Blood vessel segmentation in magnetic resonance angiography imagery
Kozaitis, Samuel P.
MetadataShow full item record
Small blood vessels may be difficult to detect in magnetic resonance angiography due to the lack of blood flow caused by disease or injury. Our method, which uses a block-matching denoising approach to segment blood vessels, works well in the presence of noise. We examined extended regions of an image to determine whether they contained blood vessels by fitting a Gaussian mixture model to a region's histogram. Then, dissimilar regions were denoised separately. This approach was beneficial in low-contrast settings. It can be used to detect higher-order blood vessels that may be difficult to detect under normal conditions.