Development and validation of an MRI-based musculoskeletal model for predicting pedicle screw loosening after lumbar fusion surgery: a multicenter study

Objective: To investigate the utility of a preoperative risk assessment model based on magnetic resonance musculoskeletal metrics in predicting postoperative screw loosening. Methods: Data from 454 patients who underwent lumbar fusion with pedicle screw fixation for degenerative lumbar diseases at our hospital from January 2019 to February 2024 were obtained. The dataset was randomly split into internal training set (60%) and internal test set (40%). Screw loosening was assessed based on X-ray at the 12-month follow-up. Vertebral bone quality (VBQ) and pedicle bone quality (PBQ) scores were used to evaluate bone quality at the vertebral body and pedicle, respectively. Paraspinal muscle metrics, including relative cross-sectional area (rCSA), fat infiltration rate (FI%), and relative functional cross-sectional area (FrCSA), were also assessed. Multivariate forward logistic regression and correlation analysis were performed to determine the potential risk factors for screw loosening and to facilitate variable selection. Based on the results of variable selection, a simplified nomogram model was constructed, and its accuracy in predicting screw loosening was compared with that of lumbar T-scores and Hounsfield unit (HU) values. In addition, an external validation cohort consisting of 193 patients at another hospital between May 2020 and June 2024 were also included to enhanced the model’s generalizability. Results: Among 273 patients from the internal training set, 43 exhibited screw loosening at the one-year postoperative follow-up. Compared to patients without screw loosening, those with screw loosening had significantly higher proportion of malnutrition, higher VBQ scores, increased PBQ scores, elevated mean FI% of L1-L5, and lower L1-L4 HU values, worse bone mineral density, as well as lower mean FrCSA of L1-L5. PBQ and mean FI% of L1-L5 are the most critical variables associated with screw loosening. Based on PBQ and mean FI% of L1-L5, we constructed a simplified nomogram model. DeLong’s test indicated that the combined indices demonstrated superior diagnostic accuracy. Consistent findings were also observed in the internal test set and external validation cohort. Conclusion: Among patients undergoing lumbar fusion surgery, PBQ scores and mean FI% of L1-L5 were identified as the most critical parameters for predicting postoperative screw loosening. The MRI-based risk assessment model provides a valuable tool for the early evaluation of individualized risks of screw loosening.