Classification of space maintainer requirement in pediatric panoramic dental images using deep learning

Space maintainers are intraoral appliances used to prevent loss of space in the dental arch after early deciduous tooth extraction. The timely and correct application of space maintainers guides the dental development process of the child and allows the permanent teeth to erupt in the ideal position. However, a detailed evaluation of biological and anatomical factors is required for the correct indication of a space maintainer. This process, which traditionally relies on the physician's assessment, is open to subjectivity and may result in erroneous decisions. In order to overcome this problem, a deep learning based approach has been developed. Within the scope of the study, a special dataset of panoramic X-ray images labeled by an expert pedodontist was created and the need for a space maintainer for each tooth was classified. The images were trained for object detection and classification using the YOLOv5 algorithm. The model was trained for 50 epochs with a training-validation ratio of 80%-20% on data labeled with MakeSense.AI. The results showed 82% accuracy in the "placeholder_not_necessary" class and 96% accuracy in the "placeholder_necessary" class. The F1-Confidence curve showed an optimum performance of 0.85 with a confidence threshold of 0.403 and a mAP@50 value of 0.931. During the training process, loss values decreased and precision-recall metrics remained stable in the range of 0.85-0.90.