Intuition in Medical Image Segmentation: Visualizing Graph Edge Weights

Weighting functions for graph-based medical image segmentation algorithms (e.g., Graph cut) have a significant effect on the segmentation, but to our knowledge no tool provides the user with intuition towards their proper selection. The large variety, their complexity, and the limited feedback hinders comparison of choices. This paper describes a package developed to visualize the effects of various edge weighting functions and parameters, in which the image of interest is overlaid with colors depicting the relative distances from the nearest seed to each voxel. By seeing the colors vary while changing parameters, the user gains intuition into the various options for the edge weighting function. A user study demonstrating the benefits of the package is presented. It is our hope that the intuition provided by the software will result in less time required to segment medical images in the clinical work-flow.