Automated Mandible Segmentation using ۳D-UNet for Virtual Surgical Planning: A Deep Learning Approach

Background and aims: Conventionally diagnosis and surgical planning for Mandibular orthognathicsurgeries such as tumor resections, displacements, and flap reconstruction of the mandiblewere based on two-dimensional images. Virtual surgical planning (VSP) is a modern approachthat enables surgeons to plan and simulate surgeries in a virtual environment before performingreal surgery. In VSP, accurate segmentation of anatomical structures, such as the mandible, iscrucial for creating accurate ۳D models that can be used for surgical planning. Mandible segmentationis a challenging task due to the complex shape and varying density of the bone. Traditionalsegmentation methods require manual intervention and are time-consuming and error-prone. Recently,deep learning methods have emerged as a promising approach for accurate and efficientmandible segmentation. Deep learning methods, such as convolutional neural networks (CNNs),have shown remarkable success in various medical imaging tasks, including segmentation. To thisend, accurate mandible segmentation using deep learning methods can facilitate virtual surgicalplanning by providing accurate ۳D models of the mandible. This research aims to implement۳D-Unet architecture for accurate and efficient mandible segmentation, which can improve theaccuracy of virtual surgical planning and ultimately improve patient outcomes.Method: In this study, we used ۳D mandible image data to train a ۳D-UNet for automatic mandiblesegmentation from CT images with the aim of facilitating virtual surgical planning (VSP).The dataset was taken from Public Domain Database for Computational Anatomy (PDDCA). Theaccuracy of this method was also assessed by using the DICE Score coefficient (DSC).Results: The proposed segmentation method outperformed the UNet and received a better Dicescore over the validation dataset.Conclusion: In this paper, we describe our proposed model for automatic mandible segmentationof CT images. Moreover, to evaluate the performance of our proposed method, a similar experimentwas performed with the U-Net. Experimental results demonstrate that our model has theaccurate fully automated segmentation of the mandible and high DSC compared to UNet.

کلیدواژه ها:

VSP ، ۳D-UNet ، Mandible segmentation ، Orthognathic surgeries