The segmentation of medical images to extract anatomical structures is an important problem in medical image analysis. Traditionally, this is done through a time-consuming and error-prone process in which an experienced physician marks a series of parallel contours that outline the structures of interest. Recent advances in surface reconstruction algorithms have led to methods for reconstructing surfaces from nonparallel contours that could greatly reduce the manual component of this process. However, no formal investigation has been conducted to determine what makes a “good” set of planes for reconstructing an arbitrary structure.
We are working to develop semi-automated tools for selecting a set of planes for a structure and authoring a protocol. Criteria to consider when choosing planes include planes that pass through feature points of the structure, planes that contain strong gradients in the image data, and planes that are orthogonal to previously selected planes. There are an infinite number of planes to choose from, so we cannot exhaustively try all possible combinations. In this talk, we present our work on an algorithm for plane selection and initial results with patient data.
Kevin Johnson, ’13
Majors: Politics, Computer Science
Sponsor: Ross Sowell