AI meets endodontics a deep learning approach to precision diagnosis

Area of Science:

• Dentistry and Artificial Intelligence
• Medical Imaging Analysis
• Machine Learning in Healthcare

Background:

• Endodontic diseases affect over 52% of the global population, with increasing prevalence.
• Accurate classification of endodontic conditions is crucial for effective treatment planning.
• Conventional diagnostic methods rely on radiographs and expert analysis, necessitating more efficient and accurate automated alternatives.

Purpose of the Study:

• To develop and evaluate an automated, data-driven model for precise classification of endodontic diseases.
• To enhance diagnostic accuracy and efficiency in endodontic treatment planning.
• To introduce a Modified Swin Transformer (MSViT) architecture for hierarchical attention and contextual learning.

Main Methods:

• Implementation of a Modified Swin Transformer (MSViT) with hierarchical attention mechanisms.
• Utilizing chaotic particle swarm optimization (CPSO) and sequential quadratic programming (SQP) for hyperparameter and feature selection.
• Training and validation on an improved root canal dataset featuring seven categories of endodontic disorders.

Main Results:

• The proposed MSViT model achieved an average classification accuracy of 97.72%.
• The model demonstrated a mean precision of 0.9749 and a mean square error of 8.2 × 10⁻⁴.
• Optimized performance was achieved with a learning rate of 0.0001 and a computational time of 867.59 seconds.

Conclusions:

• The developed MSViT architecture effectively learns fine-grained anatomical and pathological features for endodontic disease differentiation.
• The automated approach offers a significant improvement in accuracy and efficiency compared to conventional methods and baseline models.
• This AI-driven framework holds promise for advancing clinical decision-making in endodontics.