肺部支气管分割的设计与改进:3D UX-Net

Design and improvement of pulmonary bronchial segmentation in the lung:3D UX-Net

为了实现肺部支气管的精确分割,并获取支气管的内径,提出了一种基于3D UX-Net模型的肺部支气管分割设计和后处理优化方法。首先,采用3D UX-Net模型训练支气管数据;其次,结合区域选择、阈值调整及最大连通域分析等提升分割精度的全新后处理(Reprocess)流程;最后,在私有数据集上与其他算法模型进行对比实验。实验结果表明:基于3D UX-Net模型的Dice系数评价指标从0.83提升到0.85,自制和临床数据集关于支气管内径的误差在0.2mm,得出本文提出的模型和后处理方法能够快速而准确地分割支气管。

To achieve accurate segmentation of the pulmonary bronchi and obtain the inner diameter of the bronchus,a pulmonary bronchi segmentation design and post-processing optimization method is proposed based on the 3D UX-Net model.First,a 3D UX-Net model is used to train bronchial data.Second,a new post-processing(Reprocess)process is combined with region selection,threshold adjustment and maximum connected domain analysis to improve segmentation accuracy.Finally,comparative experiments with other algorithm models are carried out on private datasets.The experimental results show that the evaluation index of the Dice coefficient based on the 3D UX-Net model is improved from 0.83 to 0.85,and the error of the inner diameter of the bronchus is 0.2mm based on homemade and clinical datasets.It can be concluded that the proposed model and post-processing method are able to segment the bronchi quickly and accurately.