Segmentation of Pulp and Pulp Stones with Automatic Deep Learning in Panoramic Radiographs: An Artificial Intelligence Study.
Authors
Affiliations (6)
Affiliations (6)
- Department of Dentomaxillofacial Radiology, Faculty of Dentistry, Cyprus International University, Nicosia 99010, Cyprus.
- Department of Dentomaxillofacial Surgery, Faculty of Dentistry, Cyprus International University, Nicosia 99010, Cyprus.
- Department of Biomedical Engineering, Faculty of Engineering, Near East University, Mersin 10, Lefkoşa 99010, Turkey.
- Department of Management Information Systems, School of Applied Sciences, Cyprus International University, Nicosia 99010, Cyprus.
- Dentmetria A.Ş., İstanbul 34726, Turkey.
- Department of Dentomaxillofacial Radiology, Faculty of Dentistry, Ankara University, Ankara 06100, Turkey.
Abstract
<b>Background/Objectives</b>: Different sized calcified masses called pulp stones are often detected in dental pulp and can impact dental procedures. The current research was conducted with the aim of measuring the ability of artificial intelligence algorithms to accurately diagnose pulp and pulp stone calcifications on panoramic radiographs. <b>Methods</b>: We used 713 panoramic radiographs, on which a minimum of one pulp stone was detected, identified retrospectively, and included in the study-4675 pulp stones and 5085 pulps were marked on these radiographs using CVAT v1.7.0 labeling software. <b>Results</b>: In the test dataset, the AI model segmented 462 panoramic radiographs for pulp stone and 220 panoramic radiographs for pulp. The dice coefficient and Intersection over Union (IoU) recorded for the Pulp Segmentation model were 0.84 and 0.758, respectively. Precision and recall were computed to be 0.858 and 0.827, respectively. The Pulp Stone Segmentation model achieved a dice coefficient of 0.759 and an IoU of 0.686, with precision and recall of 0.792 and 0.773, respectively. <b>Conclusions</b>: Pulp and pulp stones can successfully be identified using artificial intelligence algorithms. This study provides evidence that artificial intelligence software using deep learning algorithms can be valuable adjunct tools in aiding clinicians in radiographic diagnosis. Further research in which larger datasets are examined are needed to enhance the capability of artificial intelligence models to make accurate diagnoses.