Unsupervised Grow-Cut: Cellular Automata-based Medical Image Segmentation
Brown Bag Lecture by Dr. Payel Ghosh | 6/21/2011 11AM-12PM | 7th Floor Conference Room, Bldg 38A
Abstract: Segmentation is the process of deriving meaningful regions from images that are homogeneous with respect to local image features such as texture, color, edges etc. Supervised segmentation methods typically extract specific regions from images through user interaction or from manually generated training images. However, obtaining ground-truth is cumbersome for medical images and is also subject to intra and inter-operator variability. Unsupervised segmentation is particularly useful for region-based image retrieval applications that derive “similar” regions/classes of unknown categories from a large database of images. In this talk I will present a cellular automata-based unsupervised image segmentation technique. The unsupervised grow-cut algorithm (UGC) starts with a random number of seed points and automatically converges to a natural segmentation. The algorithm has been tested on a subset of medical images derived from the ImageCLEFmed database, 247 chest radiographs from the JSRT database and the Berkeley segmentation benchmark dataset. The unsupervised grow-cut algorithm has been compared against two other popular unsupervised segmentation techniques: the Mean Shift method and Normalized Cut method. The talk also describes how this research work contributes to the iMedline and the Tuberculosis screening project.
Bio: Payel Ghosh joined the National Library of Medicine as Postdoctoral Fellow in July 2010. She earned her Ph.D. degree in Electrical and Computer Engineering from Portland State University, Portland, OR in June, 2010. Her research interests include medical image processing, machine learning, and evolutionary computation methods such as genetic algorithms. She is currently working unsupervised image segmentation for local region-of-interest based image retrieval. Her mentors are Dr. Sameer Antani and Dr. Rodney Long in the Communications Engineering branch of the Lister Hill National Center for Biomedical Communications.