area under the curve of 0.89 in the training cohort and 0.91 in the test cohort, demonstrating superior performance compared to traditional methods.
Sensitivity was approximately 89%, specificity was 82%, and negative predictive value was 94% in the test cohort.
Interpretation:
The radiomics-XGBoost model demonstrates superior predictive accuracy for osteoporosis detection and may serve as an effective complementary screening tool.
Limitations:
Retrospective, single-center design.
Modest sample size.
Use of a single CT scanner and acquisition protocol.
Inability to stratify patients with osteopenia, which may limit applicability.
Conclusion:
The radiomics-XGBoost model shows promise as a screening tool for osteoporosis, necessitating external validation in multicenter cohorts.
