Fast Filament Tracking Using Graphics Hardware

Recent developments in biomedical imaging are producing massive data sets containing structures, such as fine filaments, that are difficult to visualize. In this paper, we describe a method for tracking filaments and show how this technique can be used for segmentation and visualization. Instead of segmenting individual slices of a 3D data set, a large volume is processed allowing the filaments to be traced in a local frame positioned along the filament path. Images of cross sections of the filament are processed to predict the filament trajectory and trace the filament. Because only information local to the filament is processed, this technique has a speed advantage over segmentation of individual images. We show that this algorithm is easily extended to use graphics hardware in order to take advantage of parallelism and cope with the massive data sets produced using these imaging methods. Finally, we show how these techniques can be used to aid in the visualization of complex filament networks.