Separation and attachment point feature extraction from 2D cardiac MR blood flow images

We present a new method for extracting and quantifying separation and attachment points from 2D blood flow field images. The new technique calculates the signed streamline curvature at each point of the field which is a first order rather than second order quantity in a velocity field, and then searches for zero crossings of this function. Additional constraints based on the evaluated Jacobian at the candidates points are incorporated into the algorithm to eliminate the false positives attributed to noise. We also introduce the separation/attachment index which quantifies the strength of separation and attachment at the extracted points. Results of the method with synthetic data and a blood flow velocity image are presented.