An effective teeth recognition method using label tree with cascade network structure.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate teeth detection and classification in dental radiographs are crucial for medical treatment and forensic identification.
• Traditional methods face challenges due to the similarity of some teeth and the limited availability of medical data for training deep learning models.

Purpose of the Study:

• To develop an effective deep learning technique for teeth detection and classification in dental periapical radiographs, addressing data scarcity.
• To improve the accuracy of identifying all 32 teeth positions, even in complex cases with missing or altered teeth.

Main Methods:

• A novel method utilizing a label tree to decompose the classification task and a cascade network structure composed of convolutional neural networks was proposed.
• Key strategies were employed to enhance detection and classification performance, specifically designed to handle limited training data.

Main Results:

• The proposed method achieved high overall precision (95.8%) and recall (96.1%) on a small training dataset.
• This performance significantly surpasses directly training a state-of-the-art 33-class convolutional neural network.

Conclusions:

• The developed deep learning approach effectively addresses the challenge of limited data in medical imaging for teeth detection and classification.
• The method demonstrates robust performance in complex clinical scenarios, offering a significant advancement for dental diagnostics and identification.