摘要
医学图像分割是将图像中所关注的区域分割出来,以帮助医生诊断治疗。目前,深度学习技术已被广泛应用于医学图像分割[1]。文章的网络模型Grnet利用迁移学习方法,将VGGNet19-1结构与改进后的残差块结构resdiual-block1组合,构建了一个新的病理图像细胞核分割网络。最后得出Grnet在MoNuSeg测试集上的平均dice系数为74.86%,在TNBC数据集上的平均dice系数为74.74%,而Unet网络在MoNuSeg测试集上的平均dice系数为72.83%,在TNBC数据集上的平均dice系数为60.51%。实验结果表明,该网络的细胞核分割效果更好,充分验证了算法的有效性。
Medical image segmentation is to segment the area of concern in the image to help doctors diagnose and treat.Currently,deep learning technology has been widely used in medical image segmentation[1].Our network model,Grnet,uses a transfer learning approach to combine the VGGNet19-1 structure with the modified residual block structure,resdiual-block1,to construct a new nucleus segmentation network for pathological images.It can be seen from experiments that Grnet has average dice coefficient of 74.86%,average dice coefficient and 74.74% on the TNBC dataset,while Unet network has 72.83% average dice coefficient on the MoNuSeg test set and average dice coefficient of 60.51% on the TNBC dataset.Experimental results show that the nucleus segmentation of this network was better and fully validate the algorithm.
作者
胡志敏
陈小辉
HU Zhimin;CHEN Xiaohui(College of Computer and Information Technology Engineering,China Three Gorges University,Yichang Hubei 443002,China)
出处
《长江信息通信》
2022年第12期31-34,共4页
Changjiang Information & Communications