基于三重注意力机制的新冠肺炎病灶分割模型

Segmentation Model of COVID-19 Lesions Based on Triple Attention Mechanism

为了解决感染区域比正常组织对比度低的问题,提出了一种基于三重注意力机制(triple attention mechanism,TAM)的新冠肺炎(corona virus disease 2019,COVID 19)病灶分割模型DDTM-Net,并将其应用于条件生成对抗网络。MultiConv模块可以自动提取肺部切片中感染区域的特征,呈现了更丰富且包含不同类型的病灶信息。TAM同时集成了空间、通道、位置注意力模块,可以更精准地定位感染区域的病灶。使用的损失函数是由3种不同的损失函数组成的复合函数,能最小化预测图和真实标签的差异,更好地优化TM-Net模型。在COVID-19数据集上进行实验和评估的结果表明:基于TM-Net的磨玻璃影(ground-glass opacities,GGO)和固结(Consolidation)两类病灶的平均dice相似系数(dice similarity coefficient,DSC)比基于Attention U-Net和R2U-Net的DSC分别提高了1.4%和0.5%,证明TM-Net提升了新冠肺炎病灶分割的准确性。

In order to solve the problem of low intensity contrast between infected areas and normal tissues,A corona virus disease 2019(COVID-19)segmentation model TMNet is proposed based on triple attention mechanism(TAM),and applied to conditional generative adversarial network in this paper.The MultiConv module in TM-Net can automatically extract rich features of infected areas in lung slices.These features contain different types of lesion information.The designed TAM,which integrates spatial,channel and positional attention modules,can accurately locate lesions in the infected area.By composing of three types of loss functions,the loss function of TM-Net can minimize the differences between prediction graphs and real labels,thus optimizing the TM-Net.Experiment and evaluations conducted on COVID-19 data sets show that the average dice similarity coefficient(DSC)of ground glass opacities(GGO)and consolidation of TM-Net are 1.4%and 0.5%higher than the results of attention U-Net and R2 U-Net,respectively,proving the accuracy improvement of TM-Net in COVID-19 lesions segmentation.