Automated segmentation of mandibular bones in computed tomography (CT) scans is crucial for craniomandibular surgeries. Since traditional manual or semi-automated methods face challenges due to labor intensiveness and interobserver variability, deep learning, notably the U-Net architecture, offers a promising solution by accurately delineating anatomical structures in CT scans. This study compared a traditional 2D U-Net and U-Net Transformer model in implementing automated mandible segmentation of head and neck CT scans. The U-Net transformer model exhibited slightly higher Dice similarity coefficient scores and Intersection over Union (IoU) values compared to the U-Net model, indicating improved segmentation efficiency and spatial overlap with ground truth annotations. Additionally, the U-Net Transformer model also achieved superior pixel classification accuracy compared to the 2D U-Net model. These findings suggest that the U-Net Transformer model can significantly enhance surgical planning and diagnosis in craniofacial surgeries, highlighting the potential of Transformer-based deep learning architectures in automating mandibular segmentation from CT scans.