Comparison of methods for quantitative analysis of dynamic susceptibility contrast enhanced brain perfusion MRI

Quantitative cerebral blood flow (CBF) can be obtained from dynamic susceptibility contrast (DSC) MRI using for instance the truncated singular value decomposition (tSVD). Block-circulant SVD and reformulated SVD (rSVD) are modified SVD approaches. The purpose of this study is to compare the different approaches. The optimal truncation thresholds (PSVD) for tSVD and block-circulant SVD are determined using simulated data. The optimal PSVD minimizes the CBF estimation error over all residue functions and CBF values. The observed optimal PSVD values are comparable to the values published by other authors. Most authors assume that the recirculation in the concentration curves has to be removed. Using noise-free curves the CBF estimates are independent of the recirculation. However, when noise is added, a shift in optimal PSVD occurs. Selecting the PSVD of curves without recirculation will cause underestimation of the curves with recirculation. The methods are compared in their ability to reproduce CBF. The determination of the PSVD per voxel using generalized cross validation and L-curve criterion is investigated as well. The tSVD approach with the optimal fixed PSVD performs best. In the presence of negative delays, the tSVD approach overestimates the CBF. Block-circulant and rSVD are delayindependent. Due to its delay dependent behavior, the tSVD approach performs worse in the presence of dispersion as well. However all SVD approaches are dependent on the amount of dispersion. Furthermore different CBV calculation methods are compared: integration of the whole time signal, integration of the first pass, fitting of a gamma variate function and integration of the area under the tissue impulse response function obtained during deconvolution. The latter method performs best for all CBF, CBV and SNR values. Using clinical data, large variations in the CBF values obtained with the methods and between tissue regions are observed. After the comparisons it can be concluded that block-circulant SVD is the most promising approach due to the delay independent behavior.