Frequently Asked Questions
	1. Why perfusion fMRI? What experimental designs are suitable for perfusion fMRI?
	2. What are the drawbacks of perfusion fMRI? Why isn’t everyone using it?
	3. There seem to be a variety of ASL techniques out there. Which one provides the best sensitivity for functional neuroimaging studies?
	4. It is unclear how perfusion parameters are selected. What is the best way to optimize the signal to noise ratio while minimizing artifacts? Are there any trade-off or implications of the choice of the parameters?
	5. How do I generate the perfusion data from the raw image series? What is the difference between the preprocessing of perfusion and BOLD data?
	6. What is the difference between the statistical analysis of perfusion and BOLD data?
	7. Is it necessary to calculate the quantitative perfusion image series for data analysis?
	8. How can perfusion fMRI improve studies of brain areas with high susceptibility from magnetic field inhomogeneities (e.g., orbital frontal cortex)?
	1. Why perfusion fMRI? What experimental designs are suitable for perfusion fMRI?
	Several recent studies indicate that arterial spin labeling (ASL) perfusion contrast may offer certain advantages over the widely adopted BOLD method. These include: improved sensitivity for slow changes in neural activity, reduced inter-subject variability, more specific functional localization and generally reduced susceptibility effects. Studies of slowly developing processes (e.g., mood changes or procedural learning) as well as studies that require comparison of widely-spaced observations (e.g., drug effects) are particularly suitable for perfusion fMRI due to its stability over long time scales. BOLD fMRI, on the other hand, suffers power loss at low task frequencies due to drifts in the baseline. Perfusion contrast is also appealing for visualizing brain activity in regions with high static susceptibility, such as the limbic system, orbitofrontal and inferior temporal regions. Finally, clinical neuroscience studies can benefit from quantitative perfusion measurement as it provides baseline information which may be useful in interpreting different neuroimaging results seen in patient and control groups. Because perfusion fMRI time-series do not possess temporal autocorrelation in time, these data can also be analyzed using virtually any parametric or nonparametric data analysis method, an option not available for BOLD data.
	2. What are the drawbacks of perfusion fMRI? Why isn’t everyone using it?
	In addition to the simple novelty of the method, and its unavailability at most centers, there are some specific limitations of perfusion fMRI. Chief among these is the reduced magnitude of the signal change for perfusion as compared to BOLD. Thus, for experimental designs that seek to detect changes in neural activity within a subject that are evoked faster than every 30 seconds (e.g., most event-related designs), BOLD remains the superior choice. Next, the image coverage in ASL methods is inferior to that of BOLD, and is not ideal for whole brain studies. Finally, due to the pairwise acquisition of label and control images as well as the required delay time for the tagged blood to flow into imaging slices, the temporal resolution in perfusion fMRI is relatively low. Specifically, an effective temporal resolution of 4 seconds represents a practical limit. Nevertheless, recent technical developments are rapidly changing the face of ASL methods. The use of phase-delayed experimental events with respect to image acquisition can provide higher effective temporal resolution of evoked responses. Also, refinement of fast-spin echo methods will provide improvements in slice coverage and acquisition times.
	3. There seem to be a variety of ASL techniques out there. Which one provides the best sensitivity for functional neuroimaging studies?
	ASL contains a class of techniques and can be generally divided into two categories: continues and pulsed ASL. Continuous ASL, which utilizes long RF pulses for labeling, provides improved perfusion contrast at the cost of imaging time and a high level of RF power deposition. Pulsed ASL, which utilizes a nearly instant RF pulses for labeling, is easier to implement. Although the perfusion signal is smaller in pulsed ASL methods, more sample points can be acquired in the same amount of time in pulsed than in continuous ASL methods. As a consequence, these two types of ASL techniques provide comparable sensitivity and statistical power for functional activation studies. Pulsed ASL methods also contain a variety of subtypes, such as FAIR, PICORE and QUIPSS. The perfusion contrast in these techniques are fairly similar although subtle differences do exist. Keep in mind that the temporal stability of the MRI scanner plays an important role in determining the sensitivity of ASL methods.
	4. It is unclear how perfusion parameters are selected. What is the best way to optimize the signal to noise ratio while minimizing artifacts? Are there any trade-off or implications of the choice of the parameters?
	The delay time (between the labeling and image acquisition) is a specific parameter related to ASL methods. Shorter delay times, improve the perfusion signal but reduce the accuracy of perfusion quantification. If the delay time is too short for the labeled blood to reach imaging slices, activation cannot be picked up by ASL methods. In experimental settings, delay time should be chosen as the optimal point in the trade-off between these factors. Most studies have reported transit time on the order of a few hundred ms to around 1 sec. In pulsed ASL methods, the delay time should be at the upper limit of the transit time range. An empirical way to determine the optimal delay time is to do ASL at serial delay times. A rough estimation of transit time can be visually judged as the time that the large vascular signal disappears. In continuous ASL methods, shorter delay time can be used to improve the SNR as the labeling has accumulated in the brain during the long labeling period. The image acquisition time in ASL methods is limited by the T1 relaxation of the labeled blood, allowing a time window of only a few hundreds ms. The choice of slice number, image resolution and coverage has to be balanced to meet the criteria for total image acquisition time. It is generally suggested to use larger slice thickness in perfusion fMRI than in BOLD to improve the SNR and image coverage.
	5. How do I generate the perfusion data from the raw image series? What is the difference between the preprocessing of perfusion and BOLD data?
	Perfusion image time series are generally produced by pair-wise subtraction between interleaved label and control acquisitions. The simplest way is to do direct pair-wise subtraction between temporally adjacent label and control acquisitions. Artifacts can be reduced by first performing a sinc interpolation to shift the series of label (or control) images by one TR, followed by pair-wise subtraction between time-matched label and control images. “Surround subtraction” has also been proposed through linear interpolation to shift the series of label (or control) images by one TR, followed by pair-wise subtraction. However, the “surround subtraction” method will slightly inflate the false-positive rate in statistical analysis through temporal smoothing. The preprocessing of perfusion and BOLD data is similar, although motion correction may be performed separately on the label and control images in the raw image data. Also, perfusion fMRI data tend to benefit from spatial smoothing to a greater extent than does BOLD data as a consequence of the noise characteristics. Finally, the preprocessing of BOLD fMRI data may include baseline detrend, which is not necessary for perfusion data.
	6. What is the difference between the statistical analysis of perfusion and BOLD data?
	Because perfusion time series is independent in time, it is legitimate to apply the standard general linear model for statistical analysis. Unlike BOLD data, temporal smoothing is not required in the analysis of perfusion fMRI data and the degree of freedom can be preserved. The inherent “white noise” property in ASL data simplifies statistical analysis and affords the opportunity to apply virtually any traditional parametric or non-parametric statistical test, such as permutation analysis. The group analyses of perfusion and BOLD data are very similar.
	7. Is it necessary to calculate the quantitative perfusion image series for data analysis?
	It helps, but in some cases is not strictly necessary. Because the procedure to generate quantitative perfusion images generally involves calibration using a reference image, absolute perfusion series are less sensitive to variations in the scanner condition. This is important for comparison of temporally widely-spaced blocks. Perfusion quantification will also reduce the inter-subject variability in the group analysis as cerebral blood flow tends to be quite consistent in healthy adult subjects. If absolute perfusion quantification is not possible, the subtracted (relative) perfusion image series can be used for most paradigms except those involving scans on different days.
	8. How can perfusion fMRI improve studies of brain areas with high susceptibility from magnetic field inhomogeneities (e.g., orbital frontal cortex)?
	Perfusion images with gradient echo acquisitions have reduced susceptibility effect compared to BOLD images because a minimum TE is used. Spin echo sequences can be used for image acquisition in ASL methods, such as spin echo EPI and SPIRAL. Parallel imaging with array coil will further shorten the TE and image acquisition time, and subsequently reduce susceptibility artifact. Echo train and steady state imaging sequences such as FSE and FISP can also be used but generally restricted to single slice. Some specific sequence with reduced susceptibility sensitivity, such as reversed SPIRAL, offers alternative approach for perfusion imaging in orbitofrontal and temporal regions.