Computational protocol: Optimization of 4D vessel‐selective arterial spin labeling angiography using balanced steady‐state free precession and vessel‐encoding

Similar protocols

Protocol publication

[…] All images were processed using the FMRIB Software Library (FSL ) and MATLAB (MathWorks, Natick, MA, USA). Both 4D and dynamic 2D VEPCASL angiography data were motion corrected using MCFLIRT . The Brain Extraction Tool was applied to the 3D TOF data to generate a brain mask, which was then linearly registered (using FLIRT ) and applied to the VEPCASL angiography images to exclude signal from the scalp and eyes.In order to separate the blood signals arising from each feeding artery from the vessel‐encoded data, a fast maximum a posteriori Bayesian version of the general framework for vessel‐encoded analysis was applied to the magnitude images. This method is SNR efficient and compensates for rigid body motion between the vessel localization scan and the vessel‐encoded acquisitions. This analysis results in one vessel‐selective dynamic angiogram for each of the four main brain‐feeding arteries, which can be displayed separately or together using color to represent the vascular origin of the blood signal. Maximum intensity projections (MIPs) of each 4D data set were performed to aid visualization.Due to the long VEPCASL pulse train (1000 ms), the first acquired time frame shows the vessels filled with labeled blood, with subsequent time frames showing the outflow of this bolus. For a more intuitive visualization, “inflow‐subtraction” was performed, in which each time frame was subtracted from the first. In this way the time at which the bolus leaves the voxel in the original images becomes the time at which the signal appears in the inflow‐subtracted images, giving the appearance of inflow rather than outflow. This method has been previously used by our group and others . In practice small residual transient bSSFP artifacts were observed in the first time frame and these would propagate through all subsequent inflow‐subtracted images, so the second time frame was used instead.Timing information was obtained from both 4D and dynamic 2D data sets using a simple outflow metric: the signal in each voxel was temporally smoothed (Savitzky–Golay filter with polynomial order = 2 and frame size = number of readout blocks – 1) before calculating the time at which it drops to less than 50% of its maximum, as per Okell et al. . This is approximately equivalent to the time the blood takes to travel from the labeling plane to each voxel if bolus dispersion and signal attenuation due to the RF imaging pulses are minimal.In order to assess the vessel selectivity of the proposed method, regions of interest (ROIs) were manually defined in the M2 segment of both middle cerebral arteries (MCAs) and in the P2 segment of both posterior cerebral arteries (PCAs) for each subject using the transverse MIPs of the 4D data. In one subject, who had a fetal type circle of Willis on one side, the ipsilateral PCA ROI was drawn in the P1 segment to exclude true ICA signal. To exclude signal outside arteries a vessel mask was created by summing the blood signal across all feeding arteries, determining the 99th percentile of this signal intensity and thresholding the image at 95% of this value, which was found empirically to segment the large arteries well. Within the intersection of each ROI with the vessel mask the fraction of the total signal arising from each feeding artery was calculated. The mean “contamination” signal arising from arteries not expected to contribute to each ROI was then calculated: in the MCA ROIs only the ipsilateral ICA is expected to contribute and in the PCA ROIs only the VAs are expected to contribute (in the case of a typical vascular anatomy).In order to aid comparison between the bSSFP and SPGR data acquired in one subject, a measure of SNR efficiency was calculated. The “signal” was taken as the mean signal intensity of the ASL angiograms within a vessel mask (as described above), averaged across all feeding arteries and time frames. The “noise” was taken to be the standard deviation of the signal in a 10 × 10 pixel background ROI positioned in the posterior of the brain, away from any arteries, across all feeding arteries and time frames. The ratio of the signal to the noise was then divided by the square root of the acquisition time (in minutes) to yield the SNR efficiency.Finally, a qualitative comparison of arterial visualization was performed on both the 3D TOF and the 4D VEPCASL angiography data by an experienced neuroradiologist (F.S.) using a scoring system similar to that of Wu et al. : 0 – not visible/non‐diagnostic; 1 – poor (some structure visible but not clearly defined); 2 – good (diagnostic, clear vessels); 3 – excellent (very clearly delineated). All the major cerebral arterial segments were scored, including the first and second segments of the ACAs, MCAs and PCAs, along with the distal MCA segments (defined as any MCA branches beyond M2). The reader was blind to the subject numbers and data were presented in a randomized order to prevent potential bias. A nonparametric paired Wilcoxon signed rank test was performed to determine whether differences in TOF and VEPCASL angiography scores within each vessel segment were significant. […]

Pipeline specifications

Software tools FSL, BET
Application Magnetic resonance imaging
Organisms Dipturus trachyderma, Homo sapiens