Non-contrast Intracranial 3D MR Angiography using Pseudo-Continuous Arterial Spin Labeling(PCASL) and Accelerated 3D Radial Acquisition

Abstract

Pseudo-Continuous Arterial Spin Labeling (PCASL) can be used to generate non-contrast MR angiograms of the cerebrovascular structures. Previously described PCASL-based angiography techniques were limited to 2D projection images or relatively low-resolution 3D imaging due to long aquisition time. This work proposes a new PCASL-based 3D MRA method that uses an accelerated 3D radial acquisition technique (VIPR, spoiled gradient echo) as the readout. Benefiting from the sparsity provided by PCASL and noise-like artifacts of VIPR, this new method is able to obtain sub-millimeter 3D isotropic resolution and whole head coverage with a 8-minute scan. Intracranial angiography feasibility studies in healthy (N=5) and diseased (N=5) subjects show reduced saturation artifacts in PCASL-VIPR compared to a standard Time-of-Flight protocol. These initial results show great promise for PCASL-VIPR for static, dynamic, and vessel selective 3D intracranial angiography.

Introduction

Cerebrovascular disease is the 3rd leading cause of mortality in the United States accounting for approximately 200,000 deaths in the US each year as well as considerable neurologic morbidity (1). Imaging of the cerebral vasculature system is paramount for both diagnosis and treatment planning. X-ray Digital Subtraction Angiography (DSA) has long been considered the reference standard for cerebral angiography with modern protocols now allowing for extremely high resolution, 3D static imaging as well as vessel selective, dynamic 2D projection imaging. However, X-ray DSA poses risk to patients due to the need for invasive access, ionizing radiation, and iodinated contrast agents. Additionally, X-ray DSA provides an incomplete assessment of tissue damage. Magnetic Resonance Imaging (MRI) is capable of assessing disease markers including tissue viability (2), perfusion (3), and hemorrhage (4). Unfortunately, MR angiography (MRA) is severely lacking compared to X-ray DSA. Intracranial MRA is most frequently performed with three-dimensional time-of-flight (3D TOF) with multiple overlapping thin-slabs (5). This technique provides relatively high spatial resolution, however is limited by the saturation of spins in slow, complex, or in-plane flow (6) and provides limited information of vessel filling patterns.

Developing techniques hold the potential to significantly improve intracranial MRA. Contrast enhanced MRA (CE-MRA), which has proven to be robust in the extracranial vasculature, is highly challenging in the cerebral vasculature due to the rapid passage of blood from arteries to veins. However, CE-MRA has reaped the benefits of accelerated imaging strategies (7–9) and is now feasible for intracranial imaging, albeit at relatively low spatial resolutions compared to X-ray DSA. Non-contrast enhanced technique have seen renewed development since the discovery of nephrogenic systemic fibrosis (NSF) (10,11). One of these techniques, arterial spin labeling (ASL) is of particular interest for intracranial applications. Similar to TOF, ASL relies on the inflow of blood into the imaging volume; however, ASL uses separate sequences to label and image inflowing spins. By subtracting the images with different labeling sequences, angiography can be obtained with near zero background, vessel selectivity, and inflow dynamics similar to X-ray DSA. Although ASL has been in development since 1987 (12), its use in angiography has been limited due to low SNR, longs scan times, and difficulties in practical implementation.

ASL MRI can be divided into two basic types: pulsed ASL (PASL) and continuous ASL (CASL). PASL employs a single inversion pulse, and has been popular in the last decades due to its easy implementation. With recent improvements in accelerated imaging, promising intracranial results have been shown utilizing PASL with spoiled gradient echo (SPGR) for static angiography (13). Unfortunately, PASL angiography is highly sensitive to the selection of inversion time (TI), is subject to blurring due to tag decay during the readout, and the extent of the labeling is limited by the field homogeneity of the excitation radio frequency. Inversion time resolved, balanced steady state free precession (bSSFP) offers the potential to retrospectively choose inversion time with reduced blurring (14–17), but also introduces substantial artifacts from the bSSFP readout.

CASL techniques utilize flow driven adiabatic inversion to produce higher SNR images with reduced sensitivity to TI. However these techniques have been challenging in practical implementation therefore has not been used widely. Recently, a new strategy for CASL has been reported referred to as Pseudo-Continuous ASL (PCASL) (18,19), which allows practical implementation without specialized hardware. In an initial extra-cranial comparison, PCASL has been demonstrated to provide significantly higher arterial signal compared to PASL and improve diagnostic confidence (20). PCASL also holds the potential to provide hemodynamic information, and label the selected vessels similar to X-ray DSA (21) (22). Unfortunately, current intracranial angiography is limited to 2D projection imaging or low resolution 3D imaging due to scan time limitations. For example, recent 2D projection imaging has required 11 seconds per slice (21), which translates to scan times over 30 minutes for true 3D imaging even with parallel imaging.

In this work, we propose an accelerated PCASL-based 3D MRA technique that utilizes a 3D radial acquisition technique called VIPR (23). Benefiting from the highly sparse image volume created by PCASL, and noise-like artifacts specific to 3D radial acquisition, PCASL-VIPR can achieve whole head coverage and sub-millimeter isotropic spatial resolution within clinically acceptable scan time. Feasibility studies have been conducted in both normal subjects and patients with both qualitative and quantitative comparison against 3D TOF.

Methods

Sequence

PCASL angiography is performed with an interleaved acquisition set to two tagging states, control and label, which are subtracted to yield an angiographic image. Each tag session consists of four modules: background saturation, PCASL, PASL, and imaging, as illustrated in Figure 1.

Figure 1.

Labeling geometry (left, white dashed box indicates imaging slab, dashed line indicates the labeling plane); PCASL-VIPR sequence diagram (right) shows a tag session consisting of four modules: background suppression (BGS), PCASL, FAIR and image acquisition with time assignment. PCASL is set to label state or control state to acquire the label image and the control image respectively.

For background suppression, we utilized a VERSE transformed hyperbolic secant pulse (24) to selectively invert the imaging slab. This helps to reduce the signal from cerebrospinal fluid (CSF) and other background tissue which give rise to artifacts.

The PCASL module was implemented following a balanced gradient approach (19) using a train of pulsed radiofrequency (RF) and gradients as illustrated in Figure 1. In the label state, the RF phase cycling is set such that spins at the labeling plane see RF pulses with the same phase. In this condition the spins passing through this labeling plane undergo adiabatic inversion. In the control state, the RF pulse train and the gradients are the same as in the label state, while the RF phase is cycled such that spins at the labeling plane see RF pulses having a phase of π relative to the previous pulse, leading to limited effect on the passing spins. In our implementation, a Hanning window-shaped RF pulse of 500 us duration was performed periodically with a 1200 μs spacing between RF pulses. The amplitude of the RF pulse and gradients were optimized using Bloch simulation to get a labeling efficiency greater than 95%. The parameters we used in this study are as follows: average gradient=0.78 mT/m, maximum gradient = 7 mT/m, average B1=1.63 μT.

At the end of PCASL labeling, inflowing spins superior to the labeling plane are inverted. However, fresh spins will displace labeled spins during the imaging module causing signal loss in proximal vessels. This could be avoided by setting a large gap between the labeling plane and the inferior edge of the image region. However, this will lead to substantial reduction in SNR and adds a transit time parameter that will vary from subject to subject depending on the flow velocity. To avoid signal loss with SNR penalty we followed the scheme in Ref 21 to incorporate a PASL scheme at the end of PCASL to ensure a continued inflow of labeled blood during image acquisition. In our study, flow-alternating-inversion-recovery (25) (FAIR) is implemented such that a selective inversion pulse is applied on the imaging slab in label state, while a global inversion pulse in control state. By doing this, the label image contains uninverted blood spins while the control image contains inverted blood spins. The tissue spins are inverted in both images and cancelled out in the final subtracted image.

Both bSSFP and SPGR have been utilized for non-contrast angiography and can be used as readout with PCASL. bSSFP provides substantially higher signal; however bSSFP is highly sensitive to off-resonance, leads to high SAR, and can be sensitive to flow related artifacts (26). Off-resonance artifact are a particular issue in intracranial applications, due to susceptibility at the skull base and nasal fossa, movement of spins through the off-resonant field, and higher utilization of 3T scanners. To avoid these artifacts, the acquisition module consists of a low flip angle SPGR readout combined with a VIPR sampling strategy.

The VIPR sequence samples data along radial lines evenly spaced through a spherical volume each intersecting the origin of k-space. For PCASL angiography, VIPR has two major advantages over Cartesian acquisitions. First, undersampling artifacts appear as a diffuse low level background noise rather than ghosting artifacts in Cartesian acquisitions (27). This allows for undersampling acceleration when the image is sparse. The subtractive nature of PCASL creates a highly sparse imaging volume allowing the high acceleration factors needed to keep the scan time in a clinical acceptable range for 3D scans. Second, 3D radial sampling is more robust to contrast changes during the acquisition. During the readout, the signal experience T1 recovery and RF saturation due to repetitive RF pulses. This signal behavior approximately follows the exponential decay curve for spins in the volume. Previous work with Cartesian angiography (13) and fast spin echo imaging have well characterized the response to the decay as an apodization in k-space which results in the loss of spatial resolution. For 3D radial acquisition, the signal modulation leads to an increased angular undersampling artifact in the image as demonstrated in previous applications of radial acquisition for hyperpolarized He³ MRI (28) and fast spin echo (29,30). This is merely due to the fact that the decay is evenly distributed between the edge and center of the k space. The additional diffuse artifacts has no effects on the image resolution which makes VIPR advantageous for angiography where high resolution is critical.

Evaluation

5 healthy volunteers (1 women, 4 men, age range 25–35) and 5 patients with previously diagnosed AVMs (N=3) or aneurysms (N=2) were imaged after obtaining Institutional Review Board (IRB) approval and informed consent.

All exams were performed on a clinical 3T MR system (Discovery 750, GE Healthcare, Waukesha, WI, USA) with a 32-channel head coil (Nova Medical, Wilmington, MA, USA) for reception and body coil for transmission. Subjects were imaged with a clinical standard 3D TOF scan and a whole brain PCASL-VIPR scan. The standard 3D TOF scan was acquired with the following parameters: TR/TE 30/2.8; field of view (FOV) 22×22 cm²; matrix 512×256; slice thickness 1mm; flip angle 20°; bandwidth ±41.67 kHz; number of slabs 4 (an exception is the case with a basilar tip aneurysm where only 2 oblique slabs were acquired and the scan time is 4 minutes); slab thickness 40 mm; overlap thickness 10 mm; parallel imaging method SENSE (ASSET, GE, Healthcare, Waukesha, WI) with an acceleration factor of 2×. For partial head coverage of 11 cm, the examination time was 8:01 minutes.

For PCASL-VIPR scan, the labeling plane is approximately placed at C1 segment of the internal carotid arteries such that the major feeding arteries of the brain vasculature including internal carotid arteries, external carotid arteries and vertebral arteries will be labeled. A 16cm-thick imaging slab was prescribed right above the labeling plane covering the whole brain. PCASL-VIPR parameters include: labeling duration 3 s; image acquisition window 1 s; FOV 22×22×16 cm³; 3D isotropic resolution 0.68 mm; readout bandwidth ±62.50 kHz; TR/TE (ms) 5.06/1.07; fractional echo 0.75; flip angle 10°; no flow compensation. A total of 12,000 projections were collected in a scan time of 8:27 minutes. Compared to Nyquist this represents a 13× undersampling. Data acquired for the label image and control image are first subtracted in k-space then reconstructed utilizing an optimized gridding routine (31) zero-filled to 0.46mm isotropic resolution. Individual coil images were combined utilizing coil sensitivities estimated from the center of k-space (32). This coil combination helps reduce undersampling artifacts as in partially parallel imaging (PILS) (33) and leads to more optimal SNR in the final image (34).

Images from 3D TOF and PCASL-VIPR were evaluated quantitatively and qualitatively. For qualitative evaluation of image quality, two experienced readers (one senior neuroradiologist and one senior investigator with over 25 years of experience in neurovascular imaging research) were presented with source images from both 3D TOF and PCASL-VIPR scans on a PACS workstation. The grading was not blinded as the PCASL and 3D TOF exams could be readily distinguished by the presence or absence of slab artifact and residual background signal on the 3D TOF studies. PCASL-VIPR and 3D TOF examinations were evaluated using two criteria (vessel visualization and saturation artifacts), defined on a four-point scale (Table 1). Eight vessel segments were evaluated for each subject, while pathology was evaluated only for the patients. Pearson correlation test was performed to assess the inter-rater correlation of the grading of the two readers. The nonparametric Wilcoxon signed rank method was used to test for a significant difference in image quality between PCASL-VIPR and 3D TOF for each criterion, where P<0.05 was taken to be statistically significant.

Table 1.

Criteria for qualitative evaluation.

Criterion	Interpretation	Score
Vessel Visualization	Non-Diagnostic or Not Visible	1
	Poor (Structures visible, but with significant blurring or artifacts, non diagnostic)	2
	Good (Good quality diagnostic information, minimal blurring or artifacts)	3
	Excellent (Excellent quality diagnostic information, sharply defined borders)	4
Saturation Artifacts	Severe signal dropouts. Not diagnostic.	1
	Moderate signal dropouts. Loss of diagnostic accuracy.	2
	Minimal presence of signal dropouts. Does not interfere with diagnostic interpretation.	3
	No significant saturation artifacts.	4

For quantitative evaluation, regions of interest (ROIs) were placed on the source images of the five healthy subjects on the following locations: siphons (left/right), basilar tip, carotid terminus (left/right), first bifurcation of middle cerebral artery (left/right), 5-mm-region around the anterior communicating artery, and background tissue. Contrast-to-noise ratio (CNR) were measured with equation CNR=(V-B)/σ, where V represents the maximum signal in ROIs, B represents the average signal in background ROIs, and σ is the standard deviation of the signals in background ROIs. For PCASL-VIPR images, σ includes the noise and also undersampling artifacts which have noise-like appearance. The student's t test was performed to test for a significant difference in mean CNR between PCASL-VIPR and 3D TOF for each vessel groups.

Results

Figure 2 shows representative PCASL-VIPR images obtained in a healthy subject. With near-zero background, whole volume Maximum-Intensity-Projection (MIP)s could be generated at any angle. The axial, coronal, and sagittal collapsed views are presented. High spatial resolution enables visualization of small distal arteries with size of sub-millimeter. Signal intensity decays from proximal to distal vessels due to T1 recovery and RF saturation. Whole brain coverage is achieved with a 16-cm slab thickness.

Figure 2.

Axial, coronal and sagittal MIPs of PCASL-VIPR of a Healthy subject (for better visualization, a few slices at the inferior edge of the 3D image volume were excluded when performing axial MIP). Images show complete filling of the entire arterial vasculature of the brain. Signal intensity decreases from proximal to distal vessels due to increasing transit time from the labeling plane.

Figure 3 shows results of a patient with a thalamic AVM with deep venous drainage. Limited MIPs in the sagittal view of the same thickness and location were performed for the source images of both PCASL-VIPR and 3D TOF. Images were cropped to compare with X-ray DSA image which was also performed for this patient as the reference standard. The images show increased flow through the left posterior cerebral artery and the venous drainage into the vein of Galen. The structures of the AVM were equally visualized in both exams with delineation of the main feeding arteries confirmed by X-ray DSA. However, severe signal loss in the draining vein due to saturation can be observed in the TOF image.

Figure 3.

Thalamic AVM patient: sagittal limited MIPs of 3D TOF (left), PCASL-VIPR (middle) and X-ray DSA (right). The arrows point to the venous drainage of the AVM into the vein of Galen. MIP image of 3D TOF show severe signal loss in the draining vein due to saturation artifacts.

Figure 4 shows results of a patient with left frontal AVM. From the limited sagittal MIP images of both exams, it is easy to distinguish the left anterior cerebral artery as the dominant supply to the AVM. High signal in the feeders in the PCASL-VIPR MIP image suggests short transit time from the labeling plane due to fast flow. There is still mild signal loss due to saturation in the feeding arteries in the TOF image. The superficial venous drainage can be observed in both exams.

Figure 4.

Left frontal AVM patient: sagittal limited MIPs of 3D TOF (left) and PCASL-VIPR (right). The arrows point to the major feeding arteries of the AVM. Saturation artifacts can be observed in the feeding arteries in TOF image.

Figure 5 shows results of a patient with a small aneurysm located at the tip of the basilar artery. The limited sagittal MIP images and the zoomed in view at the location of the disease of both exams show excellent depiction of the aneurysm. Because of the fast flow in the aneurysm and small size, little saturation artifact was seen in TOF image that would affect the visualization of the aneurysm. The inflow jet of the aneurysm which is shown as a bright region (arrows) is depicted in both images.

Figure 5.

Patient with an aneurysm at the tip of the basilar artery: sagittal limited MIPs (upper row) and the zoomed in view of the aneurysm (lower row) of 3D TOF (left) and PCASL-VIPR (right). Both exams show excellent visualization of the aneurysm. Little saturation artifact was seen in TOF image due to the fast flow in the aneurysm and small size. The bright region (arrows) depicts the inflow jet.

Figure 6 shows results of a patient with a large aneurysm arising from the left cavernous internal carotid artery. Limited coronal MIPs of both exams are shown at the aneurysm location. A profile was drawn along a horizontal line (indicated with a white line) through the aneurysm for the two MIP images. Signal intensity was scaled for the two profiles to be compared together. The profile of TOF shows signal drop both in the region of the aneurysm and the left carotid artery compared to the left side. This is likely caused by the saturation artifacts due to the slow, recirculating flow inside the aneurysm. In the profile of PCASL-VIPR, signal is more uniform in the aneurysm and the signal drop due to slow flow is less compared to TOF. In this case, the labeling plane was unnecessarily lower (around the neck) than the other cases which caused longer transit time for the inverted spins to reach the disease. The signal drop due to T1 recovery could also be observed in the basilar artery where the flow is slower than the internal carotid arteries. This kind of signal loss could be easily diminished by proper adjustment of the labeling plane and higher signal in the aneurysm could have been obtained as well. Another reason that could result in signal loss is the reduced labeling efficiency due to off-resonance effects caused by field inhomogeneities at the labeling plane, but that hasn't been demonstrated in this case.

Figure 6.

Patient with an aneurysm arising from the left cavernous internal carotid artery: cropped coronal limited MIPs of 3D TOF (upper) and PCASL-VIPR (lower). PCASL-VIPR image shows stronger and more uniform signal inside the aneurysm and less blurring artifacts around it. The right figure shows the profile along the horizontal line across the aneurysm indicated in the left upper corner small image. Different vessels are labeled in the profile. 3D TOF profile (starred line) shows signal drop in the aneurysm and left carotid due to saturation of slow and recirculating flow. PCASL-VIPR shows weaker signal in the aneurysm and basilar artery due to the placing of the labeling plane.

Table 2 summarizes the results from the qualitative evaluation. Pearson's correlation coefficient is 0.8, and the grading scores of two readers were averaged. For vessel visualization, PCASL-VIPR was superior in three of the eight vessel groups and pathology (positive values) while 3D TOF was slightly better in the other five vessel groups (negative values). However, there was no statistical significance between any of these groups. For saturation artifacts, PCASL-VIPR was superior in four of the eight vessel groups and in the delineation of pathological conditions with a statistically significant advantage in visualization of the siphons (bold and underscored). 3D TOF was superior with no statistical significance in the remaining vessel comparisons. The scores were also averaged over all the evaluated vessel segments excluding pathology, and PCASL-VIPR performs slightly better in both criteria with no statistical significance.

Table 2.

Results of qualitative evaluation.

Evaluated Vessel Segments	Vessel Visualization (1–4)			Saturation Artifacts(1–4)
	TOF	PCASL	PCASL-TOF*	TOF	PCASL	PCASL-TOF*
Anterior cerebral artery	3.25	3.10	−0.15	3.20	3.25	0.05
Right posterior cerebral artery	3.22	3.11	−0.11	3.22	3.06	−0.16
Left posterior cerebral artery	3.10	3.05	−0.05	3.15	3.05	−0.10
Right middle cerebral artery	3.35	3.30	−0.05	3.35	3.25	−0.10
Left middle cerebral artery	3.35	3.30	−0.05	3.35	3.25	−0.10
Basilar tip	3.40	3.45	0.05	3.35	3.50	0.15
Right siphon	3.30	3.60	0.30	3.30	3.85	0.55
Left siphon	3.25	3.60	0.35	3.30	3.85	0.55
Pathology	2.80	3.50	0.70	3.00	3.60	0.60
All the above vessels excluding pathology	3.28	3.32	0.04	3.28	3.39	0.11

^*Score difference between PCASL-VIPR and 3D TOF. Positive values mean better performance of PCASL-VIPR over 3D TOF, negative values mean the opposite. Statistical significance as tested using a Wilcoxon signed rank test is indicated in bold font (P < 0.05)

Figure 7 shows the average CNR (mean ± SD) of PCASL-VIPR and 3D TOF across all subjects' vessel segment ROIs. Compared to 3D TOF, PCASL-VIPR shows higher CNR in the siphons and basilar artery (starred) with statistical significance (p value <0.05) and equal values in the others. The CNR values are also more consistent across the subjects in PCASL-VIPR. In addition, PCASL-VIPR shows a trend of signal diminishing from the siphons to the anterior cerebral arteries with increasing transit time.

Figure 7.

CNR (mean ± SD) graph of PCASL-VIPR and 3D TOF measured from five ROIs: left/right siphons, basilar tip, left/right carotid terminus (car ter), first bifurcation spot of the middle cerebral arteries (mca) and 5-mm-region around the anterior communicating artery (a comm). CNR of PCASL-VIPR is higher than 3D TOF in siphons and basilar tip (starred) with statistical significance tested by t test. It also shows signal decreases in moving from proximal to distal vessels. CNR of 3D TOF shows no flow related decay pattern; the standard deviation is higher (less consistent across the subjects).

Discussion

We have developed a 3D non-contrast-enhanced MR angiography technique that combines ASL tagging with PCASL and an accelerated 3D radial acquisition technique, VIPR. PCASL-VIPR has been tested for intracranial angiography in both healthy and diseased subjects.

Compared with 3D TOF, PCASL-VIPR was found to hold several advantages over it as demonstrated by the images of the feasibility study. Ignoring the imperfect equalization of MT effects and the tissue perfusion signal, static tissue should be completely cancelled out by subtraction procedure. With a near zero background, vessel visualization is limited by noise in the background rather than contrast with neighboring structures. PCASL has much less sensitivity to slow flow while 3D TOF suffers from saturation artifacts due to slow or in-plane flow patterns. When combined with a 3D radial trajectory, PCASL-VIPR can easily cover the entire head without the extension of scan time and achieve 3D isotropic spatial resolution simultaneously. According to our evaluation results, whole head PCASL-VIPR provides approximately equal image quality as 3D TOF in healthy subjects. When applied to patients especially, AVM patients, PCASL-VIPR provides better image quality and diagnostic information over 3D TOF which suffers from the artifacts due to saturation and multi-slab coverage. The recent large group study that investigated the reliability of using 3D TOF to follow up the obliteration process of AVMs (35) showed its insufficiency when the remaining nidus diameter is < 10 mm. The reasons include spin dephasing due to complex or turbulent flow, saturation due to slow flow, and lack of temporal resolution. Our preliminary study has already shown PCASL's advantage of less sensitivity of slow flow. We also expect better performance of PCASL-VIPR in the delineation of AVMs when it's applied to dynamic angiography. In future study, systematic analysis of the AVM components with large number of subjects will be conducted.

PCASL-VIPR holds several advantages over previously developed 3D PASL angiography techniques (13,16,17). With PASL, the delay time between spin labeling and image acquisition needs to be chosen for the tradeoff between ensuring vasculature filling and providing adequate SNR. This parameter is subject dependent, and the selection requires additional scans and also induces other artifact-issues, even with the retrospective methods. For this work, we were able to utilize a long label duration (3 s) which ensured adequate vessel filling. Longer label duration could be utilized, but have diminished gains due to T1 recovery of the tag. Second, the vessel segments with different transit times experience different amount of T1 recovery in PCASL angiography varying from minimum (the proximal vessels) to maximum (the most distal vessels). In PASL angiography, all the vessel segments experience the same amount of T1 recovery which is determined by the longest transit time. Therefore, PCASL angiography provides higher SNR than PASL angiography.

Previously reported PCASL angiography (21,22) has demonstrated utility in acquiring hemodynamic vessel-specific information in intracranial angiography. However, these techniques are limited to 2D projection imaging or multiple thick-slab imaging, and cannot be easily extended to 3D imaging due to long scan times. With the accelerating benefits from the 3D radial trajectories, PCASL-VIPR is the first method to achieve high 3D isotropic spatial resolution which greatly improves the vessel visualization compared to the 2D techniques. PCASL-VIPR also holds the potential to depict temporal dynamics and label individual vascular beds. Temporal dynamics can be achieved by varying the label duration prior to imaging. Although dynamic CE-MRA is also able to perform time-resolved imaging, temporal resolution and/or spatial resolution have to be compromised to match imaging data acquisition with the short time period of the first pass of contrast agent. Additionally, the intravenous injection protocols lead to substantial bolus dispersion limiting the benefits of improved temporal resolution. For PCASL based dynamic angiography, since the label duration is freely adjustable, inflow dynamics can be depicted with resolutions commonly only achieved with X-ray DSA. In practical use, the optimal temporal resolution is chosen to couple the sampling window which controls the amount of leading edge blurring due to the continuously advancing bolus and the diffuse artifacts caused by the inconsistency of the data. Although shortening the sampling window will proportionally increase the total scan time, the image with a short label duration will have greater sparsity and higher CNR to support higher acceleration factor when using VIPR acquisition. PCASL-VIPR is also capable of doing vessel-selective imaging based on PCASL's ability to target a specific feeding vessel, analogous to selective catheter injections in X-ray DSA. Time-resolved hemodynamic, and vessel-specific information similar to that obtained with DSA, without the use of exogenous contrast agents, make PCASL-VIPR a promising alternative to X-ray DSA.

The efficacy of PCASL-VIPR is primarily limited by long scan times, T1 recovery, RF-saturation, and labeling efficiency. Long scan times introduce sensitivity to motion artifacts and hinder time resolved imaging. In this work, we have not utilized parallel imaging or compressed sensing (36) which could be utilized to help achieving higher acceleration factors. The signal difference of PCASL-VIPR is directly related to the labeling efficiency of PCASL. Efficiency can be lost by imperfect phase tracking at the labeling plane. Phase tracking errors arise from off-resonance field at the labeling plane due to imperfect shimming or the susceptibility from eddy currents or gradient waveform errors. Corrections can be made if the amount of off resonance and gradient errors can be mapped at the labeling plane (37,38), but requires substantially more imaging time. Higher order shimming might also be useful and will be tested in future study.

While signal saturation is substantially reduced compared to 3D TOF, signal loss is still observed in distal vessels in PCASL-VIPR. Due to T1 recovery, signal decreases exponentially with the transit time from labeling plane. This limits the extent to which slow inflow can be imaged. In addition, signal saturation occurs in distal vessels due to the application of multiple RF pulses within an SPGR readout cycle. Spin that are not refreshed during the readout by inflow experience more RF saturation. This effect is more severe when the readout flip angle increases and/or the acquisition window extends. The residual tissue signal at the inferior edge of the image volume as can be seen from the coronal and sagittal MIP images came from two sources: the selective inversion performed over the imaging volume and PCASL labeling at the edge of the volume. With PCASL, the tissue spins at the labeling plane experience (α, α) RF pulses in the label phase, and (α, -α) RF pulses in the control phase. Therefore, the saturation effects caused by the two different RF pulse train are not equal. Furthermore, the selective inversion has a profile that will result in incomplete subtraction at areas just outside the imaging FOV. Subtraction of the label phase and control phase is not able to cancel out the tissue signal at the labeling plane and leaves some residual tissue signal at the inferior edges of the 3D image volume. Both of these errors would be mitigated by placing the labeling plane more inferior, at the cost of reduced flow sensitivity.

The use of balanced SSFP as seen in other non CE-MRA methods could provide higher SNR and reduced RF saturation effects. However, SSFP is generally more prone to off-resonance artifacts that are more prevalent when imaging at higher field strength. In this work, the choice of SPGR avoided this issue and also reduced the specific absorption rate compared to balanced SSFP. All acquisitions in this study were obtained at 3T, which provides higher SNR over 1.5 T and decreased T1 recovery thanks to longer T1.

In conclusion, a novel non-contrast-enhanced 3D MRA technique is presented in this study. The feasibility of this technique was validated for intracranial angiography. Other potentials such as time-resolved imaging and vessel selective imaging will be developed in future work. PCASL-VIPR will be further validated by being applied to larger patient groups and compared to conventional x-ray digital subtraction angiography.