Qayyum Abdul, Lalande Alain, Meriaudeau Fabrice
ImViA Laboratory, University of Bourgogne Franche-Comté, Dijon, France.
ImViA Laboratory, University of Bourgogne Franche-Comté, Dijon, France; Medical Imaging Department, University Hospital of Dijon, Dijon, France.
Comput Biol Med. 2020 Dec;127:104097. doi: 10.1016/j.compbiomed.2020.104097. Epub 2020 Oct 28.
Automatic segmentation on computed tomography images of kidney and liver tumors remains a challenging task due to heterogeneity and variation in shapes. Recently, two-dimensional (2D) and three-dimensional (3D) deep convolutional neural networks have become popular in medical image segmentation tasks because they can leverage large labeled datasets, thus enabling them to learn hierarchical features. However, 3D networks have some drawbacks due to their high cost of computational resources. In this paper, we propose a hybrid 3D residual network (RN) with a squeeze-and-excitation (SE) block for volumetric segmentation of kidney, liver, and their associated tumors. The proposed network uses SE blocks to capture spatial information based on the reweighting function in a 3D RN. This study is the first to use an SE residual mechanism to process medical volumetric images using the proposed 3D residual network composed of various combinations of residual blocks. Our framework was evaluated both on the Kidney Tumor Segmentation 2019 dataset and the public MICCAI 2017 Liver Tumor Segmentation dataset. The results show that our architecture outperforms other state-of-the-art methods. Moreover, the SE-RN achieves good performance in volumetric biomedical segmentation.
由于肾脏和肝脏肿瘤的异质性和形状变化，在计算机断层扫描图像上进行自动分割仍然是一项具有挑战性的任务。最近，二维（2D）和三维（3D）深度卷积神经网络在医学图像分割任务中变得很流行，因为它们可以利用大量带标签的数据集，从而能够学习分层特征。然而，3D网络由于其计算资源成本高而存在一些缺点。在本文中，我们提出了一种带有挤压激励（SE）块的混合3D残差网络（RN），用于肾脏、肝脏及其相关肿瘤的体积分割。所提出的网络使用SE块基于3D RN中的重加权函数来捕获空间信息。本研究首次使用SE残差机制，利用由残差块的各种组合构成的所提出的3D残差网络来处理医学体积图像。我们的框架在2019年肾脏肿瘤分割数据集和公开的2017年医学图像计算方法国际会议（MICCAI）肝脏肿瘤分割数据集上都进行了评估。结果表明，我们的架构优于其他现有最先进的方法。此外，SE-RN在体积生物医学分割中取得了良好的性能。
