Hanchuan Peng, Ph.D.
Janelia Farm Research Campus, Howard Hughes Medical Institute19700 Helix Drive
Ashburn, VA, 20147
Biography:
Dr. Hanchuan Peng is currently a senior computer scientist and head of a bioimage data analysis & mining lab at Janelia Farm Research Campus, Howard Hughes Medical Institute (HHMI). His research interests include bioimage informatics, neuroscience, computational biology, pattern recognition and machine learning, and neuronal networks. His recent work include image mining of in situ gene expression patterns of fly embryos, minimum-Redundancy-Maximum-Relevance (mRMR) feature selection, Bayesian structure-function analysis for brain images, and building a 3D digital cell atlas for C. elegans. His on-going projects include building a 3D high-resolution digital atlas for a fruit fly brain, multiscale image understanding and mining, etc. Dr. Peng is active in the bioimage informatics/mining field. He was a program chair of the 2005 and 2006 International Workshops on Bioimage Informatics (www.bioimageinformatics.org) held at Stanford and Santa Barbara, respectively, a guest editor of a BMC Cell Biology supplement on bioimage informatics published in July 2007. He is a co-organizer of 2008 Conference on Computer Vision for Neuroscience and 2009 Conference on Bioimage Informatics, and is on the editorial board of International Journal of Data Mining and Bioinformatics. His lab website is http://research.janelia.org/peng .
Talk Title: Automatic Annotation of Patterns in Bioimages
Abstract: Automatic annotation of patterns in images and other media (e.g. videos) is fundamentally hard. For biological image datasets that are produced in a well-controlled way, this problem is still intrinsically challenging due to computational difficulties such as how to detect and select good features, how to estimate the feature/sample distribution in high-dimensional space, and how to build effective classification scheme when there are hundreds of annotation classes corresponding to ontological, anatomical, or behavioral descriptors. We designed computer programs to automatically recognize and annotate gene expression patterns in multidimensional images, such as the in situ mRNA gene expression patterns for fruit fly embryos and the neuronal patterns in fruit fly brains. Our systems use scalable parallel classifiers, each utilizing the minimum redundant set of features that capture both the global and local spatial information of an image pattern. For other related problems such as the whole-animal single-cell level annotation of cells in an animal, we also designed novel tools to assist both the inclusion of expert knowledge in the database and the automatic identification of the anatomical structures in an image.