Deconvolution with Dictionary Learning

Time: 周三下午2点到4点

Venue: 信电大楼215学术厅

Speaker: 方若谷博士，美国康奈尔大学电子与工程学院

报告内容简介(Abstract)

Patient safety, especially radiation safety, has been a national priority through the National Institute of Health and the Food and Drug Administration. While the use of CT scans has skyrocketed over the past decades, the associated risk of cancer and adverse biological effects has raised significant concern to patients. Perfusion CT (PCT) imaging has been widely advocated to detect acute stroke or chronic brain disease yet the excessive radiation exposure is more serious than the routine CT. However reducing the radiation dosage will lead to image quality degradation that may hamper accurate and timely diagnosis of acute disease. In this talk, I will focus on how to develop a general, efficient and learning-based sparse deconvolution algorithm to achieve robust parameter map estimation and accurate detection of ischemic penumbra and hemorrhage in the brain. In particular, I will present how to utilize the complementary information in the high-dose repository of CT perfusion maps and learn a compact and adaptive dictionary for low-dose enhancement to handle low-contrast tissue and complex structural details. The algorithm has been applied to cerebral blood flow and permeability estimation, and extended with tissue-specific approach. The work is based on collaboration with radiologists and clinical experts in the radiology department of Weill Cornell Medical College.

报告人简介(Biography)

方若谷，美国康奈尔大学电子与计算机工程学院博士研究生。于2009年获浙江大学信息工程学院学士学位，竺可桢学院混合班学业排名第一，英国剑桥大学访问研究学生，国际冯氏基金会利丰奖学金获得者，香港大学2007-2008年度交流学生，浙江大学连续三年优秀学生学业一等奖学金，宝钢优秀学生奖学金，康奈尔大学Jocobs学者奖学金，纽约威尔康奈尔医学院访问学生。主要研究领域包括医学影像的增强和分析，计算机视觉，机器学习，数字医疗等。在国际知名刊物和国际会议上以第一作者发表论文10余篇, 包括Medical Image Analysis (JCR一区，影响因子4.4) 以及MICCAI (医学图像顶级会议)。2010年获国际图像处理大会(International Conference on Image Processing) 最佳论文奖, 2010年康奈尔工程研究会议最佳博士报告奖。提出的稀疏反卷积法的论文被评为2013年第二和第三季度Medical Image Analysis（医学影像分析顶级期刊）最热门论文。