Maximum entropy deconvolution for dynamic susceptibility contrast magnetic resonance imaging

Dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) describes local tissue perfusion by rapidly imaging a bolus of contrast agent as it passes through the cerebral vasculature. DSC-MRI may identify functionally salvageable tissue in acute ischemic stroke. The DSC-MRI data set is a 4-dimensional function of space and time. In order to obtain quantitative, absolute maps of tissue perfusion the measured time curves must be deconvolved for each pixel in the 3D volume. The most widely used deconvolution method for DSC-MRI, singular value decomposition (SVD), can cause truncation artifact in the deconvolved curves, resulting in inaccuracies in the perfusion measurements. We present a maximum entropy (ME) deconvolution approach to obtain quantitative measures of local cerebral perfusion. SVD is discussed in the context of Fourier theory and insight is provided into the motivation for a statistically-based iterative reconstruction algorithm. Rigorous computer simulations are performed to evaluate and compare ME and SVD in clinically realistic situations. Our results show that ME better recovers the true perfusion profiles. It achieves a smaller meansquared error than SVD and provides significantly better estimates of cerebral blood flow and blood volume.