Study on strategy of CT image sequence segmentation for liver and tumor based on U-Net and Bi-ConvLSTM

作者：

Highlights：

• A branch structure is added on the skip connections of U-net to supplement features.

• Bi-ConvLSTM based post-processing for U-net leads to more accurate contours.

• Integrating sequence information into U-net achieves better coincidence degree.

• Sequence segmentation hardly increases model parameter number compared with U-net.

摘要

•A branch structure is added on the skip connections of U-net to supplement features.•Bi-ConvLSTM based post-processing for U-net leads to more accurate contours.•Integrating sequence information into U-net achieves better coincidence degree.•Sequence segmentation hardly increases model parameter number compared with U-net.

论文关键词：Deep learning,Sequence segmentation,U-net,Bi-directional convolutional long short-term memory,Liver tumor,CT image

论文评审过程：Received 30 November 2020, Revised 16 March 2021, Accepted 4 April 2021, Available online 22 April 2021, Version of Record 8 May 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115008