基于改进U-Net网络的肺部CT图像分割算法研究

Research on lung CT image segmentation algorithm based on the improved U-Net network

针对肺部疾病的诊疗,从计算机断层扫描(Computed Tomography,CT)图像中自动检测肺部病灶,对监测疾病进展和进一步临床治疗具有重要意义。为此,基于U-Net网络,提出新型的肺部CT图像分割网络LG-Net。利用带有注意力模块的跳层连接对病灶区域提取边缘轮廓信息;其次,引入多级残差卷积与特征融合模块,弥补网络中的特征损失问题。在公开数据集上的实验表明,LG-Net提高了肺部CT图像的分割精度,相较于传统的U-Net算法具有更优异的分割性能。

For the diagnosis and treatment of lung diseases,automatic detection of lung infection from Computed Tomography(CT)images is of great significance in monitoring disease progression and further clinical treatment.Therefore,based on U-Net,a new lung CT image segmentation network LG-Net is proposed.Firstly,the edge contour information is extracted from the focal region by using the skip connection with the attention module.Secondly,the multi-level residual convolution and feature fusion modules are introduced to make up for feature loss in the network.Experiments on public datasets show that LG-Net improves the segmentation accuracy of lung CT images and has better segmentation performance than traditional U-Net algorithm.