<research> current|past
		My research falls in the general category of multimedia content analysis, modeling, and indexing. In particular I'm interested in the applications of multimedia in health-care. The following provides a brief overview of my research. Please see the publications page for more detail. Current Research Since joining IBM T.J. Watson research center, I've been working on several different areas related to multimedia content analysis and indexing. Visual event modeling and recognition in news videos Mining multimodal medical records for decision support Concept-based health records Past Research Throughout my doctoral studies I was very closely involved in one of the major and multidisciplinary projects in Columbia University. The project, which was named PERSIVAL, aimed at providing personalized multimedia summaries of heterogeneous medical data (text of medical articles, diagnosis reports, patients' history, medical images, definition of medical terms,...) to users in the health-care. I was responsible for the image and video part, for which I focused on content-based analysis, indexing, and summarization of "echocariogram" videos. I developed innovative methods for modeling the content of these videos with appropriate spatio-temporal statistical models. My research which was funded by  DLI-2 program of NSF resulted in a number of publications both in medical and multimedia related conferences. I also have 3 patents on the technology that I created during my research. Here you can read an interview with me which appeared as the feature article in the "Health Imaging & IT" magazine about the future of the  echo videos management. I strongly believe that information-based medicine and fusion of heteregeneous medical information for making better decisions in health-care is going to have a greate impact on the future of medicine. Because health related information is multi-media in nature, devising automatic content-based analysis tools to extract knowledge from the vast amount of medical data is very important in that context. The following are snippets introducing different aspects of this research. Digital Echocardiogram Video Library (DEVL) DEVL is a library of echocardiogram videos illustrating different categories of cardiac abnormalities. In DEVL the content of each echocardiogram video is automatically analyzed and indexed at the structural units. View Recognition in Echocardiogram Videos A combined generative and discriminative method is proposed and implemented for automatic recognition of the different views of echocardiogram videos. ARGs are used to model the constellation of cardiac chambers in each view. MRF is used to match models to observations and obtain the likelihood of the observation. To deal with ambiguities due to structural similarities  the constellation are projected into an anchor space in which discriminative techniques are used to distinguish between them. Activity Modeling for Constellation of Parts The ARG used for representing the constellation of parts of the object is made time-dependent. The TD-ARG is trained to capture the specific activity pattern of the constellation of parts representing a non-rigid object. Coordinated Multimedia Summary (CMS) CMS is a methodology for creating multimedia summaries from a set of heterogeneous resources. The concept space related to the target of the summary could be navigated by interacting with the summary interface.