Comprehensive Summary
This retrospective study developed and validated a radiomics-based classification model using computed tomography (CT) images to distinguish fresh vs old vertebral compression fractures (VCFs). A cohort was divided into subgroups by compression grade (mild, moderate, severe) and morphology (wedge, biconcave, crush) and 1,834 features were extracted. On the full cohort the radiomics model achieved an AUC of 0.824, while a nomogram combining radiomics + clinical factors reached AUC 0.897. Performance varied markedly by subgroup: for the severe compression grade subgroup the AUC was 0.927; mild grade had AUC 0.633; moderate grade 0.774. By morphology, the crush-type subgroup achieved AUC 0.909, wedge 0.841, biconcave 0.897. The findings underscore that both compression grade and vertebral morphology significantly influence diagnostic performance of radiomics models.
Outcomes and Implications
Radiomics applied to CT enables non-invasive differentiation of fresh versus old VCFs, which is clinically relevant for treatment planning (e.g., deciding intervention vs conservative management). The variable performance by compression grade and morphology suggests that model deployment should account for fracture severity and type to avoid over- or under-estimating accuracy. Integrating clinical data with radiomics further enhances diagnostic utility. This could reduce reliance on MRI in settings where MRI is less available, expedite fracture-acuity assessment, and support orthopedic decision-making.