Abstract

We will develop an AI-based tool to automatically assess the quality of root canal treatments (RCTs) using intraoral periapical (IOPA) radiographs. Our solution aims to address the subjectivity and inconsistency in manual diagnosis by dental professionals.

The tool will detect treated teeth, evaluate treatment adequacy (e.g., underfilled, overfilled, voids, separated instruments), and visually mark key anatomical features such as the root apex and pulp floor. We expect this system to assist clinicians in making reliable decisions while also providing a standardized platform for training and research.

Methodology

Our project consists of the following key components:

Data Collection

We will collect a dataset of 1,000 anonymized IOPA radiographs from the Faculty of Dental Sciences.

Annotation

Dental experts will annotate regions including treated teeth, canals, crowns, root apices, and potential voids.

Preprocessing

Images will be enhanced using contrast normalization and artifact reduction techniques.

Model Architecture

Treated tooth localization using object detection (e.g., CenterNet). Root canal region segmentation using transformer-based models.

Labeling & Classification

Using descriptive prompts and classifiers to determine RCT quality (e.g., overfilled, underfilled, etc.).

Methodology Pipeline Diagram

Overview of the methodology pipeline

Experiment Setup and Implementation

  • We will use pre-trained models like BiomedCLIP for zero-shot classification and MedSAM for guided segmentation.
  • Initial implementation will focus on binary classification: treated vs untreated.
  • We will simulate and visualize feature embeddings using t-SNE/UMAP.
  • Classifier accuracy will be benchmarked against expert assessments.
  • For segmentation, we will evaluate precision, recall, IoU, and Dice score.

Results and Analysis

We will evaluate:

  • Accuracy of treated tooth detection (expected ≥90%)
  • Classification performance across RCT conditions (overfilled, underfilled, voids, etc.)
  • Effectiveness of visual overlays for clinical interpretability
  • Feedback from dental experts via a usability study

Conclusion

We aim to deliver a reliable AI-assisted diagnostic tool to support root canal treatment evaluation. Our project will improve diagnostic consistency, aid dental training, and pave the way for future integration with electronic dental records and mobile diagnostic apps.

Publications

Publications will be added here as the project progresses.

Team

Team Members

Chandula Adhikari
E/19/008
e19008@eng.pdn.ac.lk
Kaumini Adikari
E/19/009
e19009@eng.pdn.ac.lk
Sandeep Dassanayake
E/19/063
e19063@eng.pdn.ac.lk

Supervisors

Dr. I. Nawinne
Supervisor
isurunawinne@eng.pdn.ac.lk
Prof. R.D. Jayasinghe
Supervisor
rdjayasinghe@dental.pdn.ac.lk
Dr. R.M.S.G.K. Rasnayake
Supervisor
rmsgk@dental.pdn.ac.lk
Prof. R. Ragel
Supervisor
rragel@eng.pdn.ac.lk
Prof. M.C.N. Fonseka
Supervisor
mcnf@eng.pdn.ac.lk
Dr. P.A. Hewage
Supervisor
pahewage@eng.pdn.ac.lk