Purpose: We aimed to develop and externally validate a tool for predicting short-term functional outcome after lumbar fusion surgery.
Methods: Data of 1520 patients underwent lumbar fusion from three institutions was analyzed. A total of 855 and 1251 radiomics features from paraspinal muscles were extracted from preoperative CT and MRI scans, respectively. Multivariable logistic regression was used to identify independent risk factors of poor functional status after surgery. We developed and externally validated a combined model by integrating radiomics score and clinical features. We evaluated the clinical utility and stability of the model using decision curve and calibration curve analysis. SHAP plot was used for interpretation of predictive results.
Results: At multivariable analysis, radiomics score and 4 clinical features were identified as independent risk factors of poor functional outcome, and then a combined model was generated. This model had excellent performance, with AUCs of 0.85(95 %CI, 0.81-0.88), 0.82(95 %CI, 0.77-0.84), 0.79(95 %CI, 0.73-0.84) and 0.80(95 %CI, 0.76-0.83) in the derivation dataset and three independent test datasets, respectively. Moreover, this model showed great calibration and utility, outperforming the clinical model and radiomics score alone (both p < 0.05).
Conclusion: The combined model allows for accurate prediction of functional outcome after lumbar fusion surgery. The model could guide clinical decisions about the necessity of surgery for potential functional recovery.
Keywords: Functional outcome; Lumbar fusion; Machine learning model; Radiomics.
Copyright © 2024. Published by Elsevier B.V.