Automated perfusion-weighted MRI using localized arterial input functions

PURPOSE: To investigate the utility of an automated perfusion-weighted MRI (PWI) method for estimating cerebral blood flow (CBF) based on localized arterial input functions (AIFs) as compared to the standard method of manual global AIF selection, which is prone to deconvolution errors due to the effects of delay and dispersion of the contrast bolus. MATERIALS AND METHODS: Analysis was performed on spin- and gradient-echo EPI images from 36 stroke patients. A local AIF algorithm created an AIF for every voxel in the brain by searching out voxels with the lowest delay and dispersion, and then interpolating and spatially smoothing them for continuity. A generalized linear model (GLM) for predicting tissue outcome, and MTT lesion volumes were used to quantify the performance of the localized AIF method in comparison with global methods using ipsilateral and contralateral AIFs. RESULTS: The algorithm found local AIFs in each case without error and generated a higher area under the receiver operating characteristic (ROC) curve compared to both global-AIF methods. Similarly, the local MTT lesion volumes had the least mean squared error (MSE). CONCLUSION: Automated CBF calculation using local AIFs is feasible and appears to produce more useful CBF maps.