Pulp-Stone Detection in Panoramics Using Deep Learning:A Multi-Institutional Study
Büşra ÖZTÜRK, Ali Altındağ, Özer Çelik, İbrahim Şevki Bayrakdar, Kaan Orhan · International Dental Journal · 2024
Pulp stones are calcified masses of different sizes within the dental pulp cavity. This study aims to assess the efficacy of the YOLOv8 deep learning algorithm in accurately discerning the presence of calcifications within the pulp-chambers depicted in panoramic radiographs. A dataset of 1000 panoramic radiographic images was labeled using the CranioCatch program, an innovative tool developed in Eskişehir, Turkey. The dataset was partitioned into three subsets using a random allocation strategy: 80% for training, 10% for validation, and 10% for testing purposes. The images underwent a 3x3 clash enhancement operation aimed at improving the visibility and definition of the annotated regions. The YOLOv8 algorithm for pulp stone detection was trained by transfer learning method. The F1 score, sensitivity, and precision metrics for the artificial intelligence model, implemented with the YOLOv8 architecture, yielded in the evaluation of the test dataset, with values of 0.9087, 0.8726, and 0.9480, respectively. In the realm of panoramic radiography, deep learning-driven artificial intelligence algorithms have demonstrated notable efficacy, particularly in identifying pulp stones, showcasing remarkable success. Additionally, leveraging artificial intelligence-backed clinical decision support system software holds promise in enhancing the efficiency and efficacy of dental practitioners.