Deep Learning-Based Segmentation of Periapical and Periodontal Regions in Radiographic Images with U-Net

The paper introduces a U-Net-based system for periapical and periodontal dental tooth segmentation to enable precise biological width measurements. The Faculty of Dentistry at MSA University generated 2,158 preprocessed periapical images and 598 preprocessed periodontal images that were further augmented through processing to make the model more stable. The U-Net model achieved excellent performance indicators during its 100 epoch training with a learning rate of 1 × +10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−7</sup> as it produced training accuracy (0.9952), Dice coefficient (0.9740), IoU (0.9493), precision (0.9745), recall (0.9743) and loss value at (0.0260). All validation metrics showed an accuracy result of 0.9683 with associated Dice coefficients o f 0 .8250 a nd Intersect over Union scores of 0.7036. The successful performance of the model indicates potential future deployment for biological width detection through Mask R-CNN which will improve periodontal condition diagnosis and treatment strategies.