Incremental Learning for Panoramic Radiograph Segmentation

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Dentistry

Background:

• Manual detection of dental and orthodontic anomalies is inefficient and error-prone.
• Automated methods are needed to improve accuracy and efficiency in dental diagnostics.

Purpose of the Study:

• To develop a fundamental method for automated detection and treatment of dental and orthodontic problems.
• To leverage deep learning for accurate anomaly detection in dental radiographs.

Main Methods:

• Utilized incremental learning approaches with Mask R-CNN as backbone networks.
• Employed transfer learning by incorporating newly annotated data to enhance model performance.
• Focused on automatically labeled data for model construction, addressing data scarcity.

Main Results:

• Achieved encouraging and adequate findings for filling and tooth segmentation tasks.
• Demonstrated improved model performance through transfer learning, despite limited radiograph data.
• Successfully constructed a more accurate model from automatically labeled data.

Conclusions:

• The proposed deep learning method, utilizing incremental learning and Mask R-CNN, offers a viable solution for automated dental anomaly detection.
• Transfer learning significantly enhances model accuracy, particularly in data-scarce environments common in dental imaging.
• The approach shows promise for improving the efficiency and reliability of diagnosing dental and orthodontic issues.