Removal of CSF pixels on brain MR perfusion images using first several images and Otsu's thresholding technique

Abstract

Brain MR perfusion imaging is used to evaluate local perfusion in patients with cerebral vascular disease. Quantitative measurements on the hemodynamic parameters and volume of brain with abnormal perfusion provide an estimation of the severity of the brain perfusion defect. However, quantitative measurements of these focal cerebral hemodynamic parameters are limited by the presence of cerebrospinal fluid (CSF) pixels. We noticed that the CSF has a higher signal than other tissue types on the first perfusion image, which is usually discarded in routine parametric image calculations. This signal difference, however, can be used to segment CSF pixels on the perfusion images. An image division was used to generate ratio images to compensate for spatially dependent signal variation caused by the inhomogeneity of excitation radiofrequency field. By applying an appropriate signal threshold to the ratio images, CSF pixels can be identified and removed from the parametric images. With the removal of CSF pixels, the volume of delayed-perfusion brain parenchyma can be better visualized and the interference from the CSF can be avoided. The proposed technique is simple, fast, automatic, and effective, and no extra scanning is needed to use this technique.