Developing and validating a combined model with CT-based paraspinal muscle radiomics and clinical features to predict residual low back pain after percutaneous kyphoplasty.

Abstract

To develop and validate a CT-based radiomics model for predicting long-term residual low back pain (RLBP) in patients with osteoporotic lumbar vertebral compression fractures (LVCFs) following percutaneous kyphoplasty (PKP). This study enrolled 385 patients with LVCFs who underwent PKP. Patients were randomly allocated to a training cohort (n = 269) and a testing cohort (n = 116). Radiomic features were extracted from preoperative CT scans of the paraspinal muscles. Candidate predictors, including demographics (age, gender, BMI), clinical factors (Oswestry Disability Index [ODI], smoking status, fracture history), and radiological measures (mean CT value, muscle area, fat infiltration fraction [FIF]), were analyzed. Three models were developed: a clinical model derived from multivariable logistic regression, a radiomics model constructed from the Rad-score, and a combined model incorporating both. Model performance was evaluated with the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). Four independent clinical predictors were identified: ODI score, mean CT value, FIF, and multi-level fractures. The combined model showed the best performance, with an AUC of 0.905 in the training cohort and 0.891 in the testing cohort. In the testing cohort, the AUC of the combined model was significantly higher than that of the clinical model (0.891 vs. 0.787, P = 0.044) and showed a trend toward superiority over the radiomics model (0.891 vs. 0.828, P = 0.066). Calibration curves and DCA confirmed the combined model's good calibration and clinical utility. A nomogram was created to facilitate clinical use. A combined model incorporating clinical risk factors and paraspinal muscle radiomics features can accurately and robustly predict the risk of long-term RLBP after PKP. The model shows promise as a noninvasive preoperative tool to aid in individualized risk stratification and clinical decision-making.