Automatic Segmentation of Teeth from Panoramic X-Ray Images Employing Deep Learning Models

Dental practitioners rely on panoramic radiographs for diagnosing impacted teeth, infections, and mouth/jaw cancers. The development of deep learning-based algorithms has enabled practical identification for evaluating a patient's oral health and aiding in effective treatment plans. There has not been enough work done to create collaborative models that improve learning performance by combining the strengths of existing models. Using seven different segmentation models (ResU-Net++, U-Net++, MultiResU-Net, DeepLabV3+, Poly-PVT, PraNet, and TransResU-Net), this paper presents a comprehensive analysis of teeth segmentation using deep learning models, comparing, and contrasting their features, strengths, datasets used, and future research directions. The TransResU-Net model outperforms state-of-the-art segmentation methods with an F1 score of 0.9266 (precision: 0.9235; recall: 0.9346; accuracy: 0.9721) and Jaccard index of 0.8658.