Medical Image Computing and Signal Processing Lab

A thin-structure filter based on Non-Local Means (NLM) distances, applied to maximum intensity projected Magnetic Resonance Angiography (MRA) images. The aim is to enhance visibility of low-intensity vessels with higher background levels. Contrary to the conventional NLM filter, a point filtering scheme is adopted using weights derived from mean NLM distances computed over a fixed neighborhood region around each pixel. The derivation of weights is performed sequentially in two cascaded steps, separately for intensity equalization and local background suppression respectively. Restoration of vascular structures at finer scales is achieved by employing recursively estimated correction terms derived from Non-Local gradients applied to multi-channel phase contrast images.