Development and validation of an artificial intelligence algorithm for cervical vertebral maturation staging using lateral cephalograms.

BACKGROUND

Cervical vertebral maturation (CVM) assessment using lateral cephalograms offers a reliable method for evaluating skeletal maturity without additional radiation exposure. However, traditional manual analysis is time-intensive and prone to variability. With the growing role of artificial intelligence in medical imaging, this study aimed to develop and validate a deep learning algorithm capable of automatically determining CVM stages from lateral cephalometric radiographs, improving diagnostic efficiency and accuracy.

METHODS

In total, 525 lateral cephalograms from individuals aged 7-17 years (249 males and 276 females; mean age, 12.67 years) were analyzed. An artificial intelligence-powered annotation platform, developed using PLAINSIGHT, was employed to identify 19 anatomical landmarks and perform 20 linear measurements on the C2, C3, and C4 vertebrae. The VGG19 convolutional neural network model was trained using 1300 augmented images generated from 420 original cephalograms. Model validation was performed on an independent dataset comprising 105 cephalograms.

RESULTS

The trained VGG19 model achieved an overall accuracy of 86% in CVM staging, with optimal performance observed between 80 and 100 training epochs. Confusion matrix analysis indicated the highest classification accuracy in CVS stages 4, 5, and 6. The model demonstrated an overall F1 score of 0.85, with the highest score in CVS6 (0.93) and the lowest in CVS1 (0.79), reflecting robust predictive capability across multiple maturation stages.

CONCLUSION

The VGG19-based deep learning model showed strong potential for automating CVM assessment using lateral cephalograms. Its high accuracy and reproducibility suggest its utility as a clinical decision-support tool for evaluating skeletal development in growing individuals.