Partial volume correction for image-generated arterial input functions

We propose a method for Partial Volume correction and intensity recovery that models blood vessels as small cylinders of known diameter. We use a Bayesian classifier that explicitly models the effects of the point spread function on these cylinders. Although the method requires prior knowledge of the cylinder/arterial width, there is no requirement for any registration. A further advantage is that Region Of Interest (ROI) definition can be limited to only a few axial slices, thus minimizing time averaging. Furthermore, ROI selection requires only approximate placement around the target artery, encompassing both artery and background tissue, so that recovered data values are not operator-dependent. We present results for classifier performance on simulated phantom data of hot cylindrical inserts in a warm background with different contrast to noise ratios.