Sparse methods on manifolds of images and shapes for significant anatomy detection linked to disease

点击翻译按钮获取中文摘要

Medical imaging plays a key role in modern medicine, being part of many diagnoses and treatments. While imaging hardware and acquisition methods have seen a tremendous development in the last decade, this is not matched by advances in image processing software. Therefore, in clinical practice, much image processing work is either not performed at all, or performed manually due to inadequate software. To realize the full benefit of these new medical imaging methods, new and better image analysis methods need to be developed. My current and planned research focuses on developing modern medical image analysis methods for studying anatomical and physiological variability in humans due to normal aging or disease. Such studies involve computational methods that extract and interpret image information to identify and quantify change. Our methods make use of anatomical similarity between humans parts and organs (such as ventricles or hippocampus among brain structures) to define a shape model. This shape model is defined in a continuous way and captures both global and local variability between individuals. We develop methods to statistically study disease-induced changes in anatomical shapes. This will provide a way to better understand the influence of certain diseases. This proposal introduces both theoretical advances of main medical image analysis methods (segmentation, shape analysis, multidimensional statistics) as well as practical modern solutions for a brain imaging study through our collaborative medical and biomedical collaborators. In addition, it opens the opportunity of training graduate students in an interesting multidisciplinary environment.