Maximum likelihood deconvolution of dynamic contrast MRI data

Bolus tracking of contrast agent with MRI can be used to measure local cerebral haemodynamic parameters including flow and volume. This can provide useful information for assessment of treatment options related to ischemic damage following stroke. The analysis of the acquired dynamic data requires the use of de-convolution to reconstruct the residue function (R) of the contrast agent. Measurement of the tissue time course and the arterial input function are obtained by T2 or T2* weighted sequences. Reconstruction of R provides estimates of flow, volume and mean transit time. The raw MRI signal intensity is well approximated by Rician statistics. The standard approach to estimation involves logarithmic transformation and least squares deconvolution. At low signal to noise this approach may not be efficient. A maximum likelihood (ML) deconvolution method involving an iterative re-weighted non-linear least squares algorithm has been developed. Studies are presented to evaluate improvements achieved by this approach relative to the standard deconvolution method. Mean square error properties of the residue function as well as the derived functionals of flow and blood volume parameters are considered. The results show that over a range of realistic signal to noise values, significant improvements in estimation accuracy are achieved by ML.