iCVM: An Interpretable Deep Learning Model for CVM Assessment Under Label Uncertainty.

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical Vertebral Maturation (CVM) assessment is vital for orthodontic and orthopedic treatment planning.
• Accurate CVM determination aids in evaluating craniofacial skeletal development.

Purpose of the Study:

• To develop an automated deep learning model for Cervical Vertebral Maturation (CVM) assessment.
• To improve the accuracy and interpretability of CVM staging in orthodontic diagnostics.

Main Methods:

• A novel convolutional neural network, iCVM, based on residual networks was developed.
• Techniques including dropout layers and label distribution learning were employed to address data limitations and label ambiguity.
• Grad-CAM was utilized for model interpretability analysis.

Main Results:

• The iCVM model demonstrated superior performance on the new CVM-900 dataset across various metrics.
• Model predictions showed high consistency with established clinical criteria, indicating strong interpretability.
• The study released the CVM-900 dataset to facilitate future research.

Conclusions:

• Deep learning, specifically the iCVM model, offers a promising approach for automated and accurate CVM assessment.
• The interpretability of the model supports its clinical utility in orthodontic growth and development evaluations.
• The CVM-900 dataset provides a valuable resource for advancing research in automated CVM analysis.