一种基于TransUnet的臂丛神经超声图像分割网络

A Brachial Plexus Nerve Ultrasonography Segmentation Network Based on TransUnet

在医疗图像分割领域中,以臂丛神经(Brachial Plexus,BP)超声图像为例的部分超声图像中存在对比度低、边缘模糊和噪声多等问题,使得对目标区域的准确分割十分困难。为此,基于TransUnet网络框架将Transformer模块引入U-Net网络编码端,利用其自注意力机制更好地捕捉图像中的全局特征,提高模型的特征提取能力;同时将空洞卷积应用到网络的跳跃连接来增大感受野,降低特征图中的噪声影响,为解码端提供更显著的特征。实验表明,与传统的U-Net、SegNet以及基于Transformer的MedT(Medical Transformer)相比,设计的网络模型具有更高的Dice系数和IoU值,Dice系数较前三者最高提升了13.2%。

In the field of medical image segmentation,there are problems such as low contrast,blurred edges,and noise in some ultrasound images,such as the Brachial Plexus(BP)ultrasound images,which make accurate segmentation of the target area difficult.Therefore,in this paper,we introduce Transformer modules into the encoding end of the U-Net network based on the TransUnet framework,using its self-attention mechanism to better capture global features in the image and improve the model s feature extraction capabilities.At the same time,we apply dilated convolutions to the skip connections of the network to increase the receptive field,reduce noise in the feature map,and provide more prominent features for the decoding end.Experiments show that compared with traditional U-Net,SegNet,and MedT(Medical Transformer)based on Transformer,the network model designed in this paper has a higher Dice coefficient and IoU value,with the Dice coefficient improved by up to 13.2%compared to the three approaches mentioned before.