A tooth and root canal panoramic image segmentation method based on ResNet50
By using the DeepLabV3+ model based on ResNet50 for panoramic image segmentation of teeth and root canals, the accuracy and efficiency problems in the diagnosis of pulp and periapical diseases in existing technologies are solved, and high-precision automatic semantic segmentation of teeth and root canals is achieved, which is suitable for real-time clinical auxiliary diagnosis.
Patent Information
- Application Number
- CN202610445884.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-07
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for diagnosing pulp and periapical diseases rely on the experience of clinicians. Two-dimensional panoramic radiographs are easily affected by anatomical structures, making it difficult to accurately identify the number, location, and morphology of root canals. Traditional algorithms are sensitive to noise and have low recognition accuracy. Existing deep learning methods are inefficient and lack pre-trained model optimization.
We used the DeepLabV3+ model based on ResNet50 for panoramic image segmentation of teeth and root canals. Through data augmentation and class weight optimization, we used the pre-trained ResNet50 backbone network for automatic semantic segmentation of teeth and root canals. We also combined the ASPP module to capture multi-scale contextual information and alleviate the gradient vanishing problem.
It improves the accuracy and efficiency of tooth and root canal segmentation, enabling more precise clinical auxiliary diagnosis with a segmentation accuracy rate of 93.7%, and is suitable for the identification of complex tooth overlap and root canal curvature.
Smart Images
Figure CN122416010A_ABST