Deep Learning-Based Detection of Separated Root Canal Instruments in Panoramic Radiographs Using a U²-Net...

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Endodontic Treatment

Background:

• Separated endodontic instruments pose a significant challenge in root canal therapy, impacting treatment success and prognosis.
• Detecting these fragments on panoramic radiographs is difficult, especially in complex cases or for less experienced clinicians.

Purpose of the Study:

• To develop and validate a deep learning model, specifically the U 2 -Net architecture, for automated detection and segmentation of separated endodontic instruments.
• To assess the model's performance on panoramic radiographs acquired from diverse imaging systems.

Main Methods:

• Retrospective analysis of 36,800 panoramic radiographs, with 191 meeting inclusion criteria.
• Manual segmentation of separated instruments using the Computer Vision Annotation Tool.
• Training and evaluation of the U 2 -Net model using Dice coefficient, IoU, precision, recall, and F1 score.

Main Results:

• The U 2 -Net model achieved a high Dice coefficient (0.849) and IoU (0.790), indicating excellent segmentation accuracy.
• The model demonstrated strong performance with a precision of 0.877, recall of 0.847, and F1-score of 0.861.
• Robust performance was observed across radiographs from various imaging systems.

Conclusions:

• The U 2 -Net model shows high accuracy in segmenting separated endodontic instruments on panoramic radiographs.
• The model's consistent performance across different imaging systems suggests significant clinical utility for detection and treatment planning.
• Further multicenter validation is recommended to confirm the generalizability of the findings.