Related Experiment Video
Updated: May 31, 2026
09:10
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
PRAD++: Toward Robust Periapical Radiograph Analysis Through Dataset and Model Advancements
IEEE Transactions on Medical Imaging
|May 29, 2026
Summary
This study introduces PRAD++, a large dataset for dental periapical radiograph analysis, and PRNet++, a deep learning model that achieves state-of-the-art performance in segmentation and classification tasks.
Area of Science:
- Dentistry
- Medical Imaging
- Artificial Intelligence
Background:
- Deep learning (DL) models require large annotated datasets, but high-quality periapical radiograph (PR) datasets are scarce due to annotation costs and image quality issues.
- PRs are crucial in endodontics, yet limited data hinders DL development for PR analysis.
- Existing DL models struggle with the complexities of PR interpretation.
Purpose of the Study:
- To address the scarcity of annotated PR data by introducing PRAD++, a large-scale dataset.
- To develop an advanced DL model, PRNet++, for end-to-end PR analysis.
- To improve the accuracy, robustness, and clinical interpretability of DL models in PR analysis.
Main Methods:
- Created PRAD++, a dataset with 10,000 PR images featuring 9 pixel-level segmentation categories and 17 image-level classification labels, annotated by clinical experts.
- Developed PRNet++, an end-to-end network utilizing Multi-scale Wavelet Convolution (MWCN) and Channel Fusion Attention (CFA) for multi-scale feature integration.
- Incorporated an Expert Prior Injection (EPI) loss to integrate domain-specific dental knowledge, linking segmentation to classification for enhanced accuracy.
Main Results:
- PRNet++ achieved an average Dice Similarity Coefficient (DSC) of 81.25% for segmentation on the PRAD++ dataset.
- The model obtained macro- and micro-averaged PR-AUCs of 66.58% and 79.10% for classification, outperforming state-of-the-art methods.
- PRNet++ demonstrated superior robustness and interpretability, particularly in challenging clinical categories, validated by ablation and visualization studies.
Conclusions:
- PRAD++ and PRNet++ provide a valuable resource and a powerful tool for advancing DL-based dental image analysis, specifically for periapical radiographs.
- The proposed PRNet++ architecture, with its MWCN, CFA, and EPI loss, effectively addresses the limitations of existing models.
- The results highlight the potential of integrating expert knowledge into DL models for improved clinical applicability and interpretability in dentistry.
