Automated Cavity Detection and Classification Using Deep Learning.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Deep Learning for Diagnostics

Background:

• Accurate dental cavity detection is crucial for early diagnosis and treatment planning.
• Traditional deep learning methods often use binary classification on whole images, missing localized details.
• There is a need for improved AI models that can analyze dental X-rays at a finer granularity.

Purpose of the Study:

• To develop and evaluate a multi-scale AI-assisted approach for dental cavity classification and detection.
• To compare the performance of image-level versus tooth-level analysis using the Ultralytics YOLOv11 framework.
• To assess the effectiveness of AI in localized cavity detection within dental radiography.

Main Methods:

• Utilized the Ultralytics YOLOv11 framework for object detection and classification tasks.
• Developed and compared two primary methodologies: image-level (panoramic) and tooth-level analysis.
• Evaluated model performance using metrics such as accuracy, mean Average Precision (mAP@50), and recall on a test dataset.

Main Results:

• The single-tooth classification model achieved a test accuracy of 0.854, with panoramic classification slightly higher at 0.864.
• For cavity detection, the tooth-level model demonstrated superior performance with an mAP@50 of 0.845, significantly outperforming the panoramic approach (0.669).
• The tooth-level model achieved a notably higher recall (0.743 vs. 0.394), indicating better localization of cavities.

Conclusions:

• Tooth-level AI analysis offers significant advantages for accurate dental cavity detection compared to image-level approaches.
• The study highlights the trade-offs between segmentation-based and direct image-based AI methods in dental diagnostics.
• Future research should focus on refining segmentation techniques and validating models across diverse clinical imaging conditions.