Hierarchical Deep Learning Model for Tooth Classification in Intra-Oral Dental Photographs

Articles in Press

Document Type : Original Article

Authors

1 Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, Malaysia

2 Faculty of Computer Science and Information Technology, Universiti Tun Hussein Onn Malaysia, Malaysia

3 Department of Agricultural Engineering, Bahauddin Zakariya University, Pakistan

4 Dental Simulation and Virtual Learning Research Excellence Consortium, Department of Dental Science, Advanced Medical and Dental Institute, Universiti Sains Malaysia, Malaysia

Abstract

Tooth classification in dental informatics presents numerous challenges due to the geometrical and texture distribution complexity. This article presents a hierarchical deep learning (HDL) model for tooth classification that leverages contextual and spatial information in intra-oral dental photographs. Discriminative deep feature blocks are employed hierarchically in the HDL model: YOLOv6, YOLOv5-based CBAM, and YOLOv5-based hybrid pooling technique. The model operates on dental photographs captured from various viewpoints and comprises three primary levels: level-0, level-1, and level-2 corresponding to two, four, and seven-tooth classifications, respectively. The hierarchical structure allows the sequential extraction of contextual information, improving robustness and feature quality. The refined features are consolidated into a single, comprehensive representation. This consolidated representation encapsulates the essential characteristics of the individual tooth, making it highly effective for tooth classification tasks. The proposed HDL model significantly outperforms the non-hierarchical standalone models in the test set, achieving an overall mAP, recall, and F1-score of 88%, 0.999, and 0.94% respectively. The most significant improvement is notably observed in classifying occluded and less visible teeth like the first and second Molar tooth class. The HDL model can be integrated into educational software to provide in-depth information about the morphology of teeth.

Keywords

Main Subjects

International Journal of Engineering

Articles in Press, Accepted Manuscript
Available Online from 02 April 2024

Files

Share

How to cite

Statistics