Limit of Detection of Localized Absolute Changes in CBF Using Arterial Spin Labeling (ASL) MRI

Hence, while a larger regression kernel might be more suitable for baseline measurements, it can underestimate changes in CBF due to activation or disease. The goal of this ongoing study is to investigate the limit of detection of absolute CBF changes using ASL. Here we show results from simulated data analyzed using both the PVE-correction method (with varying regression kernel sizes) and the conventional one that does not correct for PVE. Validity of the results is shown on an experimental stroke model from a patient with left thalamic infarct. Methods - Simulations: A theoretical ASL CBF image was constructed based on simulated EPI control and label images. Simulation of the EPIs was based on SPM5 templates for gray matter (GM), white matter (WM) and CSF tissue probability maps; tissue magnetization ratios were assumed mCSF:mGM:mWM=1.68:1.23:1.00 2 . GM and WM CBF were assumed 107 and 30 (mL/100g·min), respectively 2 . Matlab code was written that allowed for a 2D step function in CBF to be applied with varying width (in the xy plane) and location across the brain. Results shown here correspond to a step function of width = 5x5x1 voxels (corresponding to an activated region = 15x15x9 mm 3 .) The patient was a 54 yo, right-handed male, diagnosed with sub-acute left-thalamic infarct. Written consent was obtained as approved by the institutional IRB. Imaging was done on a 1.5T scanner (Philips) using a standard transmit-receive coil as previously described 3 . Briefly, single shot spin-echo EPI CASL images were acquired with: TR/TE/=4s/36ms, θ=90o; FOV=220×198 mm 2 ; acq.matrix=64 × 58; 15 slices, thickness/gap = 8mm/1mm; post-label delay=1000ms; labeling time=2.0s. A high resolution, 3D T1 (SPGR) image was acquired: TE/TR=3 ms/34 ms, θ =45°; 100 slices; slice thick.=1.5mm/1mm gap; FOV=240×240mm 2 ; acq. matrix=256 × 256. PVE-correction: Pure tissue-specific CBF maps were obtained for the simulated and patient data as previously described 2 . To assess the effect of regression kernel size, analysis was run with varying kernel sizes in the xy-plane; the z-dimension=1 (corresponding 9mm). ASL CBF images were also obtained from the conventional, PVE-uncorrected data. Results: Fig.1 shows results from simulated data. A stepwise increased activated region, ΔCBF=50%, was centered on a 100% GM voxel. The larger the kernel size, the lower the amount of activation recovered (see also Fig.2). Furthermore, for larger kernel sizes, the observed activated regions were spread onto the unactivated voxels due to the smoothing effect. Results were compared with those from the conventional, PVE- uncorrected method (squares). The PVE-uncorrected method yielded results similar to the corrected method for larger regression kernels.