Velocity extraction from spin-tagging MRI images using a weighted least-squares optical flow method

Magnetic resonance imaging (MRI) can provide truly non-invasive measurements of internal flow fields. The extraction of velocity from spin-tagging images requires the quantitative tracking of grid nodes. A weighted least-squares optical flow method was used in this work to estimate the displacements of the grid nodes and tags from synthetic and real spin-tagging MRI images. To investigate the accuracy of the proposed method, synthetic spin-tagging images were generated using the Poiseuille law analytical profile. Three synthetic sequences with different levels of noise were generated and the average and maximum absolute errors were estimated for points corresponding to grid nodes and tags. Different sizes and shapes of region of interest (ROI) were investigated to determine the optimal size and shape for reliable extraction of velocity both for synthetic and real spin-tagging MRI images. The optimal ROI size was found to be 13x13 pixels2 . The average and maximum absolute error for the velocity in vertical direction for synthetic data using the optimal ROI size ranged from 5.46% to 14.42% and from 6.39% to 31.96% respectively.