AUTOMATED SEGMENTATION OF THE RIGHT HEART USING AN OPTIMIZED SHELLS AND SPHERES ALGORITHM

We have developed a novel framework for medical image analysis, known as shells and spheres. This framework utilizes spherical operators of variable radius centered at each image pixel and sized to reach, but not cross, the nearest object boundary. Statistical population tests are performed on adjacent spheres to compare image regions across boundaries. Previously, our framework was applied to segmentation of cardiac CT data with promising results. In this paper, we present a more accurate and versatile system by optimizing algorithm parameters for a particular data set to maximize agreement to manual segmentations. We perform parameter optimization on a selected 2D slice from a 3D image data set, generating effective parameters for 3D segmentation in practical computational time. Details of this approach are given, along with a validated application to cardiac MR data.