基于多尺度上下文特征学习的胆管及胆结石图像分割

Bile Ducts and Hepatoliths Segmentation via Multi-Scale Context Feature Learning

肝胆管结石病是外科的常见病、多发病,实现对CT序列图像中的胆管及胆结石的自动化精准分割具有重要的意义。本文提出一种基于改进卷积神经网络的医学图像分割网络,充分考虑了CT切片间的上下文信息。网络由两对编码器-解码器组构成,包含多尺度上下文特征提取和多尺度特征融合等过程。多尺度卷积可以捕捉胆管及结石的局部细节信息和全局位置信息;ConvLSTM引入相邻切片间的上下文信息以提升分割精度。实验结果表明,该算法在胆管及胆结石的分割任务上具有较好的性能。

Hepatolithiasis is a common and frequently-occurring disease in surgery,and it is of great significance to realize automatic and accurate segmentation of bile ducts and gallstones in CT sequence images.This paper proposes a medical image segmentation network based on an improved convolutional neural network,which fully considers the context information between CT slices.The network consists of two pairs of encoder-decoder groups,including processes such as multi-scale context feature extraction and multiscale feature fusion.Multi-scale convolution can capture local detail information and global position information of bile ducts and stones and ConvLSTM introduces context information between adjacent slices to improve segmentation accuracy.Experimental results show that the algorithm has better performance in segmentation tasks of bile ducts and gallstones.