Evaluation of multi-task learning in deep learning-based positioning classification of mandibular third molars.

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Pell and Gregory, and Winter's classifications are essential for safe mandibular third molar extraction.
• Accurate classification of mandibular third molars aids surgical planning and risk assessment.
• Deep learning models offer potential for automating radiographic analysis.

Purpose of the Study:

• To evaluate the classification accuracy of convolutional neural network (CNN) deep learning models for mandibular third molars.
• To compare the diagnostic performance of single-task versus multi-task learning approaches.
• To assess the utility of VGG 16 model for classifying based on Pell and Gregory, and Winter's criteria.

Main Methods:

• A dataset of 1330 cropped panoramic radiographs of mandibular third molars was curated.
• Convolutional neural network (CNN) models, specifically VGG 16, were employed for classification.
• Single-task and multi-task learning strategies were compared using metrics like accuracy, precision, recall, F1 score, and AUC.

Main Results:

• Single-task learning demonstrated superior diagnostic accuracy compared to multi-task learning across all evaluated metrics (p < 0.05).
• Recall and F1 scores for tooth position classification showed moderate effect sizes in both single-task and multi-task learning.
• The study confirmed the efficacy of Pell and Gregory, and Winter's classifications when implemented within specific deep learning tasks.

Conclusions:

• Single-task deep learning models are highly effective for classifying mandibular third molars based on established radiographic criteria.
• CNNs, particularly the VGG 16 model, show promise in automating the classification of mandibular third molars.
• This research provides a foundation for utilizing AI in enhancing the safety and efficiency of third molar extraction procedures.