| 研究所简介 |
| 天津市重点实验室 |
| 所长致辞 |
| 组织机构 |
| 联系我们 |
|
2010年秋季先进机器人与MEMS技术系列学术讲座（32）
添加日期：2010-09-13 作者：宋洪生老师 来源：
2010年秋季先进机器人与MEMS技术系列学术讲座（32）
南开大学机器人与信息自动化研究所 Institute of Robotics and Automatic Information System 2010年秋季先进机器人与MEMS技术系列学术讲座 Seminar Series：Advanced Robotics & MEMS 报告人：李硕 博士 单位：Digital Imaging Group of London 题目： 医学图像处理的新进展，Automated medical image analysis 地点：主楼227 时间：2010年9月17日（周五）下午3:00~5:00 报告将在医学图像的分割、配准和基于机器学习的图像理解方面介绍他们小组近来的理论成果，并给出这些理论在心脏影像分析与辅助诊断方面的实际应用，摘要如下： Abstract: Automated medical image analysis has attracted many leading researchers in the field with great progress developed in the last decade. The three major pillars of medical image analysis are segmentation, registration and machine learning. The talk will start with a brief state-of-art review of each pillar. And then the recent progress of each of them developed in our group will be introduced, compared and validated with the real application and challenging practical data. 1) Fast, accurate variational image segmentation: Level set, a continuous solution, and graph cut, a discrete solution, are two major progresses in numerical solutions to solve partial differential equation. We introduce several advanced modeling, which embed the domain knowledge naturally to solve the current challenges such as real time segmentation and co-segmentation. 2) Parametric deformable registration. A parameterization of deformation fields for diffeomorphic image registration will be introduced, through which a registration can be parameterized with just two constrained parameters. Validation showed the great efficiency of this advanced solution. Moreover, the general framework can be incorporated into the current existing general registration framework to enforce the constraints learned from domain and expert to improve the performance and efficiency. 3) Hybrid machine learning: A customized machine learning system will be introduced, which combines the strength of information theory, Bayesian modeling and neural network to accommodate the domain and expert’s knowledge. Finally, the talk will be concluded with their integrated applications in medical image analysis, with a major focus and coherent example on cardiac image analysis. 李硕，2006年初于加拿大Concordia 大学计算机系获得博士学位，从事模式识别，计算机视觉及其在影像医学中的应用等方面的研究工作。同年加入加拿大通用电气公司(GE)医疗部任科学家，主攻医疗图像的智能分析。2006底，他被聘请为University of Western Ontario医学院的兼职教授，2007年被该校研究生院授予博士生导师资格。他同时还是加拿大Lawson Health Research Institute 的兼职研究员。2007年他创建了伦敦医疗图像研究部（digital imaging group of London）并担任主任，该部门从最初的2个人发展到现在的22人。从2007到现在，他是一个国家项目，三个省级项目，和二个通用电气公司“下一代产品开发项目”的首席科学家，先后5次在技术创新和管理方面获得公司嘉奖。李硕在领域内的顶级期刊和会议(IEEE transactions, IJCV, CVPR,MICCAI,IPMI)发表了30多篇论文并获得10余项专利，部分技术已转移到GE公司的医疗设备中。他们团队的主页是http://dig.lhsc.on.ca/index.php |