Automatic image segmentation using edge detection by tissue classification in local neighborhoods

A highly sensitive edge detector has been developed that uses tissue classification of pixels based on analysis of data in their local neighborhoods. The author extended the concept of tissue classification over the whole image to classification at the local pixel level by examining data in small overlapping neighborhoods. Edges are detected wherever it can be determined that more than one pixel population is present in a local neighborhood. Signal amplitude can vary across the object as long as variation is less than that at the edge. The detector has been applied to multislice, multiphase gradient echo images of the human heart. A segmentation strategy specified slice processing order graded regions of interest (ROIs) in gradient echo nuclear magnetic resonance image data and used successfully detected ROIs to guide subsequent detection. In the 1688 slices processed 95.6% of boundary pixels were automatically detected