Cervical spinal cord angiography and vessel-selective perfusion imaging in the rat

Abstract

Arterial spin labeling (ASL) has been widely used to evaluate arterial blood and perfusion dynamics particularly in the brain, but its application to the spinal cord has been limited. The purpose of this study was to optimize vessel-selective pseudo-continuous arterial spin labeling (pCASL) for angiographic and perfusion imaging of the rat cervical spinal cord. A pCASL preparation module was combined with a train of gradient echoes for dynamic angiography. The effects of echo train flip angle, label duration, and a cartesian or radial readout were compared to examine their effects on visualizing the segmental arteries and anterior spinal artery (ASA) that supply the spinal cord. Lastly, vessel selective encoding with either a vessel-encoded pCASL (VE-pCASL) or super-selective pCASL (SS-pCASL) were compared. Vascular territory maps were obtained with VE-pCASL perfusion imaging of the spinal cord, and the inter-animal variability was evaluated. The results demonstrated that longer label durations (200 ms) resulted in greater signal to noise ratio in the vertebral arteries, improved conspicuity of the ASA, and showed better quality maps of blood arrival times. Cartesian and radial readouts demonstrated similar image quality. Both VE-pCASL and SS-pCASL adequately labeled the right or left vertebral arteries which revealed the inter-animal variability in the segmental artery with variations in their location, number, and laterality. VE-pCASL also demonstrated unique inter-animal variations in spinal cord perfusion with a right-sided dominance across the n=6 animals. Vessel-selective pCASL successfully achieved visualization of the arterial inflow dynamics and corresponding perfusion territories of the spinal cord. These methodological developments provide unique insights to the inter-animal variations in the arterial anatomy and dynamics of spinal cord perfusion.

Graphical Abstract

Vessel-selective pseudocontinuous arterial spin labeling (pCASL) was optimized for angiographic and perfusion imaging of the rat cervical spinal cord. Vessel-selective pCASL successfully visualized the arterial inflow dynamics and corresponding perfusion territories. These methodological developments provide unique insights to the inter-animal variations in the arterial anatomy and dynamics of spinal cord perfusion.

Introduction

Imaging of the neurovasculature and tissue perfusion has critical importance in many diseases or injury states, and while MRI techniques have been developed and used routinely for the brain, their application to the spinal cord have been far more limited. In particular, arterial spin labeling (ASL) allows noninvasive measurement of both arterial blood flow and tissue perfusion by magnetically labeling blood water as an endogenous tracer¹. Pseudo-continuous arterial spin labeling (pCASL) is the recommended ASL technique due to its high signal to noise ratio (SNR) and less burden on clinical hardware². Recently, pCASL was optimized for the rat cervical spinal cord to evaluate spinal cord blood flow (SCBF) as a biomarker for acute spinal cord injury (SCI)³ and has also been demonstrated in the rat thoracic cord⁴. The goal of the current work was to refine pCASL methods to visualize the dynamics and anatomical variability of the feeding arteries of the rat cervical cord for applications to both angiography and perfusion.

For spinal cord angiography, pCASL was previously optimized to visualize the arterial vasculature supplying the rat cervical spinal cord using combinations of label duration (LD) and post-label delay (PLD)³. While that study successfully demonstrated the unique vascular pattern in the cervical spinal cord, the acquisition of images across separate scans led to poor efficiency with long scan times and was only performed in a single animal. In the brain, time-efficient pCASL angiography techniques have been demonstrated, some of which achieve high levels of acceleration using compressed sensing and sparsity information⁵. Several of these methods also exploit radial k-space sampling for improved robustness to motion, distortion, and blurring, which could be advantageous for the spinal cord. Golden angle radial sampling also has the additional advantage of sub-sampled retrospective reconstruction for time-resolved 4D dynamic MR angiography to capture arterial phases and provide excellent delineation of the vasculature with shorter scan times⁶. For example, the Combined Angiography and Perfusion using Radial Imaging and ASL (CAPRIA) approach captures dynamic angiography and perfusion⁷, and golden angle radial stack-of-stars acquisitions have been evaluated in patients with arteriovenous malformations (AVM) at 7T ⁸. Since golden angle radial sampling has not been yet studied for the spinal cord angiography, previous developments of the technique for more efficient and robust sampling call for similar optimizations to the spinal cord.

Beyond perfusion and angiography, revealing the vascular territories of the spinal cord also has potential clinical value. Post-mortem examinations in both rodent⁹ and human cadaveric specimens¹⁰ demonstrate a high degree of individual variability of the spinal cord arterial anatomy. In the cervical spine, the vertebral arteries are the predominant origin of segmental or radicular arteries that supply the spinal canal. These rejoin together to form the anterior spinal artery (ASA) which in turn is the major supply to the spinal cord gray matter. The spinal cord vasculature is highlighted by a high degree of anastomosis and collateral flow, and its individual variation has important implications for surgical interventions¹¹.The number, laterality, and level of segmental arteries are variable across individuals but a slight right dominance¹² and concentration of segmental arteries in the mid-cervical levels (C2-C5) have been identified^9,10. To reveal the vascular anatomy and perfusion territories using ASL, two vessel-selective techniques based on pCASL have been developed with consistent applications to the brain: vessel-encoded pCASL (VE-pCASL)¹³ and super-selective pCASL (SS-pCASL)¹⁴. In VE-pCASL, periodic modulation of gradient blips between the labeling radiofrequency pulses imparts a differential inversion efficiency across two vessels of interest along a single axis, which can be further combined using encoding strategies for multiple vascular territory imaging^15,16. In SS-pCASL, single-artery labeling is achieved by dynamically rotating the in-plane gradients to maintain inversion efficiency in a small focal spot centered on the artery of interest¹⁴. These methods have not been previously applied to the spinal cord in either humans or animals.

The goal of this work was to optimize pCASL to reliably measure spinal cord arterial anatomy and perfusion in the rat cervical spinal cord. First, we combined pCASL preparation with multiple gradient echo readouts for efficient and dynamic information, with an initial emphasis on the vertebral arteries. Subsequently, we compared Cartesian and radial sampling to capture the segmental artery anatomy. Vessel selective encodings of the left or right vertebral arteries were compared using VE-pCASL and SS-pCASL. Finally, maps of spinal cord perfusion using VE-pCASL were demonstrated. Collectively, the optimizations advance the methods available for spinal cord angiography and perfusion imaging, and further demonstrate the unique inter-animal variability in spinal cord vasculature that may have important applications if extended to the human spinal cord.

Materials and Methods

Animals

All animal procedures were approved by the Institutional Animal Care and Use Committees (IACUC) at the Medical College of Wisconsin and the Clement J. Zablocki Veterans Affairs Medical Center. A total of 7 healthy female rats (age range 1 – 6 months) were used for the experiments. Group sizes are indicated for each set of results.

MRI

MRI was performed on a Bruker 9.4T Biospec System (Bruker Biospin) operating Paravision version 6.0.1. Animals were positioned prone in a custom head holder with a bite bar inside a 3.8 cm diameter Litzcage coil (Doty Scientific, Inc). Animals were anesthetized with 4-5% isoflurane for induction and maintained at 2-2.5% during imaging with a target of 40-50 breaths per minute for stable physiology monitored using a respiratory pillow. Electrocardiogram (ECG) electrodes were utilized for monitoring animal physiology and for gating.

Pseudo-continuous arterial spin labeling (pCASL)

A pCASL module used gradient and radiofrequency parameters previously optimized for rodent brain¹⁷ and spinal cord perfusion³: 400 μs Hanning pulse, 0.8 ms interpulse spacing ($\Delta t$), B1_ave=5 μT, G_max=45 mT/m, and G_ave=5 mT/m. The animal was positioned within the magnet ensuring that the C7 level was 11.7 mm off-isocenter ($\Delta z$) which maintained pulse phase increments ($\Delta \phi$) of 0° for labeling and 180° for the control condition according to Dai et al¹⁸ ($\Delta \phi =\gamma G_{\text{ave}}\Delta z\Delta t$). No additional pulse phase corrections were used to account for magnetic field inhomogeneity effects¹⁷. Map based shimming was used for all studies with both the label plane and spinal cord, contained within the same shim voxel.

A dynamic readout was used that included a series of spoiled gradient echoes acquired after the pCASL preparation, here defined as a single segment (Fig. 1A). Either a single-slice cartesian or golden angle radial acquisition was used with the label and control conditions acquired sequentially. For cartesian imaging, each phase encoding step was acquired multiple times within each segment, resulting in images with different effective post-label delays (PLDs) (Fig. 1B). The number of total segments was equal to the number of phase encoding steps. For radial imaging(Fig. 1C), the trajectory (i.e. spoke) for each of the gradient echo readouts was rotated by a multiple of the golden angle (111.25°). Successive segments were also shifted in time and spokes rotated to ensure that over the full acquisition, all spokes maintained a linear spacing of PLDs with a unique angle. Although the radial acquisition also permits binning of spokes to reconstruct a dynamic series, it was not used in these studies. Additional scan parameters are reported for each set of experiments.

Figure 1.

Pulse sequence and k-space ordering. A: Each readout segment consisted of multiple spoiled gradient echo acquisitions and was preceded by an ASL preparation, with label and control images obtained consecutively. B: In a cartesian acquisition, each gradient echo separated by time TR_inner repeatedly acquired the same line of k-space for different effective post-label delays. Segments were repeated for all phase encoding steps. C: In a radial acquisition, each gradient echo was rotated to acquire a spoke through the center of k-space. Segments were repeated and shifted in time to obtain a unique set of spoke angles and effective post-label delays over the full acquisition.

Dynamic pCASL Angiography

Previous results in the rat cervical cord³ using separate scans each with a different combination of label duration (LD) and PLD demonstrated transit times of approximately 200 ms within the anterior spinal artery (ASA). First, we acquired 25 PLDs per segment to confirm similar temporal effects with this dynamic readout in a single animal using the following parameters: TR= 300 ms (determined by gating), TE=2.5 ms, TR_inner=12 ms, resolution=0.313x0.313 mm², matrix=128x128; FOV=40x40 mm², slice thickness=2 mm, flip angle=15°. A saturation pulse positioned above the label plane (pre-sat) suppressed the tissue signal and reset any residual magnetization. The pre-sat pulse was triggered at 50 ms after the RR wave. A chemical shift-based fat suppression immediately preceded each of the excitation RF pulses. Based on these images, all subsequent angiography experiments used 16 PLDs for a total readout window (i.e. segment duration) of 192 ms and a scan time of 3:30 minutes (varied by gating) except as otherwise noted.

The use of a constant flip angle (CFA) for each gradient echo was further compared to a variable flip angle (VFA) scheme that maintains more consistent signal within a segment¹⁹. Experiments compared CFA (10°, 15°, 30°, and 50°) and VFA (Max=50°) using the cartesian readout.

The effects of two different label durations were also compared. Incorporating VFA, an LD of 50 ms was used to visualize both the inflow and outflow of the bolus in the vertebral arteries, whereas an LD of 200 ms prolonged the bolus within the arteries and improved contrast in the distal arteries including the ASA. All other imaging parameters were identical except TR=420 ms.

Subsequently, the effect of radial or cartesian trajectories were compared using both VFA and an LD of 200 ms. Both acquisitions used a higher resolution (0.156x0.156 mm²; 192x192 matrix) that also necessitated a longer echo time (3.9 ms), and flow compensation was used along all axes. The number of segments for both acquisitions was matched for direct comparison and each scan required 4:27 (varied by gating) minutes due to their higher resolution.

Vessel-selective Angiography and Perfusion

Vessel-selective pCASL modules included additional gradients in the in-plane directions between the radiofrequency pulses. For VE-pCASL, these were adjusted to label the left and right vertebral arteries at the C7 label plane which were separated by approximately 6 mm on average and resulted in a gradient moment of approximately 1.5 mT/m·ms, depending on the precisely measured separation for each animal¹³. Four images were acquired with the left or right arteries labeled in addition to a non-selective and a control labeling condition. For non-selective and control labeling, the in-plane gradients were omitted. For SS-pCASL, the focal spot of the labeling was centered on the left or right vertebral artery and used the same 1.5 mT/m·ms gradient moment. Similarly, separate images were acquired with the left, right, non-selective, and control labeling. Two separate scans used either a coronal or axial imaging plane.

Perfusion imaging with VE-pCASL was evaluated in six animals. A longer LD (600 ms) was used along with a global pre-saturation pulse (3850 ms before excitation) and an inversion pulse (1550 ms before excitation) to specifically null the cerebrospinal fluid signal. The image acquisition used an 8-shot rapid acquisition with refocused echoes (RARE) readout using center-out k-space encoding order (TR/TE = 4000/4.15 ms, resolution = 0.203 x 0.203 mm², FOV = 29.2 x 26.0 mm², slice thickness = 2 mm, 8 repetitions). Two spatial saturation bands were placed dorsal and ventral to the spinal cord to suppress motion-induced ghosting and no physiological gating was used. VE-pCASL perfusion images included four images (left-VA, right-VA, nonselective, and control) which were acquired with 2 separate PLDs (200 and 400 ms). An additional image without ASL or inversion preparation was acquired at the end of each series with a longer TR (6000 ms) to estimate tissue proton density (M₀). The total scan time was 20:16 minutes for each PLD.

Data analysis

Images were analyzed using custom Matlab scripts. For angiography imaging, the complex subtraction between label and control were calculated ($\Delta M=\mid (M_L-M_C)\mid$), where $M_L$ and $M_C$ are the label and control signal intensity, respectively. Manual regions of interest (ROIs) in the vertebral arteries were drawn on images to quantify signal to noise (SNR) as the mean value divided by the standard deviation of an ROI placed outside of the tissue. To quantify the signal uniformity along the length of the vertebral arteries, the slope of SNR was derived from a line profile covering C1 to C7 using least squares fitting. To quantify resolution of the vessels, the full width half maximum (FWHM) was obtained from a line contour drawn on the vertebral arteries, avoiding regions with apparent susceptibility artifacts.

To estimate blood arrival times (BAT) in the arteries, Bloch simulations were employed using the CAPRIA tools²⁰ to simulate the label bolus across a range of arrival times, with the T_1blood of 2.38s for 9.4T²¹ and either the 50 or 200 ms label duration. The effects of multiple RF pulses after each ASL preparation were included, whereas the effects of laminar flow or dispersion were neglected. For each voxel, BAT was identified as the simulation having the least-squared difference to the measured dynamic signal. For short labeling durations, both the inflow and outflow contributed to the estimated arrival times whereas for label durations longer than the arrival time, the outflow effects were the predominant feature inherent in the BAT estimation.

Vessel-selective images were decoded to identify contribution from each feeding artery by the pseudoinverse of the encoding matrix (A) as follows:

$$y=A\times x=\left[\begin{array}{ccc}\hfill -1\hfill & \hfill 1\hfill & \hfill 1\hfill \\ \hfill 1\hfill & \hfill -1\hfill & \hfill 1\hfill \\ \hfill -1\hfill & \hfill -1\hfill & \hfill 1\hfill \\ \hfill 1\hfill & \hfill 1\hfill & \hfill 1\hfill \end{array}\right]\times \left[\begin{array}{c}\hfill R\hfill \\ \hfill L\hfill \\ \hfill T\hfill \end{array}\right]$$

Where $y={[y_1{y}_2{y}_3]}^T$ represents the signal measured by each encoding step; the encoding matrix, consisting of −1 and 1 represent inverted (label) and unperturbed (control) arterial spins, respectively; $x={[RLT]}^T$ represents the decoded signals for the right vertebral ($R$), left vertebral ($L$), or tissue signal ($T$). The decoded images with VE-pCASL and SS-pCASL were additionally color-coded for left (red) and right (green) labeling. Maps of inversion efficiency ($IE$) were calculated for each vessel-selective encoding

$$IE=\frac{\mid M_C-M_L\mid }{2\cdot M_C}$$

with complex values from the control and label conditions. ROIs manually placed in each vertebral artery were used for quantification.To examine relative signal changes under label or control conditions, the complex-valued signal intensity from each condition was also normalized to the nonselective control condition resulting values between 1 and −1.

For perfusion imaging, spinal cord blood flow (SCBF) was quantified using the standard kinetic model for pCASL²²

$$SCBF=\frac{6000\cdot \lambda \cdot \Delta M\cdot e^{\frac{PLD}{T_{1b}}}}{2\cdot \alpha \cdot M_0\cdot T_{1b}\cdot \left(1-e^{-\frac{LD}{T_{1t}}}\right)}$$

where $\lambda$ is the blood-brain partition coefficient (0.9 ml/g), $\alpha$ is inversion efficiency measured as 0.80 and 0.67 for nonselective pCASL and VE-pCASL labeling modules, respectively, $\Delta M$ is the pairwise magnitude subtracted perfusion-weighted signal between label and control images, $PLD$ is the post label delay. The longitudinal relaxation time constant of arterial blood ($T_{1b}=2380$ ms)²¹ or spinal cord tissue (${\mathrm{T}}_{1\mathrm{t}}=1380$ ms)³ used constant values from the literature for 9.4T. $M_0$ was estimated from proton-density images acquired at the end of each scan. SCBF maps were computed separately for each PLD.

SCBF maps were spatially registered to a custom template space, with details previously presented^3,24. Briefly, manual single-point landmarks were placed in the cord in native space to define the vertebral levels and seed the spinal cord centerline. Images were registered to a template based on the Atlas of Rat Spinal Cord available at https://osf.io/mesz4/. Automated cord straightening, rigid alignment, and non-linear warping to the template used the Spinal Cord Toolbox²⁵, with the final images resampled at a resolution of 50 x 50 x 150 μm³. Voxelwise analysis was performed with all 6 animals to obtain mean SCBF and the percentages of voxels with left, right, or combined dominance based on their SCBF values from the decoded maps. A template region of interest encompassing the whole spinal cord was used with a lower threshold of 80 ml/100g/min to omit voxels without measurable perfusion, noting the mean SCBF in the isoflurane-anesthetized rat is approximately 200 ml/100g/min³.

Results

pCASL Prepared Dynamic Angiography

To first confirm arterial dynamics and contrast with the dynamic pCASL angiography sequence, 25 gradient echo images with PLDs between 12 and 300 ms were acquired (n=1). Subtracted label and control images revealed blood in the vertebral arteries decayed based on their distance from the label plane (Fig. 2A) and at all levels to within the noise floor beyond approximately 150 ms after labeling (Fig. 2B). Based on these results, subsequent experiments constrained PLD to less than 200 ms.

Figure 2.

Arterial flow changes measured with pCASL in the vertebral arteries across post-label delays (PLDs). A: Control image was averaged across 25 PLDs between 12 and 300 ms. Individual ${\Delta M}_{cplx}$ images were calculated by complex subtraction between label and control images. B: Longitudinal magnetization change at different vertebral level across the PLD (C6: blue, C4: red, C2: green). Each ROI is presented in the control image in Panel A. Note blood signal decayed to within the levels of noise after approximately 150ms after labeling. (n=1)

First, a constant flip angle was compared to a variable flip angle readout train (Fig 3). The VFA readout train had more uniform SNR across the visible vertebral arteries, both spatially and temporally. CFA 30° had high signal in the lower cervical levels (C7) but contrast was low in the higher cervical levels (C2) at the longest PLDs (189 ms). VFA had less contrast at early PLDs but was more consistent over time (Fig. 3A), particularly evident at 189 ms. Across n=3 animals per acquisition (Fig. 3B), and averaged over the cervical levels C1-C7, SNR increased as the CFA flip angle increased from 10 to 50°. VFA had greater uniformity across cervical levels C1-C7 evaluated by linear regression (slope= 0.047, R²=0.12) compared to the CFA₃₀ (slope=−0.114, R²=0.48) but had slightly higher variability compared to CFA₁₀ (slope=0.014, R²=0.03) methods (Fig. 3C). Although the ASA and segmental arteries could be seen in the control images, this was due to a time-of-flight effect and neither CFA nor VFA using the LD of 50 ms showed strong contrast in these more distal vessels using the subtracted angiographic contrast.

Figure 3.

Flip angle comparison. A: Control and mean images were averaged over PLDs (max. 200 ms). Individual ${\Delta M}_{cplx}$ images were calculated by complex subtraction between label and control images. B: SNR measured in the vertebral arteries at C1 through C7, showing the inter-animal variability with different flip angle (FA) conditions. C: SNR change across the vertebral level, demonstrating the spatial variability with different FA conditions. (n=3)

To improve SNR in these arteries, a longer label duration (200 ms; LD₂₀₀) was examined with the VFA readout and compared to the shorter label duration (LD₅₀). The longer label duration provided higher SNR in the vertebral arteries, and the segmental arteries and ASA were also more visible, both within the individual subtracted maps and in the mean image from all PLDs (Fig. 4A). Quantitatively, SNR measured in the vertebral arteries for LD=200 ms (34.57 ± 6.71) was greater compared to LD₅₀ (11.95 ± 0.97) (Fig. 4B). Similar results were evident in the ASA, with SNR for LD₂₀₀ (8.15 ± 1.52) being greater than LD₅₀ (3.33 ± 0.48) (Fig. 4B). A LD₂₀₀ showed greater SNR at each level of the vertebral arteries between C6 and C2, compared to a LD₅₀. (Fig. 4C). Blood arrival times (BAT) were also estimated, and qualitatively, the BAT maps appeared to have greater SNR with LD₂₀₀ compared to LD₅₀. Across all animals (n=6), BAT values along the length of the ASA were slightly longer and more consistent with LD₂₀₀ (106.95 ± 6.78 ms) compared to the LD₅₀ (96.65 ± 8.46 ms).

Figure 4.

Label duration (LD) comparison. A: Individual ${\Delta M}_{cplx}$ images were calculated by complex subtraction between label and control images. Mean images were averaged over PLDs (max. 200 ms). B: SNR measured in the vertebral arteries (VA) at C1 through C3 and in the anterior spinal artery (ASA), showing the inter-animal variability with different LD conditions. C: SNR change across the vertebral level, demonstrating the spatial variability with different LD conditions. (n=6)

Golden angle radial sampling of pCASL prepared angiography was directly compared with the Cartesian readout, both with higher in-plane resolution(Fig. 5A) using LD₂₀₀ and VFA (n=3). Qualitatively, the $\Delta \mathrm{M}$ images of Cartesian and radial sampling were visually comparable with one another, with only minor differences. The SNR in the vertebral arteries (C2 - C6) were not significantly different from one another (25.55 ± 7.90 and 21.26 ± 7.21 for Cartesian and radial, respectively) (Fig. 5B&C). Likewise, the FWHM of the vertebral arteries were comparable between Cartesian (2.42 ± 0.27) and radial (2.58 ± 0.24) acquisition (Fig. 5D). Considering radial sampling is more robust to motion artifact, it was used for further experiments.

Figure 5.

Comparison of image quality between Cartesian and golden angle radial sampling. A: Representative ${\Delta M}_{cplx}$ images of Cartesian and radial sampling. White lines indicate the location of profile lines in Panel B. B: Full width half maximum (FWHM) calculated at the left and right vertebral arteries to compare SNR between Cartesian and radial sampling. C&D: SNR and full-width half-maximum (FWHM) comparison between Cartesian and radial sampling (n=3).

Vessel-selective pCASL Angiography and Perfusion

The two vessel-selective methods, VE-pCASL and SS-pCASL, were compared in their ability to capture additional arterial anatomy beyond non-selective labeling (n=6). Both methods demonstrated clear and consistent labeling of either the right or left vertebral artery (Fig. 6A). First, the arterial contrast for both techniques was similar to nonselective labeling. In the example shown, both vessel-selective methods similarly revealed two segmental arteries originating as branches of the right vertebral artery. Additionally, in the coronal plane, the confluence of the vertebral arteries into the basilar artery was also partially evident. In the axial plane, both vertebral arteries were well-labeled by each technique. Across all animals (n=6), IE quantified in the axial images revealed diminished values in the vertebral arteries for both VE-pCASL (0.67 ± 0.08) and SS-pCASL (0.63 ± 0.05) compared to non-selective labeling (0.80 ± 0.06) (Fig. 6B). By normalizing the signals from all conditions to the non-selective control condition, the decrease in inter-animal mean IE did not appear to be systematic for either the label or control conditions in both of the vessel-specific methods.

Figure 6.

Comparison of nonselective and vessel-selective pCASL labeling. A: Each column represents control, nonselective (NS), vessel-encoded (VE), and super-selective (SS) pCASL labeling conditions. Red and green represents the left and right vertebral arteries. B: Inversion efficiency (IE) values for each condition measured in the vertebral arteries in the axial images. C: The numbers, locations, and laterality of the segmental arteries found with our optimized vessel-selective pCASL (label duration = 200 ms and post-label delay < 200 ms). (n=6)

To examine the segmental artery anatomy, six animals were imaged with the coronal VE-pCASL, and the level and laterality of the segmental arteries were manually counted (Fig. 6C). Twelve segmental arteries were identified across all animals (Table 1), with an average of 2 segmental arteries per animal (range: 1-3). There was no preferred laterality, and the majority of segmental arteries were present at C4 and C5 levels.

Table 1.

Segmental arteries observed using coronal angiography maps

Laterality and location of segmental arteries (n=6)
Animal ID	C1/C2	C2/C3	C3/C4	C4/C5	C5/C6	C6/C7
Rat1				R	R
Rat2			L	L	L
Rat3				L
Rat4			R		LR
Rat5				R
Rat6				R	L

In the same animals, spinal cord perfusion (Fig. 7A) was measured using VE-pCASL in the sagittal plane with 2 PLDs of 200 and 400 ms. The contribution of each artery is shown as red (left) and green (right) corresponding to vascular territories of the left and right vertebral arteries. In this example, a region is highlighted in which the dorsal white matter perfusion originates from the right vertebral artery (green) whereas the gray matter and ventral white matter at the same cervical level originates from the left vertebral artery (red). Across all animals, the SCBF maps with non-selective encoding had mean whole-cord values of 152.8 ± 79.3 ml/100g/min and 168.0 ± 57.2 for PLD 200 and 400 ms, respectively. Further, using the measured inversion efficiency (0.67) for quantification of VE-pCASL, SCBF of the combined left and right maps showed similar values and trends as the non-selective encoding at 159.8 and 171.4, for the PLD of 200 and 400 ms, respectively (Fig. 7C). In the decoded SCBF vascular territory maps, straightened and spatially registered, the unique perfusion patterns across different animals were evident, with pronounced variability in the level and laterality of SCBF (Fig. 7B). Voxelwise analysis of these maps revealed a right-sided dominance with 52.7% of all voxels above the minimum SCBF threshold compared to 35.4% for the left encoding, and only 11.9% of voxels had contributions from both left and right (Fig. 7C). Right-dominant voxels also had larger mean SCBF (232.1 ml/100g/min) compared to left-dominant voxels (172.7).

Figure 7.

Spinal cord perfusion territory mapping. A: Example of perfusion contrast ($\Delta \mathrm{M}$) in native space with nonselective (NS) and vessel-encoded (VE) pCASL labeling showsmost voxels have perfusion that originates in either the right (red) or left (green) vertebral arteries with minimal overlap. An arrow indicates a region of dorsal white matter with perfusion originating on the right side, and with the perfusion of gray matter and ventral white matter at the same cord level originating on the left side. B: Registered SCBF maps from all 6 animals show unique perfusion patterns. C: Whole-cord ROIs values for non-selective and vessel-encoded were similar. Voxelwise analysis reveals the contributions from left, right, or both vertebral arteries, with SCBF values below 80 ml/100g/min excluded. (n=6)

Discussion

The results overall demonstrate the development of pCASL angiography and perfusion to capture the arterial anatomy and vascular territories of the rat cervical spinal cord. We demonstrate a high degree of individual animal anatomical variability in both the segmental arteries and their perfusion territories. The distribution of the segmental arteries measured in this study was similar to those measured previously in the rat using post-mortem examinations⁹ and human cadaveric studies¹⁰ which show the predominance of arteries supplying the spinal cord originate in the mid-cervical levels. However, the curvature of the spine and the single slice acquisition in this study may have limited detection of arteries at the most caudal or rostral levels. There were no apparent differences in laterality of the segmental arteries in this study. In human studies, there is a clear left-sided vertebral artery dominance that is unrelated to cerebral dominance (i.e. handedness)²⁶ and a slight right-sided dominance in the number of segmental arteries in the cervical cord¹². All of the arterial connections between the ASA and segmental arteries were of a T shape, and we did not observe any evidence of loop structures prominently seen in human cadaver studies¹².

Consistent with prior angiography studies in the human brain²⁷, a pCASL preparation with multiple gradient echo readouts benefited from a variable flip angle readout train and sufficient labeling duration to retain signal in the distal arteries. The measured transit times in the segmental arteries and anterior spinal artery were approximately 200-400 ms using a label plane at C7 level. With a short label duration (50 ms), both the leading and trailing edge of the label bolus could be observed but reduced the available SNR in the distal arteries. However, the longer label duration (200 ms) improved contrast in the distal arteries, but only the trailing edge of the bolus is measurable although this was sufficient to reliably measure arrival times. Dynamic MRA studies in the brain using golden angle radial acquisition have the advantages of both reducing motion artifacts and utilizing undersampling temporally-binned MRA contrast^28-30. We demonstrated similar image quality between Cartesian and radial sampling with similar resolution of the vertebral artery. Since a homogeneous volume transmit/receive coil was used in this study, undersampling methods that rely on receive coil arrays were not reliable. Further developments, including three-dimensional acquisitions, undersampling, and alternative reconstruction methods instead of a simple re-gridding may be beneficial for more efficient pCASL angiography in the future or in applications to the human spinal cord.

Vessel-selective pCASL techniques have not been previously examined in the rodent. This study examined 2 variants, vessel-encoded and super-selective ASL, which both successfully labeled the left and right vertebral arteries. Since VE-pCASL relies on the distance between the two arteries along a single axis to impart differential labeling, it was well-suited to the anatomy of the vertebral arteries. SS-pCASL also distinctly labeled each vertebral artery, but VE-pCASL has the additional advantage that all collected images contribute to the overall SNR due to the encoding and decoding methods³¹. As expected, IE was reduced in the VE (0.67) and SS (0.63) methods compared to nonselective encoding (0.80). Although the laterality of the segmental arteries was readily deciphered from their anatomical locations without vessel-selective angiography, it may nonetheless have additional utility in evaluating the ASA or in vascular disorders. For example, vessel-selective angiography can visualize watershed areas around the junction between ascending and descending flow on ASA³.

The vascular territory maps provided unique insight to the perfusion of the spinal cord that were only partially evident from the angiographic or non-selective labeling images. The mean perfusion values between the non-selective and vessel-selective pCASL were similar, noting that IE was measured directly and used for quantification, which differs from the original study that estimated IE from the difference between the non-selective and vessel-selective perfusion maps¹³. The perfusion territory patterns were evident in the composite maps and showed remarkable anatomical variability across animals. Further, although the perfusion territories were largely consistent within each spinal cord level, interesting patterns were evident. In one animal, the dorsal white matter was perfused from the right side whereas the gray matter and ventral white matter at the same cord level were perfused from the left side (arrow in Fig. 7A). Across the six animals, there was a clear right-sided dominance with approximately half of the voxels having perfusion originating in the right vertebral artery compared to the 35% with left-dominance. Further, perfusion was higher in the right-sided voxels than those from the left. Only approximately 12% of voxels did not have a clear dominance. These are likely watershed regions occurring where right and left segmental arteries alternate. Values chosen for thresholding and other analytical considerations may also influence these estimated values. Although adjacent left or right perfusion territories could not be resolved in this study, combining vessel selectivity (shown here) with time-encoded ASL shown in a previous study³ may provide further insight to the anatomy and vascular dynamics within the spinal cord.

Despite continued development of spinal cord perfusion and angiography methods for the rodent, translating these techniques to humans has been limited. First, our experiments were conducted exclusively using female animals. Although in prior rodent studies there were no differences due to sex⁹, human cadaveric studies¹⁰ have reported a higher number of segmental arteries in females compared to males. The effects of sex, age and variability of neurovascular anatomy may have relationships with injury or diseases of the spine, which remains to be understood in more detail. It should also be noted that isoflurane anesthesia increases perfusion several-fold compared to its value in the normal, awake animal³². Using spinal cord probes³³ or injectable tracers and post-mortem analysis³⁴, the perfusion of the spinal cord gray matter (40-50 ml/100g/min) appears to be of a similar range as cerebral perfusion across several species^34-37. Thus, the values reported here of between 150-200 ml/100g/min substantially overestimate SCBF compared to the non-anesthetized spinal cord gray matter, while also providing greater perfusion contrast and sensitivity than would be expected in human studies. Additionally, while the velocity of blood within the major arteries of the neck appear to be similar in rodents and humans, the longer distances between the lower and upper cervical level in the human will dictate much longer transit times, likely similar to those for the brain. The decreased field strength of human systems shortens the blood T₁ time constant and imparts another challenge for human ASL, and greater spinal cord motion and CSF pulsation associated with cardiac and respiration³⁸ are additional significant barriers to spinal cord perfusion imaging in the human. Contrast-enhanced time-of-flight MRI has previously been used to visualize the ASA in the human cervical cord³⁹, although it is not routinely used diagnostically. ASL has certain advantages which have been documented in this work and elsewhere in the preclinical setting to capture robust perfusion maps that may be useful to further examine therapies for spinal cord injury or other disorders that affect perfusion. In human cord, it remains to be seen whether ASL can provide unique insights and utility beyond those of more established vascular imaging methods.

Abbreviations

ASA

Anterior Spinal Artery

ASL

Arterial Spin Labeling

ATT

Arterial Transit Time

BAT

Blood Arrival Time

CFA

Constant Flip Angle

CSF

Cerebrospinal Fluid

FSL

FMRIB Software Library

LD

Label Duration

pCASL

Pseudo-Continuous Arterial Spin Labeling

PLD

Post Label Delay

SCBF

Spinal Cord Blood Flow

SCI

Spinal Cord Injury

SNR

Signal-to-Noise Ratio

SS-pCASL

Super-selective Pseudo-Continuous Arterial Spin Labeling

VE-pCASL

Vessel-Encoded Pseudo-Continuous Arterial Spin Labeling

VFA

Variable Flip Angle

Data Availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.