3D Soft Segmentation and Visualization of Medical Data Based on Nonlinear Diffusion and Distance Functions

Abstract

Visualization of medical 3D data is a complex problem, since the raw data is often unsuitable for standard techniques like Direct Volume Rendering. Some kind of pre-treatment is necessary, usually segmentation of the structures of interest, which in turn is a difficult task. Most segmentation techniques yield a model without indicating any uncertainty. Visualization then can be misleading, especially if the original data is of poor contrast. We address this dilemma proposing a geometric approach based on distance on image manifolds and an alternative approach based on nonlinear diffusion. An effective algorithm solving Hamilton-Jacobi equations allows for computing a distance function for 2D and 3D manifolds at interactive rates. An efficient implementation of a semi-implicit operator splitting scheme accomplishes interactivity for the diffusion-based strategy. We establish a model which incorporates local information about its reliability and can be visualized with standard techniques. When interpreting the result of the segmentation in a diagnostic setting, this information is of utmost importance.