基于Swin Transformer和UNet的肺结节分割方法

Nodule Segmentation Method Based on Swin Transformer and UNet

肺结节的准确分割是后续良恶性分析和诊断的关键。由于基于卷积神经网络的分割模型受限于局部特征提取特性,忽略了全局特征。因此,本文提出了一种新的肺结节语义分割框架ST-UNet网络,将Swin Transformer嵌入UNet中,构成一种新颖的Swin Transformer和CNN并行的双编码器结构。结果表明:该模型不仅对肺结节的分割具有较好的性能,而且对医生进行肺结节的早期诊断具有重要的临床意义和应用价值。

Accurate segmentation of pulmonary nodules is the key to subsequent benign and malignant analysis and diagnosis.Because the segmentation model based on convolutional neural network is limited by local feature extraction,the global feature is ignored.Therefore,this paper proposes a new semantic segmentation framework for pulmonary nodules ST-UNet network,and emparts Swin transformer into UNet to form a novel dual encoder structure of Swin Transformer and CNN in parallel.The results show that this model not only has a good performance in the segmentation of pulmonary nodules,but also has important clinical significance and application value for doctors in the early diagnosis of pulmonary nodules.