Quantification of cerebrospinal fluid flow in dogs by cardiac‐gated phase‐contrast magnetic resonance imaging

Abstract Background Cerebrospinal fluid (CSF) flow in disease has been investigated with two‐dimensional (2D) phase‐contrast magnetic resonance imaging (PC‐MRI) in humans. Despite similar diseases occurring in dogs, PC‐MRI is not routinely performed and CSF flow and its association with diseases is poorly understood. Objectives To adapt 2D and four‐dimensional (4D) PC‐MRI to dogs and to apply them in a group of neurologically healthy dogs. Animals Six adult Beagle dogs of a research colony. Methods Prospective, experimental study. Sequences were first optimized on a phantom mimicking small CSF spaces and low velocity flow. Then, 4D PC‐MRI and 2D PC‐MRI at the level of the mesencephalic aqueduct, foramen magnum (FM), and cervical spine were performed. Results CSF displayed a bidirectional flow pattern on 2D PC‐MRI at each location. Mean peak velocity (and range) in cm/s was 0.92 (0.51‐2.08) within the mesencephalic aqueduct, 1.84 (0.89‐2.73) and 1.17 (0.75‐1.8) in the ventral and dorsal subarachnoid space (SAS) at the FM, and 2.03 (range 1.1‐3.0) and 1.27 (range 0.96‐1.82) within the ventral and dorsal SAS of the cervical spine. With 4D PC‐MRI, flow velocities of >3 cm/s were visualized in the phantom, but no flow data were obtained in dogs. Conclusion Peak flow velocities were measured with 2D PC‐MRI at all 3 locations and slower velocities were recorded in healthy Beagle dogs compared to humans. These values serve as baseline for future applications. The current technical settings did not allow measurement of CSF flow in Beagle dogs by 4D PC‐MRI.

the central nervous system, and transport of biologically active substances. The choroid plexus as the sole source of CSF is debated, but its constant production creates a pressure that dictates the direction of the fluid flow through the ventricular system to the subarachnoid space (SAS). 1,2 This so-called CSF bulk flow occurs over minutes. In addition, there is a pulsatile flow modulated by the expansion of intracranial vessels during every systole because of a mild increase in intracranial volume. [3][4][5] As compensation, CSF is displaced from intracranial to spinal CSF spaces in systole and back in diastole, leading to a bidirectional pulsatile flow. [5][6][7] Likewise, the respiratory cycle contributes to CSF flow-variations of intrathoracic pressure evoke changes in the epidural venous plexus. 8 Abnormal CSF flow patterns and alteration of CSF flow velocity occur in a variety of diseases, mainly due to stenosis or obstruction of CSF spaces, which might result in secondary disorders of the ventricular system like hydrocephalus or cyst-like lesions within the forebrain, spinal cord, or both. [9][10][11][12][13][14][15][16][17][18] Phase-contrast magnetic resonance imaging (PC-MRI) is a non-  21,22 In humans, PC-MRI can determine the degree and location of CSF stenosis within the ventricles and the SAS, provide information for surgical planning, and assess postsurgical success. 7,9,10,14,16,17,23 Toy breed dogs and brachycephalic dogs in particular show abnormalities of CSF spaces in MRI similar to humans. [24][25][26][27][28] However, only limited velocity data are available in dogs and the only existing study in dogs demonstrated a lower peak velocity in Cavalier King Charles Spaniels (CKCS) compared to control dogs at the foramen magnum (FM). Furthermore, this investigation revealed that the combination of slower CSF peak velocity in the dorsal SAS at the level C2-C3 intervertebral disc and higher peak velocity at the level of the FM in CKCS was associated with syringomyelia in CKCS. 29 Despite these promising results, PC-MRI is not implemented as a diagnostic tool to detect CSF flow abnormalities in dogs with hydrocephalus, syringomyelia, or other conditions indicative of stenosis of CSF flow.
Challenges for the acquisition of PC-MRI in dogs with sufficient signal to measure flow velocities arise from the smaller CSF spaces, higher heart rate in dogs, the slower flow velocity compared to humans, [29][30][31] and the lack of normative data to compare with.
The aim of this prospective study is to optimize 2D and 4D PC-MRI for the smaller CSF spaces and slower flow velocity in dogs, to provide normative values from neurologically healthy dogs with normal appearance of CSF spaces on MRI.

| Phantom study
The 2D and 4D PC-MRI sequences were adapted for the anatomical differences in dogs compared to humans on a phantom mimicking the smaller CSF spaces and slower CSF flow velocity.
A 20 × 15 × 15 cm plastic box was filled with gelatin, embedding 3 infusion lines with a diameter of 2.5 mm, 1.5 mm, and 1.0 mm as straight segments with a U-turn. The infusion lines were connected to a medical perfusion pump outside the MRI room (Syramed μSP6000; arcomed, Regensdorf, Switzerland). The perfusion syringes were filled with water with a similar viscosity to CSF. The liquid flowed from the small to the large diameter infusion lines to reduce the effect of higher resistance in the small lines. Flow velocity was calculated with a linear stream-profile conversion table (volume to flow velocity). Measurements were repeated when the velocity was gradually reduced.
The phantom was placed in a 3T magnetic resonance system (Philips Ingenia scanner, Philips AG, the Netherlands) with a SENSE-head and neck coil. A cardiac frequency with a heartbeat of 100 bpm was simulated by a physiological simulation setting on the Philips software. 3D T1-and 3D T2-weighted images were performed as planning sequences. The 2D and 4D PC-MRI sequences were aligned perpendicular to the flow in the perfusion lines. First to be tested were sequence parameters for better signal-to-noise ratio (SNR) with a voxel size of 1 × 1 × 5 mm with different numbers of acquisitions (NSA 1 and 2), field of view (FOV 140/140, 150/150), and flip angle (15 and 20 ). The measurements were performed once for each parameter and subjectively evaluated for good SNR. Finally, to find the closest velocity encoding the speed in the infusion lines, without getting aliasing but still having good SNR, different velocities encoding from 10 cm/s to 1 cm/s were tested. Additionally, for the 4D PC-MRI sequence, both available accelerating techniques, kt-BLAST and SENSE, were tested for their effect on scan time and signal. 19,20,32 The image acquisition and assessment of image quality was performed by the first author, a PhD student with a Master's degree in veterinary medicine under supervision and advice of a senior radiologist (ECVDI Diplomate). Technical support concerning the 4D sequence was provided by a senior scientist in biomedical engineering with expertise in motion encoding.
Concerning the 2D sequences, further support was provided by a Professor of Radiology and Biomedical Engineering with a research focus on blood and CSF flow dynamics using flow-sensitive MRI techniques.
A decision of sufficient image quality was made based on the velocity-to-noise ratio, that is, the ability to detect a change of gray scale value in time of a pixel undergoing a phase shift. The final decision on acquisition parameters was made in consensus between first and last authors.

| Animals
The study was conducted with 6, adult, purpose-bred, research Beagle dogs. Sex was equally distributed. The dogs were also used in another parallel research study, which did not interfere with our research aims and allowed the reduction of the number of experimental animals used (3R reduce). The study population had a mean age of 7.42 years (median = 7.65; range 5-10) and a mean bodyweight of 13.6 kg (median 13.4; range 11.0-18). All dogs were declared healthy on clinical neurological examination and showed no structural brain abnormalities on MRI (3D T2-weighted and T1-weighted images; Figure 1 Fresenius Kabi AG, Switzerland) 5 mL/kg/h) was administered during anesthesia. Body temperature was maintained between 37.5 C and 39 C and heart rate was maintained between 70 and 110 beats per minute. If cardiac frequency increased to over 110 bpm, a fentanyl (Fentanyl Sintetica; Sintetica, Switzerland) bolus of 2 μg/kg IV was administered and repeated, if necessary. Anesthesia monitoring included cardiovascular and respiratory variables, which were measured continuously and recorded by a multiparameter monitor (Datex S5, Anandic Medical Systems AG, Switzerland) that included pulsoxymetry, noninvasive blood pressure measurement, capnography, inhalant gas analysis, spirometry, and vectorcardiography.
Mean blood pressure was kept above 65 mm Hg while heart rate was kept below 110 beats per minute. Anesthesia depth was adapted as necessary.

| Magnetic resonance imaging
All dogs were scanned in sternal recumbency with the head slightly extended (<180 to the spinal cord) and elevated so that no pressure developed on the jugular veins. 3,4 The cardiac frequency was moni-

| Postprocessing and image analysis
The postprocessing of the 2D data set was performed on an external workstation (MR WorkSpace 2.6.3.5, Philips Medical System, the F I G U R E 1 T2-weighted mid-sagittal magnetic resonance image of the brain and upper cervical spine of a Beagle. The yellow lines illustrated the location of the phase-contrast sequence at the mesencephalic aqueduct, the most caudal aspect of the foramen magnum and at the atlantoaxial junction. Measurements were aligned perpendicular to the flow direction Netherlands). The 4D PC data sets were processed by the GT-flow software (Version 1.3.11, Gyrotools Ltd., Zurich, Switzerland). On the acquired data from the 6 dogs, all ROIs were placed 3 times at least 3 days apart and the mean of the peak velocity is reported.
Postprocessing of the 4D data set could not be performed due to high velocity noise and low SNR on the phase images, which led to undetectable velocity in CSF spaces.

| Statistical analysis
Descriptive analysis was performed with all values obtained. The data were expressed as mean ± SD or median with range.

| DISCUSSION
The present study provides normative data of CSF flow in a group of middle-aged neurological healthy dogs as baseline for comparison in future studies. Ventricular and subarachnoid CSF space abnormalities are common findings in dogs; however, their relationship to alterations of CSF flow is not well understood and poorly investigated.
PC-MRI is routinely used as a method to quantify CSF flow in humans but needs adaptation for use in veterinary patients, because of the small CSF spaces, 30,31 slow CSF flow velocities, and higher heart rate compared to humans. 29 The small CSF spaces require smaller voxel sizes that lead to a reduced SNR. Compensation can be achieved by an increased FOV or increased NSA, which produces prolongation of scan time. 19,20,34 The peak velocities in dogs were approximately 20% lower compared to humans. 9,15,16,21,29,35 Lower flow velocities lead to higher velocity noise and require lowering of the velocity encoding. Reducing the velocity encoding results in an increased TR, which is accompanied by increased scan time. 19,20,22,34 Finally, PC-MRI is cardiac gated, meaning that a cardiac gradient is applied in synchronization with the R-R interval, the time elapsed between 2 successive R waves of the QRS signal on the electrocardiogram. A higher heart rate leads consequently to a smaller R-R interval and therefore a shorter time to collect flow data and might result in longer acquisition time or even abortion of the sequence. 19,20 The phantom was built to mimic the conditions in veterinary patients, and sequence parameters were adapted until a satisfactory SNR within a clinically reasonable acquisition time of 3 to 4 minutes was achieved.
The sequence parameters developed on the phantom were then applied in dogs, demonstrating that the phantom allowed us to optimize the 2D PC-MRI sequence without having a dog in the scanner under general anesthesia. This is in line with 3R requirements and demonstrates that phantom studies in MRI could help to refine protocols before use on living animals and consequently reduce anesthesia time and risk. Furthermore, the use of the phantom allowed us to demonstrate that 2D PC-MRI provides data corresponding to the cal- Therefore, the results presented here are limited to 2D PC-MRI with CSF flow velocities measured at 3 different locations. These locations play an important role in the pathomechanism of common disease processes such as stenosis, hydrocephalus and syringomyelia. [9][10][11]13,[15][16][17][18] In addition, measurements at different locations within 1 individual are important, as differences in peak velocity indicate stenosis. 29 The measurement at the cranial cervical spine might provide additional information about intracranial compliance. 15 At each of the 3 locations, the peak velocities measured in the study population of middle aged healthy Beagle dogs were lower than those measured in humans. This is especially true for the level of the mesencephalic aqueduct, where values ranging from 3 to 7 cm/s were obtained in healthy human participants. 9,14,37,38 The reason for the slower flow might be species-related differences such as skull shape, the orientation of the brain to the spinal cord, or might be due to the smaller size of dogs compared to humans. Age of the individual could also play a role since a study comparing CSF flow velocity in human infants found a higher CSF flow velocity than that in adults. 33 To reduce individual differences in CSF flow velocities, we only included dogs of the same breed and similar age and therefore similar conformation (size, bodyweight, and skull shape). These inclusion criteria resulted in a small sample size of 6 individual dogs, and the peak velocities measured at each of the 3 locations showed only small differences between dogs resulting in a small standard variation. Our results obtained from this small but consistent study sample will enable future comparison with measurements from larger samples with individuals of different ages, body weight, or skull shape. This will allow investigation of factors associated with CSF flow velocities in dogs. Apart from differences between individual animals, other factors influence the PC-MRI measurements of CSF flow. First, variation might arise from the site of measurements, particularly at the level of the mesencephalic aqueduct, where normal velocities in humans encompass a broad range from 3 to 7 cm/s. This is probably caused by setting the phase encoding gradient at either the rostral part, the ampulla, or the caudal aspect. 5,10,17,[38][39][40] The obtained range of 0.51 to 2.08 cm/s in the present study group for the peak velocity at the level of the aqueduct therefore seems acceptable and was obtained by defining landmarks during acquisition. Second, variation might arise from manually placing the ROI in the postprocessing step. Therefore, the use of semiautomatic software for ROI placement has been suggested. 41 The software was not available for us, and ROI placement was performed 3 times with low intraobserver variability of 0.68 for all SAS spaces and below 0.13 within the mesencephalic aqueduct. In the present study, manual ROI placement was a low source of error, but variability between different observers might be higher. Well-defined landmarks for both, measurements and ROI placement, seem to be of high importance to be able to compare velocities. In a previous study, CSF peak velocities of 59 CKCS dogs were compared to 5 controls, 4 Beagle dogs, and 1 mixed-breed dog. The flow velocities measured at the FM and mean peak velocity in the ventral and dorsal SAS of the FM in the control group was 0.75 ± 0.24 cm/s and 0.59 ± 0.13 cm/s, respectively. 29 The obtained peak velocities of the control group were lower compared to our study group and velocities obtained in CKCS were even lower. 29 Reasons for the higher velocities found in the present study might be due to patientrelated factors such as the individual size of CSF space, jugular venous flow, and compliance of the intracranial space [3][4][5]15 as well as the heart rate and respiration rate of the patient, 8,33,39,40 all factors known to influence the CSF flow velocity. A difference in the examination protocol might have influenced the measurements, all dogs in the previous study were positioned with the head in a flexed position. In our study, to facilitate comparison with future measurements, the dogs were examined with the head in an extended position, according to our institutional protocol for brain MRI. However, the extended head position in sternal recumbency resulted in a narrowing of the dorsal SAS, so that in 2 dogs no CSF flow data from the phase images could be extracted. It seemed that the dorsal SAS at the level of the atlantoaxial junction is especially narrow due to kinking and the dens axis, which displaced the spinal cord slightly dorsally. 30,31 For this reason, we performed an additional measurement at the level close to the intervertebral disc space of C2/C3 in 1 of the 6 dogs. This measurement resulted in a good signal of CSF flow in phase images. Therefore, cervical CSF flow measurements close to the intervertebral junction C2/C3 with well visible SAS spaces might be more reliable and clinically useful rather than the atlantoaxial junction. Furthermore, examinations in T A B L E 1 Mean peak velocity in cm/s and SD of all 6 dogs for the different locations  if the systemic blood pressure is below 60 mm Hg. 42 To avoid differences related to physiological parameters, a standardized anesthesia protocol, monitoring and consistent conditions are important. Measurements performed under constant heart rate seem important to be able to detect differences in flow velocities at different locations in 1 animal. The obtained flow velocities in our dogs suggest that if at 1 location the flow velocity is lower (see Table 1, dog 5), the velocities at the other locations are lower too, and vice versa (see Table 1