Research & Publications
Dr. James Duncan works with engineering and mathematical principles and uses mainly signal/image processing techniques to derive useful image feature information and capture model-based information in concise mathematical forms. He uses nonlinear optimization methods to implement reasoning strategies. Dr. Duncan's image processing and analysis research can be divided into three general areas:
Segmentation of meaningful regions and/or objects in images. He looks at deformable model-based approaches to boundary finding and aim at locating the complete boundary of a deformable object efficiently and reproducibly. This is done by incorporating basic structural or parametric models into an optimization-based search strategy. Initially, Dr. Duncan aimed at 2D parametrized boundary finding, but now he has extended it into an approach for segmenting deformable surfaces from three-dimensional biomedical image data sets. Current applications include segmentation of the left ventricle of the heart from 4D cine Magnetic Resonance images (MRI) and segmentation of the temporal lobes of the brain from static MR images. Future research will focus on isolating structures using images obtained from laser-scanning confocal microscopes.
Image-based measurement and quantification of anatomical, physiological and/or clinically meaningful parameters. A primary example of this research is the tracking and modeling of non-rigid motion for the purpose of quantifying cardiac left ventricular (LV) regional function from 2-D and 3-D diagnostic image sequences. Such quantification permits measurements that are useful in understanding the basic relationships between the state of the heart muscle (myocardium) and overall LV function; these measurements can be important for managing patients with ischemic heart disease. Dr. Duncan's methodology makes use of mathematical models related to the motion of 3-D elastically deformable objects and is adaptable to the nonlinear, non-rigid regional motion of the LV.
Development of decentralized approaches for forming complete, integrated computer vision/image analysis systems. Dividing computer vision and image analysis systems into modular hierarchies is useful computationally and also allows to model particular image understanding tasks better. Hierarchical systems are necessary to perform such complex tasks as recognizing anatomy, quantifying the shape and motion of the heart, or performing an integrated segmentation of anatomical structures using multiple imaging modalities. He has recently developed an approach based on the mathematical concept of game theory, where functionals representing each module's tasks both compete and cooperate to make decisions about particular image analysis goals.
Specialized Terms: Computer vision; Image processing and medical imaging, with an emphasis on biomedical image analysis
Extensive Research Description
Biomedical Engineering; Diagnostic Imaging; Magnetic Resonance Imaging; Radiology; Temporal Lobe