Birgitte Fuglsang Kjlby1, Sren Christensen2, Irene Klrke Mikkelsen1, Kim Mouriden1, Peter Gall3, Valerij G. Kiselev3, Leif stergaard1
1CFIN, Department of Neuroradiology, Aarhus University Hospital, Aarhus, Denmark; 2Department of Neurology and Radiology, University of Melbourne, Melbourne, Australia; 3Department of Diagnostic Radiology, Medical Physics, University Hospital Freiburg, Freiburg, Germany
In perfusion DSC-MRI, the precision (random error) and accuracy (systematic bias) of perfusion estimates rely critically on the noise regularization used in the deconvolution process. Existing methods are commonly optimized for the best reproducibility of true perfusion values. We show that this accuracy is obtained at the expense of precision, which negatively impacts the ability to identify critical hypoperfusion thresholds. We propose a frequency-domain optimized regularization favoring precision. This approach reveals that optimal regularization depends critically on signal to noise ratio, sampling rate and AIF shape. Application of this method to simulated data improves discrimination of hypoperfused tissue.