Development of a Combined Model for Adjacent Segment Disease Following Lumbar Spine Surgery Utilizing Preoperative CT and MRI

Objectives Lumbar fusion is among the most frequently performed spinal surgeries, with adjacent segment disease (ASD) being a prevalent postoperative complication. To date, no studies have simultaneously employed preoperative CT and MRI for predicting ASD following lumbar fusion. This study aims to investigate the utility of clinical indicators, conventional CT and MRI indicators, and unsupervised metrics derived from unsupervised machine learning algorithms based on CT and MRI in constructing a predictive model for ASD post-lumbar spine surgery. Methods A retrospective analysis was conducted on patients who underwent lumbar interbody fusion for lumbar degenerative disease at our institution between 2021 and 2023, with a minimum follow-up period of 24 months for both clinical and imaging assessments. Data collected included preoperative demographic and laboratory information, conventional CT and MRI indicators, as well as unsupervised metrics processed through different unsupervised machine learning algorithms (Gaussian mixture model, K-means, and Otsu) based on CT and MRI, which provided measures of mean, volume, and volume percentage of muscle and fat. This study compared the differences in these parameters between the ASD and non-ASD groups to identify independent predictors, which were then integrated with clinical indicators to develop a comprehensive predictive model. The predictive efficacy of the models was assessed and compared using the area under the curve (AUC) and the Delong test. Results A cohort of 375 patients was analyzed (17 patients in ASD group and 358 patients in non-ASD group). Only gender, sarcopenia and psoas major muscle index (PMI) were found to be correlated with ASD complications (all P  < 0.05). The AUC for the PMI was 0.662 (95% CI: 0.550–0.773). For unsupervised metrics, the highest AUC was derived from the CT-based Otsu model, with an AUC of 0.719 (95% CI: 0.615–0.824). The AUC of combined model was further enhanced to 0.812 (95% CI: 0.717–0.908). Conclusion The integration of clinical indicators, conventional CT and MRI indicators, and unsupervised metrics demonstrates potential for non-invasive prediction of ASD in patients who have undergone lumbar spine surgery.