Guideline proposes AI tool for scoliosis measurement with cross-center validation

This guideline presents a cross-center external validation study of SpineNET, a lightweight, interpretable keypoint-based AI framework built on the YOLOv13 paradigm for measuring Cobb angles in adolescent idiopathic scoliosis. The validation involved 1,200 radiographs for training/validation and 150 independent external test radiographs, with the AI tool compared against consensus measurements from three senior spine surgeons. The guideline focuses on the technical performance of this measurement-assisted tool rather than clinical outcomes or treatment recommendations.

Key findings from the validation include a Cobb angle mean absolute error (MAE) of 1.44°, an intraclass correlation coefficient (ICC) of 0.970, and minimal bias of 0.03°. The framework also demonstrated high object detection performance with mAP@50 of 94.2% and mAP@50-95 of 85.7%. These metrics suggest the AI tool can provide measurements comparable to expert surgeon consensus in this validation setting.

The authors note the tool could serve as an efficient and clinically interpretable measurement-assisted tool for high-throughput adolescent idiopathic scoliosis screening and follow-up in resource-constrained settings. However, the guideline does not report on safety outcomes, adverse events, or tolerability. Funding sources and conflicts of interest are also not reported, which limits assessment of potential biases.

For clinical practice, this guideline presents preliminary validation data for an AI measurement tool rather than treatment recommendations. The findings suggest potential utility for radiographic measurement assistance, but clinicians should interpret these results cautiously as they represent technical validation rather than clinical outcome evidence. Implementation would require consideration of local validation, integration with clinical workflows, and ongoing monitoring of measurement accuracy across diverse patient populations.