您现在的位置:首页>新闻中心>科研动态

科研动态

【医学部学术讲座】——Volumetric ConvNets for Automated Segmentation from 3D MR Images

文章来源: 作者: 发布时间:2017年12月18日 点击数: 字体:

主讲嘉宾:Yu Lequan香港中文大学ph.D

时间: 2017年12月28日周四下午14:30-16:30

地点: 深圳大学西丽校区A6-811

主持人: 雷柏英 副教授

主讲嘉宾简介:

Yu Lequan is currently a third year Ph.D. student in the Department of Computer Science and Engineering, The Chinese University of Hong Kong, supervised by Prof. Pheng Ann Heng. His research interest includes deep learning and the application in medical image computing. He is also interested in deep reinforcement learning.

He received the B.Eng degree from Department of Computer Science and Technology in Zhejiang University in 2015. He also has internship in Siemens Healthcare, Medical Imaging Technologies, Princeton, USA.

报告内容简介:

Automated segmentation from 3D medical image is very important in clinical practice. However, these tasks are very challenging due to the limited training dataset in medical image domain and the complicated anatomical variations of organs. In this talk, I will introduce some works about using volumetric ConvNets for automatic 3D segmentation.

First, I will introduce the volumetric ConvNets with mixed residual connections to cope with the prostate segmentation problem. Compared with previous methods, our volumetric ConvNet has two compelling advantages. First, it is implemented in a 3D manner and can fully exploit the 3D spatial contextual information of input volumetric data to perform efficient, precise and volume-to-volume labeling. Second, the combination of mixed residual connections (i.e., long and short) can greatly improve the training efficiency and discriminative capability of our network by enhancing the information prorogation within the ConvNet both locally and globally. Then I will introduce the densely-connected volumetric ConvNet, referred as DenseVoxNet to automatically segment the cardiac and vascular structures from 3D cardiac MR images. From the learning perspective, the DenseVoxNet has three compelling advantages. First, it preserves the maximum information flow between layers by a densely-connected mechanism and hence eases the network training. Second, it avoids learning redundant feature maps by encouraging feature reuse and hence requires fewer parameters to achieve high performance, which is essential for medical applications with limited training data. Third, we add auxiliary side paths to strengthen the gradient propagation and stabilize the learning process and thus improve the segmentation performance.

【打印文章】 【添加收藏】