Automated Tooth Labeling for Kennedy’s Classification in Panoramic Radiographs

Overview

This retrospective study evaluates an automated method using Mask R-CNN for tooth detection, labeling, and segmentation on panoramic X-rays to determine Kennedy’s classification of partially edentulous jaws. The approach integrates instance segmentation with post-processing and rule-based classification, aiming to support prosthodontic treatment planning.

Background

Panoramic X-rays provide a comprehensive 2D overview of dental structures and are widely used in dental diagnostics. Accurate tooth numbering using the FDI system is essential for communication and prosthodontic planning, with Kennedy’s classification categorizing partially edentulous jaws based on abutment teeth distribution. Manual tooth labeling and classification are error-prone, especially under heavy workloads, motivating the use of AI techniques such as convolutional neural networks. Mask R-CNN is a state-of-the-art method capable of simultaneous detection, classification, and segmentation of overlapping teeth in panoramic images.

Data Highlights

A total of 209 panoramic X-rays from patients aged 18 to 65 were annotated by experienced clinicians, labeling teeth, implants, and prosthetic statuses according to the FDI system and Kennedy’s classification. The dataset was split into five folds for training and evaluation of the Mask R-CNN model. Training utilized NVIDIA GTX 1080/2080 Ti GPUs with stochastic gradient descent and data augmentation including horizontal flips. Post-processing steps included filtering by confidence score, duplicate position removal, and geometric overlap checks to refine predictions.

Key Findings

• Mask R-CNN successfully performed instance segmentation and labeling of teeth in panoramic radiographs, even with overlapping structures.
• Post-processing algorithms improved the accuracy of tooth detection by removing duplicates and low-confidence predictions.
• The automated system enabled determination of Kennedy’s classification for upper and lower jaws based on detected abutment teeth.
• Exclusion of implants, pontics, and root residues from training data focused the model on relevant abutment teeth for prosthodontic planning.
• The study used a rigorous annotation and peer review process to ensure high-quality ground truth data for model training and evaluation.

Clinical Implications

Automated tooth labeling and Kennedy’s classification from panoramic X-rays can reduce human error and workload in prosthodontic diagnostics. Integration of AI-based segmentation supports consistent and efficient treatment planning by providing objective and reproducible dental status assessments. This approach may facilitate communication between clinicians and dental technicians by standardizing tooth identification and classification.

Conclusion

The study demonstrates that Mask R-CNN combined with post-processing can accurately detect, label, and segment teeth in panoramic radiographs to automatically determine Kennedy’s classification. This method holds promise for enhancing prosthodontic diagnostics and workflow efficiency.

References

1. Khurshid et al. 2023 -- Automated Kennedy’s Classification
2. Kennedy 1925 -- Classification of Partially Edentulous Arches
3. He et al. 2017 -- Mask R-CNN
4. Fraunhofer MEVIS -- CuraMate Data Curation Platform