loading page

An Automatic Annotation Model of Teeth and Dental Conditions in Dental Panoramic Radiographs: A Proposal of Segmentation Tool
  • Özer Çelik
Özer Çelik

Corresponding Author:[email protected]

Author Profile

Abstract

Purpose: The study aimed to design and propose an automatic annotation model of segmentation teeth in dental panoramic radiographs using deep-learning artificial intelligence (AI) algorithms. Methods: A total of 1183 anonymized panoramic radiographs were used to develop the AI model (CranioCatch, Eskisehir, Turkey) for an automatic annotation model of teeth and dental conditions including implant, crown, pontic, and filling in dental panoramic radiographs. U-Net implemented with PyTorch library was used for developing an automatic annotation model. The confusion matrix was used as a metric of the success of the AI model. Results: The sensitivity, precision, and F1-score values for teeth annotation at 70% intersection over union (IoU) value were 1, 0.9894, and 0.9947, respectively. At 50% IoU value, the sensitivity, precision, and F1-score values were found as 1, 1, 1, respectively. The sensitivity, precision, and F1-score values for dental condition annotation at 70% IoU value were 0.9963, 0.84, and 0.9115, respectively. At 50% IoU value, the sensitivity, precision, and F1-score values were found as 0.9963, 0.9446, 0.9698, respectively. Conclusion: To develop an artificial intelligence model for dental image classification, image segmentation, and object detection generally needs lots of labeled images. To overcome the struggles of manual annotation, the proposed annotation tool can be used to make it easier to increase the number of labeled data to further enhance the success of AI studies.