Radiation-free Cobb angle estimation in adolescent scoliosis using surface topography and a linear regression model.

Scoliosis, the most common spinal deformity in adolescents, requires frequent radiographic follow-up, exposing patients to cumulative ionizing radiation with potential long-term risks. In response, recent efforts have explored radiation-free alternatives for Cobb angle estimation, but most fail to meet the clinical threshold of a minimum significant change of 5°. In this study, we aimed to develop and internally validate a fully automated method for predicting Cobb angle using 3D surface topography (ST) data and a linear regression model (LRM). Principal component analysis was used to reduce the dimensionality of the ST data, and several machine learning models were compared, including neural networks, XGBoost and Stacking. The LRM showed the best performance in the test set, with a mean absolute error (MAE) of 3.97°, a root Mean square error (RMSE) of 4.70°, and a strong correlation with the ground truth (r = 0.91). Residual analysis confirmed normality and homoscedasticity, supporting the robustness of the model. Importantly, the MAE fell below the clinically significant threshold of 5°, indicating the model's ability to detect minimal but critical changes in spinal curvature. These results outperform most previous non-radiographic methods and suggest that a simple, interpretable LRM, combined with open-source tools and ST data, offers a viable and scalable solution for non-invasive scoliosis monitoring. If externally validated, this method could reduce reliance on x-rays, thereby reducing radiation exposure while maintaining assessment accuracy.