Radiomics classification of fresh and old vertebral compression fractures: Impact of compression grade and morphology on diagnostic performance.

OBJECTIVES

To develop a radiomics model for identifying fresh or old vertebral compression fractures (VCFs) from CT images, thereby assisting physicians in making more effective decisions.

METHODS

Patients with VCFs who underwent both CT and MRI within one week were retrospectively enrolled from June 2018 to February 2023. VCFs were categorized as subgroups according to compression grades (mild, moderate or severe) or morphology types (wedge-shaped, biconcave or crush). For each subgroup, a radiomics classification model was built based on 1834 radiomics features extracted from the training dataset. And the diagnostic performance was evaluated in the testing dataset using receiver operating characteristic (ROC).

RESULTS

The radiomics model trained on the entire cohort achieved an area under ROC curve (AUC) of 0.824. A nomogram integrating radiomics feature and clinical characteristics reached an AUC of 0.897. We graded the degree of compression as mild, moderate, and severe VCFs. The best performance was observed in the severe subgroup, with an AUC of 0.927, while the AUCs for mild and moderate were 0.633 and 0.774, respectively. In the morphology subgroups, the crush-type VCFs demonstrated the best performance, achieving an AUC of 0.909, while the AUCs for wedge-shaped and biconcave were 0.841 and 0.897, respectively.

CONCLUSION

The radiomics models effectively distinguished fresh and old VCFs, performing better when combined with clinical features. However, different grades and morphologies of VCFs showed distinct CT imaging patterns that could impact model performance, warranting consideration in future research and clinical applications.