An Artificial Intelligence Model for the Automatic Classification of the Cervical Vertebral Maturation Stages

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiographic Analysis
• Orthodontic Diagnostics

Background:

• Cervical Vertebral Maturation (CVM) staging is crucial for assessing skeletal growth.
• Manual CVM assessment from lateral cephalometric radiographs (LCRs) can be subjective and time-consuming.
• Developing automated methods for CVM classification is essential for improving efficiency and objectivity.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) model for automated CVM stage classification on LCRs.
• To compare the performance of the AI model against human assessments.
• To establish a reproducible framework for objective skeletal maturation assessment.

Main Methods:

• A two-stage pipeline was employed using 1140 LCRs.
• The YOLOv11s.pt model localized the region of interest (ROI) for CVM analysis.
• The ResNet-101 deep learning model was trained on extracted ROIs for CVM stage classification.

Main Results:

• The ResNet-101 deep learning model achieved an overall accuracy of 84% for CVM stage classification.
• High agreement was observed between AI and human readings, with quadratic weighted kappa of 0.93 and ICC values of 0.95.
• Intra- and inter-examiner reliability for human assessment was high (weighted kappa: 0.992 and 0.953).

Conclusions:

• The developed AI model provides an accurate and objective method for automated CVM stage classification.
• This AI-based framework enhances the objectivity and efficiency of skeletal maturation assessment.
• The integration of YOLOv11 and ResNet-101 offers a novel, automated workflow for CVM analysis.