Evaluation of partially edentulous jaws with artificial intelligence algorithms

Aim: Artificial intelligence (AI) has rapidly influenced various fields, including radiology and dentistry, enhancing diagnostic and planning processes. This study aims to perform automatic Kennedy classification of panoramic images from patients with missing teeth using deep convolutional neural networks (DCNN). Materials and Methods: A dataset of 1300 panoramic images was used to develop our model. We applied transfer learning techniques to the yolov5 architecture to classify missing teeth based on detection, numbering, and location, enabling accurate Kennedy classification. CranioCatch software was used for image labeling, and model success metrics were calculated through the complexity matrix method. Results: The yolov5 model achieved sensitivity, precision, and F1 scores of 0.85, 0.96, and 0.90, respectively, for missing tooth detection and enumeration. For Kennedy classification, these metrics were 0.91, 0.87, and 0.89, respectively. The model's overall success rate for Kennedy labeling of the maxilla and mandible was 89%, demonstrating robust performance. Conclusion: Accurate classification of toothed and edentulous jaws is essential in prosthetic planning. Automating this process through radiographic-based classifications and integrating these findings with intraoral scanner results can enhance the reliability of AI-driven prosthetic applications. This approach minimizes human error in the digital workflow, supporting autonomous or semi-autonomous planning. Our study significantly contributes to the development of future deep learning-based panoramic reporting systems in dentistry. We believe such systems will expand the use of artificial intelligence in clinical settings, fostering more reliable and consistent treatment planning. Keywords: Artificial intelligence, Kennedy classification, Artificial neural networks, Deep learning