Predicting postoperative coronal imbalance in Lenke 1/2 adolescent idiopathic scoliosis: A machine learning model with clinical interpretability

Objectives Selective posterior thoracic fusion (sPTF) for Lenke 1/2 adolescent idiopathic scoliosis (AIS) aims to reconcile multi-planar correction with motion preservation. Nevertheless, postoperative coronal imbalance (CIB) frequently compromises these objectives. This study developed an interpretable machine learning architecture to stratify CIB risk and identify key predictors. Methods Data from 282 patients were analyzed. Following dual-stage dimensionality reduction (Boruta and LASSO) on 24 candidate predictors, ten machine learning architectures were trained and evaluated using split-sample internal validation. Model efficacy was evaluated via area under the curve (AUC), Brier score, and decision curve analysis, using SHapley Additive exPlanations (SHAP) framework for algorithmic transparency. Results The LightGBM model demonstrated favorable performance, achieving peak AUC of 0.885 (training) and 0.824 (internal validation). SHAP identified three key predictors: the spatial relationship between the lowest instrumented vertebra and the last substantially touching vertebra (LIV-LSTV), regional lumbar adaptability (Lumbar Modifier), and skeletal maturity (Risser grade). Lower LIV-LSTV values, Lumbar Modifiers C, and lower Risser grades were associated with a higher predicted risk of CIB. Conclusion This framework incorporates information on distal instrumentation selection, lumbar curve morphology, and skeletal maturity. The model may provide a preliminary basis for future preoperative CIB risk estimation and support a more informed assessment of the trade-off between deformity correction and preservation of lumbar motion segments. Independent external validation is required before the model can be used to guide surgical decision-making.