GPU-Accelerated and memory optimized vessel enhancement filters for micro-CT tomography

In this paper we present results relative to speeding up multiscale vessel enhancement filters for angiography images by using Graphic Processing Units (GPUs). These filters, proposed initially by Frangi [1], can be used to preferentially enhance image features that have a tubular-like structure, and therefore result in the enhancement of vessels. Vessel enhancement filters are used commonly as a pre-processing step for vessel segmentation. As this process is computationally and memory intensive, filtering high-resolution 3D micro-CT (mCT) images can take several hours. We propose a method for speeding up this process using GPUs which results in significant speed gains and which is memory efficient.