Medical image segmentation, the process of dividing a medical image into meaningful objects such as organs, tumors, etc., is a critical tool that allows medical professionals to provide customized medical care to patients. In the past this highly technical, individualized care has required experts to manually analyze the images, a process that is very expensive in both time and money. Over the past decade, enormous technological advances have been made in biomedical imaging, leading to a large amount of new and improved medical data which has created a demand for algorithms which can process this data faster and more thoroughly. Researchers have worked extensively to develop these medical image segmentation algorithms, but current algorithms suffer from the following drawbacks: 1) they do not have the capability of effectively representing diverse shapes of a wide variety of medical objects and/or 2) they require substantial interaction from an expert user. This research will develop a novel medical image segmentation algorithm that can be applied to various types of medical images and will be able to be executed by any user with basic computer literacy. Many important objects will be able to be handled with the same algorithm, such as livers, prostates, and vertebrae. This research allows medical experts to spend less time analyzing a wide variety of medical images and more time directly working with patients.
The algorithm will work for any medical imaging object of interest whose shape can be decomposed into a small number of components with a very simple geometric structure. For example, livers may be slightly different from person to person, but almost all livers can be represented as a union of two or three "star-shaped" components. A component is defined to be star-shaped if there is a center point in the component such that the line segment connecting the center to every other point in the component is contained within the object. If the center of a single star-shaped component is known, then the whole component can be very quickly identified by computer algorithms, but as the number of components increases, the simultaneous computation of all the components becomes much more difficult. This research will develop algorithms which can automatically compute the centers of the star-shaped components for many medical imaging objects such as livers, prostates, and vertebrae, and further will develop algorithms that can simultaneously identify all the components for the objects. The result will be a single algorithm that will be applied to many scenarios and can be executed by non-technical users.
