Teeth Segmentation in Panoramic Dental X-ray Using Mask Regional Convolutional Neural Network

Background and purpose: Accurate instance segmentation of teeth in panoramic dental X-rays is a challenging task due to variations tooth morphology overlapping regions. In this study, we propose new algorithm, for instance, the different X-rays. Methods: An model was trained using architecture Mask Region-based Convolutional Neural Network (Mask-RCNN). The data training, validation, testing were taken from Tuft database (1000 radiographs). number predicted label 52 (20 deciduous 32 permanent). size test sets 760, 190, 70 images, respectively, split performed randomly. 300 epochs, batch 10, base learning rate 0.001, warm-up multistep scheduler (gamma = 0.1). Data augmentation by changing brightness, contrast, crop, image size. percentage correctly detected Dice set used as quality metrics model. Results: set, classified 98.4%, while score 0.87. For both left mandibular central lateral incisor permanent teeth, index result 0.91 accuracy 100%. right first molar, second third indexes 0.92, 0.93, 0.78, with an 100% all three teeth. incisor, canine, molar 0.89, 0.91, 0.85, Conclusions: A successful identification X-ray developed validated. This may help speed up automate tasks like counting identifying specific missing improving current clinical practice.