Abstract Purpose Radiation-free tools, such as scoliometers, ultrasound, and Moiré topography, have been explored for monitoring Adolescent Idiopathic Scoliosis (AIS), but none have replaced the need for serial spinal radiographs. This study aimed to evaluate the accuracy and criterion validity of a new AI-powered digital health application using 3D surface topography to predict Cobb angles, with the goal of reducing radiation exposure and enabling home-based curve monitoring. Methods In a single-center observational study, 125 patients with confirmed or suspected scoliosis underwent smartphone-based 3D surface-topography scans in standing and forward-bending positions. Poor-quality scans ( n = 20) were excluded. Radiographic Cobb angles were used as the gold standard. After random allocation, 79 scans formed the training set and 26 formed the validation set; external data (142 controls, 188 scoliosis patients) were added to strengthen training, and 25 controls were added to the test set. Accuracy was expressed as mean absolute error (MAE) and correlation with radiographs. Criterion validity was assessed by sensitivity, specificity, and AUC at clinically meaningful thresholds (10°, 25°, and 40°). Results Across 51 test scans (AIS + controls), the algorithm showed a strong correlation with radiographs ( r = 0.922, 95% CI 0.866–0.955) and an MAE of 5.9° (95% CI 4.5–7.3). In AIS-only curves of 10–50°, the MAE was 6.4° (95% CI 4.4–8.3). At 10°, sensitivity was 0.962, specificity was 0.960, and AUC was 0.978. At 25°, sensitivity was 0.706, specificity was 0.853, and AUC was 0.917. At 40°, sensitivity was 0.667, specificity was 1.000, and AUC was 1.000. Conclusion This AI-driven 3D surface topography demonstrated high validity for non-radiographic Cobb angle prediction, particularly in mild-to-moderate AIS. It supports safer, more frequent, home-based monitoring, though refinements are needed for severe curves and patients with a higher BMI.
Duval et al. (Fri,) studied this question.