Blending Parametric Model and Neural Refinement: A Coarse-to-Fine Approach for Predicting Facial Changes in Orthognathic Surgery

Jiawen Yang, Dongming He, Xuanyu Tian, Lei Ma, Jingyi Yu, Xudong Wang, Yuyao Zhang

Abstract: Accurate prediction of postoperative facial soft tissue changes is crucial for effective preoperative planning in orthognathic deformity surgery. In this paper, we propose a novel coarse-to-fine framework for predicting these changes. The coarse module employs the parametric statistical face model, SCULPTOR, to generate an initial prediction of the postoperative face. SCULPTOR effectively captures the bidirectional linear relationship between the outer facial surface and underlying bone structures, while modeling surgical deformations. In the fine module, the coarse predictions are first converted into UV maps, which are then refined using FocalNet as the facial feature extractor (FFE), incorporating a craniofacial transformation module (CTM) to capture the non-linear relationship between the skeleton and facial surface. Experimental results on patients undergoing orthognathic surgery demonstrate that our method surpasses state-of-the-art techniques in predicting postoperative facial outcomes.

External IDs:dblp:conf/isbi/YangHTMYWZ25