基于改进Mask R-CNN的牙齿识别与分割-《中国医学物理学杂志》

文章信息/Info

Author(s):: ZHAO Shuxu; LUO Qing; WANG Xiaolong; School of Electronics and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China

Keywords:: Keywords: tooth recognition tooth segmentation Mask R-CNN skip-connection structure SE module

摘要:: 针对当前的研究方法在牙齿全景X光片上提取的信息较为单一，而未曾考虑将牙齿的类别信息与形状位置信息融合提取的问题，提出一种实例分割方法同时实现牙齿识别与分割。主要通过融合跳跃结构和SE（Squeeze and Excitation）模块对Mask R-CNN实例分割模型中的分割分支进行改进，并以牙齿功能与FDI牙位两种类别编码方式，采用400张牙齿全景X光片数据进行实验仿真。实验结果表明改进后的模型相比于其他模型，可以同时有效地进行牙齿分类和分割，实现牙齿类别、形状、位置信息的融合提取，改善了Mask R-CNN实例分割模型在分割分支中语义信息提取不足的问题。

Abstract:: Abstract: In view of the fact that the information extracted from dental panoramic X-ray image by current research methods is relatively fewer, and the problem of without considering the fusion extraction of classification information and shape and position information of teeth, an instance segmentation method is proposed to realize the teeth recognition and segmentation simultaneously. The segmentation branch of Mask R-CNN instance segmentation model is improved by fusing skip-connection structure and SE (Squeeze and Excitation) module. Two teeth coding methods of tooth function and FDI tooth position were used for the experimental simulation on 400 dental panoramic X-ray images. The experimental results show that compared with other models, the improved model can effectively complete teeth classification and segmentation at the same time, realize the fusion extraction of tooth category, shape and position, and overcome the problem of insufficient semantic information extraction in the segmentation branch of Mask R-CNN instance segmentation model.