基于U-Net融合Transformer的肺结节分割方法研究
Research on Pulmonary Nodule Segmentation Method Based on U-Net Combined with Transformer
李晓东 1丁鹏2
作者信息
- 1. 山东中医药大学智能与信息工程学院,山东 济南 250355
- 2. 山东中医药大学第二附属医院 后勤党总支,山东 济南 250001
- 折叠
摘要
目的 提出肺结节分割模型,实现肺结节分割.方法 在U-Net神经网络中加入编码器、空洞卷积以及Swin Transformer模块,提出一个将空洞卷积、编码器和注意力机制相结合的模型,并在LUNA16公共数据集上验证模型性能.结果 改进的模型在LUNA16公共数据集上进行肺结节分割的准确度(Accuracy,ACC)、特异性(Specificity,SP)、交并比(Intersection Over Union,IOU)和Dice系数(Dice Similarity Coefficient,DSC)分别为0.9651、0.9572、0.8354、0.8971.结论 该分割模型在ACC、SP、IOU和DSC方面表现优异,可辅助医生诊断,在临床肺结节分割方面具有一定的参考价值.
Abstract
Objective To propose a pulmonary nodule segmentation model to realize pulmonary nodule segmentation.Methods The encoder,hole convolution and Swin Transformer module were added to the U-Net neural network,and a model combining hole convolution,encoder and attention mechanism was proposed,and the performance of the model was verified on LUNA16 public data set.Results The accuracy(ACC),specificity(SP),intersection over union(IOU)and Dice similarity coefficient(DSC)of the segmentation results of this model on LUNA16 public dataset were 0.9651,0.9572,0.8354 and 0.8971,respectively.Conclusion The segmentation model has an excellent performance in ACC,SP,IOU and DSC,can assist doctors to diagnose,and has certain reference value in clinical pulmonary nodule segmentation.
关键词
肺结节/肺结节分割/U-Net神经网络/Swin/Transformer模块Key words
pulmonary nodule/pulmonary nodule segmentation/U-Net neural network/Swim Transformer module引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据