Development and Validation of a Nonradiographic Deep Learning Model for Skeletal Maturity Assessment in Iraqi Children Using Smartphones

Introduction: We aimed to develop and validate a nonradiographic deep learning model to assess skeletal maturity among Iraqi children using smartphones. Methods: We followed 499 Iraqi children (ages 8–15), collecting five smartphone images of their hands and teeth alongside tabular data points. Our “Stacking_v4” model (a hybrid of convolutional neural networks and XGBoost) was trained to sort participants into three growth spurt stages, using the middle phalanx of the third finger method as our ground truth for comparison. Results: The model achieved 96.00% accuracy (95% confidence interval: 90.1%–98.9%) on the test set. F1 scores for the various growth stages (ranging from 0.949 to 0.968) suggest the model is quite steady across the board. Interestingly, the data showed a consistent 96.00% accuracy for both boys and girls, which may indicate gender fairness. Conclusion: While more testing is likely needed, this approach appears to offer a safe, low-cost alternative for Iraqi orthodontists. Using accessible technology like smartphones might bridge the gap in clinical decision-making without traditional radiation risks.