Cepstral estimation of arterial input functions in brain perfusion imaging

Abstract

To investigate voxel-specific arterial input functions (AIFs) obtained through blind deconvolution using complex cepstrum liftering. Blindly estimated AIFs have the potential of reducing dispersion effects in perfusion maps and are completely user-independent. The separability of AIFs and tissue residue functions (TRFs) in the cepstrum domain is exemplified using synthetic data, wherein the AIFs are modeled as gamma variate functions and the TRFs are modeled as exponential or linear functions. A novel separation filter is suggested. Initial results of different blind methods are illustrated using data from a stroke patient. The AIFs and the TRFs partly overlap in the complex cepstrum. The AIFs, obtained using the new separation filter, are closer to those obtained using noncepstral blind separation. The overlap of AIFs and TRFs in the complex cepstrum makes reliable separation of the two functions more difficult than previously described. Comparison to noncepstral blind methods suggests that similar optima are found using the new separation filter.