VALSE 论文速览 第159期：动态蛇形卷积用于管状结构分割

论文题目：

Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation

作者列表：

戚耀磊 (东南大学)、何宇霆 (东南大学)、戚晓明 (东南大学)、张媛 (东南大学)、杨冠羽 (东南大学)

B站观看网址：

https://www.bilibili.com/video/BV1BT4y1s7A6/

论文摘要：

Accurate segmentation of tubular structures, like blood vessels and roads, is essential for various applications, ensuring precision in downstream tasks. However, this task is complex due to the presence of slender local structures and variable global morphologies. To address this, we introduce DSCNet, which enhances perception in three stages: feature extraction, feature fusion, and loss constraint. In feature extraction, dynamic snake convolutions emphasize slender and tortuous local structures to capture tubular features accurately. For feature fusion, a multi-view approach integrates information from different perspectives to retain essential details across various morphologies. A continuity constraint loss, based on persistent homology, enforces topological continuity

参考文献：

[1] Yaolei Qi, Yuting He, Xiaoming Qi, Yuan Zhang, Guanyu Yang, “Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation,” IEEE/ CVF Conference on International Conference on Computer Vision 2023 (ICCV2023).

论文链接：

[https://arxiv.org/abs/2307.08388]

中文翻译版本链接：

[https://yaoleiqi.github.io/publication/2023_ICCV/DSCNet_Chinese.pdf]

代码链接：

[https://github.com/YaoleiQi/DSCNet/]

视频讲者简介：

Yaolei Qi is a Ph.D. student at Laboratory of Image Science and Technology, Southeast University, Nanjing, China. He received his B.E. degrees in Southeast university, Nanjing, China in 2019. His research lies at the intersection of deep learning and artificial intelligence in medical image processing with a special focus on dealing with the tough task in tubular structure analysis that are knowledge-driven and data-efficient. His interests include innovative network design, weakly-supervised learning and computer-assisted diagnosis. He has published articles in journals and conferences like IEEE TIP, ICCV, MICCAI.

个人主页：

https://yaoleiqi.github.io/

特别鸣谢本次论文速览主要组织者：

月度轮值AC：李爽 (北京理工大学)