A segmentation network integrating edge attention mechanism and transfer learning for mandibular nerve segmentation in oral panoramic images
QIU Yuanlue
WANG Zhan
YANG Shuo
FENG Qianjin
LI Bingchun
LAI Mingwen
ZHAO Lei
WANG Yong
Abstract：Objective This study proposes a deep learning method that integrates edge attention mechanisms and transfer learning to improve segmentation accuracy in mandibular nerve segmentation from oral panoramic images,in order to address the problems of boundary ambiguity and limited training data.Methods The Edge-Guided Attention UNet(EGAUNet)was employed to enhance the ability to capture the edge details of the mandibular nerve in oral panoramic images.Additionally,a transfer learning strategy was integrated to effectively improve the segmentation accuracy under limited training data conditions.Results The method that integrate the edge attention mechanism and transfer learning,was validated using an internal dataset.The experimental results show that after pre-training with transfer learning based on the FIVES dataset,the method obtained the following segmentation performance metrics on the test set of oral panorama data developed in this study:a mean Dice coefficient of 0.711,a recall of 0.725,an intersection over union of 0.575,a positive predictive value of 0.707,and a 95%Hausdorff distance of 12.422.Conclusion The deep learning method that integrates edge attention mechanisms and transfer learning,significantly assist dentists in accurately localizing the mandibular nerve during procedures such as dental implant placement and tooth extraction,thereby reducing the risk of intraoperative nerve injury and enhancing surgical safety.
Keywords：mandibular nerve segmentationedge attention mechanismtransfer learningoral panoramic imagingdeep learning
Publication Date：2025-12-30
Online Publishing Date：2025-12-11(First online date of this platform, not the publication date of the document)
Pages：9( 572-580 )