摘要
U-Net自诞生以来就在医学分割领域十分热门,尽管原生的U-Net在医学图像分割任务上已经有着非常不错的表现,但是在肝脏肝肿瘤分割任务上仍然有着改进空间。首先肝脏肝肿瘤分割任务中每张CT切片的肝脏和肝肿瘤的大小不一、形状各异,所以需要提取多尺度信息,而U-Net网络所有进行卷积操作的卷积核尺寸都是一样的,因此将金字塔卷积模块替换了传统卷积,以此达到提取多尺度信息的目的。由于有些切片中肝脏肝肿瘤和背景相比样本数量少得多,为此将交叉熵损失函数与Dice损失函数相结合来解决样本数量不平衡带来的问题。U-Net使用下采样操作,因此保证下采样时能保留更多有用的上下文信息至关重要,为此引入了CBAM注意力模块,它同时具备空间注意力和通道注意力。通过在LiTS2017数据集上的大量实验证明了提出模型的有效性。
U-Net has performed very well in the medical image segmentation,but there is still a room for its improvement in the liver segmentation and liver tumor segmentation. The size and shape of the liver and liver tumor in each CT slice in the liver and liver tumor segmentation are different,so it is necessary to extract multi-scale information,However,the convolution kernel size of all convolution operations in the U-Net network is the same. Therefore,the traditional convolution is replaced with the pyramid convolution module to achieve the purpose of extracting multi-scale information. Since the number of samples of liver and liver tumor in some sections is much fewer than those of background,the cross entropy loss function is combined with the Dice loss function to solve the problem caused by the imbalance of the number of samples. The downsampling operation is used in U-Net,so it is important to ensure that more useful context information can be retained during downsampling. Therefore,the CBAM attention module is introduced,which makes it have both spatial attention and channel attention. The effectiveness of the proposed model is proved by a large number of experiments on the LiTS2017 dataset.
作者
郭鹏
邵剑飞
GUO Peng;SHAO Jianfei(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)
出处
《现代电子技术》
2023年第5期85-88,共4页
Modern Electronics Technique
关键词
医学图像分割
肝脏分割
肝肿瘤分割
U-Net
特征提取
CBAM注意力
实验分析
medical image segmentation
liver segmentation
liver tumor segmentation
U-Net
feature extraction
CBAM attention
experimental analysis
