Vessel length on SNAP MRA and TOF MRA is a potential imaging biomarker for brain blood flow

Abstract

Purpose:

To explore feasibility of using the vessel length on TOF or SNAP MRA as an imaging biomarker for brain blood flow, by using arterial spin labeling (ASL) perfusion imaging and 3D phase contrast (PC) quantitative flow image as reference.

Methods:

In a population of thirty subjects with carotid atherosclerotic disease, the visible intracranial arteries on TOF or SNAP were semi-automatically traced and the total length of the distal segments was calculated with a dedicated software named iCafe. ASL blood flow was calculated automatically using the recommended hemodynamic model. PC blood flow was extracted by generating cross-sectional arterial images and semi-automatically drawing the lumen contours. Pearson correlation coefficients were used to assess the associations between the different whole-brain or hemispheric blood flow measurements.

Results:

Under the imaging protocol used in this study, TOF vessel length was larger than SNAP vessel length (P < 0.001). Both whole-brain TOF and SNAP vessel length showed a correlation with whole brain ASL and 3D Phase Contrast (PC) blood flow measurements, and the correlation coefficients were higher for SNAP vessel length (TOF vs ASL: R = 0.554, P = 0.002; SNAP vs ASL: R = 0.711, P < 0.001; TOF vs 3D PC: R = 0.358, P = 0.052; SNAP vs 3D PC: R = 0.425, P = 0.019). Similar correlation results were observed for the hemispheric measurements. Hemispheric asymmetry index of SNAP vessel length also showed a significant correlation with hemispheric asymmetry index of ASL cerebral blood flow (R = 0.770, P < 0.001).

Conclusion:

The results suggest that length of the visible intracranial arteries on TOF and SNAP MRA images can serve as a potential imaging marker for brain blood flow.

1. Introduction

Brain blood flow is an important indicator of tissue vitality and its measurement has wide application in clinical practice and research. To date, several MR imaging techniques are available for measuring blood flow in the brain, including dynamic susceptibility contrast, arterial spin labelling (ASL) and phase contrast imaging (PC) [1]. ASL allows non-invasive quantification of the absolute cerebral blood flow (CBF) in parenchymal tissue [2]. Studies have demonstrated the feasibility of ASL in acute and chronic cerebrovascular disease settings [3]. PC MRI is a technique that utilizes bipolar gradients to measure the velocity and subsequently the blood flow rate in large arterial and/or venous vessels. It can also be used to visualize vessel structure [4].

Non-contrast MRA technique, such as time-of-flight (TOF), is an imaging modality that has been widely used for assessing the morphology of head and neck arteries [5]. The imaging mechanisms of most non-contrast MRA techniques are to some extent dependent on blood flow [5]. For example, the blood signal intensity, thus visibility of the vessels, on TOF is strongly affected by the number of excitation RF pulses experienced by the blood spin, which is determined by the traveling length and velocity [6]. From this perspective, vasculature features, such as vessel length and number of branches, on the non-contrast MRA images may serve as surrogates of blood flow. In fact, associations between such vasculature features on TOF images and aging have been previously reported [7,8], and may be attributed to the well-known age-related blood flow reduction [9,10].

Simultaneous non-contrast angiography and intraplaque hemorrhage (SNAP) is an MR imaging technique that allows for MRA imaging and detection of intraplaque hemorrhage in a single scan [11,12]. SNAP has been shown to have comparable capabilities to TOF for detecting luminal stenosis in the intracranial arteries [13]. In terms of the imaging mechanism, SNAP is also sensitive to flow velocity [14,15]. In particular, SNAP is based on the phase-sensitive inversion recovery (PSIR) technique [16], and the signal intensity of distal arteries on SNAP is influenced by the blood spin’s transit time within the inversion slab and the time interval between two consecutive inversion preparation RF pulses [14,15].

Because of the difference in imaging mechanisms and blood signal measuring locations (i.e. parenchyma, large proximal vessels, or distal intermediate-sized vessels), the above-mentioned intracranial vascular features derived from TOF or SNAP MRA have the potential to reflect unique brain blood flow information or possess a different measurement performance than traditional brain blood flow measuring techniques, such as ASL and PC. Therefore, it is of interest to establish and evaluate such vascular feature-based imaging biomarker of brain blood flow. Simulation studies have demonstrated the velocity sensitivity for both TOF and SNAP [14,15]. However, these vascular features have not been validated in vivo by comparing to existing blood flow measuring techniques.

This study aims to explore the feasibility of using the vessel length from clinical routine TOF and SNAP as surrogate imaging markers of brain blood flow by examining their correlations with blood flow measurements from ASL and 3D PC MRI in patients with carotid artery atherosclerotic disease. The strength of correlation of vessel length from either TOF or SNAP with traditional blood flow measurements such as ASL or 3D PC could provide insight into the degree to which the vessel length depends on cerebral blood flow in the general population.

2. Materials and Methods

2.1. Study population

This study was based on a sub-cohort of asymptomatic subjects enrolled in an ongoing study of carotid atherosclerosis. The specific inclusion criteria for this study were: 1) at least one carotid artery had stenosis >15% and neither carotid artery had >79% stenosis, according to clinical ultrasound, CT or MRI; 2) asymptomatic with regard to the carotid disease; 3) age ≥ 18. The exclusion criteria were: 1) subjects with contraindication to MRI; 2) limited life expectancy, severe chronic illness or chronic disability that will preclude or pose a significant hardship to complete study procedures; 3) history of neck radiation therapy, or with prior bilateral carotid endarterectomy or stenting; 4) pregnant subjects; 5) subjects with systemic inflammatory disease, or vasculitis, or atrial fibrillation. The study protocol was approved by the local Institutional Review Board and informed consent was obtained from all participating patients.

2.2. MR imaging

All MR imaging scans were performed on a 3T Philips Ingenia CX scanner (Philips, Best, The Netherlands), using a 32-channel head coil. Three-dimensional pseudocontinuous ASL (pCASL) was performed to obtain parenchymal blood flow, 3D PC was performed to obtain blood flow in the large- or intermediate-size arteries, while 3D TOF and SNAP MRA imaging were performed to obtain intracranial vessel length. For the labeling module of pCASL sequence, the imaging parameters were: balanced pCASL, RF duration 0.48 ms, RF interval 1.21 ms, RF flip angle 27.81°, mean gradient strength 0.36 mT/m, maximum gradient strength 5 mT/m. The other imaging parameters of pCASL and the other sequences are summarized in Table 1. Since the distal intracranial arteries were the major focus of the vascular imaging sequences in this study, the field-of-views of the SNAP and 3D PC scans were positioned to cover more distal arteries and only a small segment of the internal carotid artery (ICA) and basilar artery, as illustrated in Fig. 1.

Table 1.

MR imaging parameters.

	3D TOF	SNAP	pCASL	3D PC
Readout	SPGR	SPGR	GRASE	SPGR
FOV, mm3	190×180×105	180×180×70	240×240×125	180×180×70
Voxel size, mm3	0.5×0.5×1	0.8×0.8×0.8	3×3×5	0.5×0.5×1
TR/TE, ms	20/3.5	10/5.6	4220/12	13/6
FA, degree	18	11/5	90	12
SENSE factor	2.2	2	2.3	5a
Receiver bandwidth, Hz/pixel	288	319	119	96
Special parameters	Number of slabs: 6	IRTR: 1814 ms; TI: 500 ms; Global inversion	Number of repetitions: 8; Labelling duration:1800 ms; PLD: 2000 ms	VENC: 100 cm/s in all three directions
Acquisition time	6 min 37 s	3 min 45 s	4 min 55 s	5 min 4 s

^aCompressed sensing acceleration factor.

SPGR: spoiled gradient echo; GRASE: gradient echo spin echo; FOV: field-of-view; TR: repetition time; TE: echo time; FA: flip angle; IRTR: time interval between consecutive inversion RF pulse; TI: inversion time; PLD: post labeling delay; VENC: encoding velocity.

Fig.1.

Typical positioning of the field-of-view in the feet-head direction for the 3D TOF sequence (the red frame), SNAP sequence (the blue frame) and 3D PC sequence (the blue frame), in this study.

In addition, whole-brain 3D T1W brain anatomical images were obtained from a recently developed sequence named iSNAP [17] with 0.8 mm isotropic resolution. An example of the T1W images is shown in Supplementary Figure S1.

2.3. Image analysis

The degree of carotid stenosis, based on North American Symptomatic Carotid Endarterectomy Trial criteria (NASCET), was recorded by a radiologist using 3D black blood carotid MR vessel wall images. The 3D black blood carotid MR imaging technique we used and details of the stenosis measurement were the same as the previous study by Zhao et al [18].

Parenchymal CBF map was calculated from the ASL images using the recommended hemodynamics model [2]. Since CBF quantification for white matter with ASL is less reliable [2], we only focused on gray matter (GM) CBF in this study. The ASL images were co-registered with the T1W images from which the GM was segmented (see example segmentation results in Supplementary Figure S1). Both the image registration and brain segmentation were performed on SPM12 (https://www.fil.ion.ucl.ac.uk/spm/). Then mean ASL GM CBF of whole brain, right- and left-hemisphere were calculated. The total brain volume was also obtained from the brain segmentation results by adding up the volume of GM and white matter.

All intracranial vascular images, including 3D TOF, SNAP and 3D PC, were co-registered using SPM12. Centerlines of the visible intracranial arteries on 3D TOF images were semi-automatically traced by a trained reviewer using a custom-developed tool, iCafe, for intracranial vascular feature extraction [19]. Once the centerlines were traced, landmarks were manually labeled to identify the branches of the intracranial arteries. Then the centerlines were mapped to the co-registered SNAP and 3D PC images. On SNAP, the mapped centerlines were further manually corrected such that the resulting centerlines matched with visible arteries on the raw image. Then the length of the visible distal arteries (i.e. the arteries distal to the A1-A2 connection point, M1-M2 connection point or P1-P2 connection point) on 3D TOF or SNAP was summed for the whole brain, right- and left-hemisphere, respectively. For a fair comparison between 3D TOF and SNAP, the spatial coverage was cropped to be the same before calculating the vessel length. An example of the centerline tracing results on TOF and SNAP is shown in Fig. 2. Measurements of intracranial vascular features using iCafe have been shown to have excellent inter-scan and intra-operator reproducibility as well as good to excellent inter-operator reproducibility [20].

Fig. 2.

Representative axial maximum/minimum intensity projection image (top) of TOF (left) and SNAP (right) and the corresponding vessel tracing results (bottom). The different intracranial arteries are indicated with different colors in the vessel tracing results.

In terms of 3D PC, the mapped centerlines were used to generate a cross-sectional image at near the middle of each of the following artery segments: ICA, basilar artery, A1 and first A2 (i.e. the one connected to anterior communicating artery) of anterior cerebral artery, M1 of middle cerebral artery, P1 and first P2 (i.e. the one connected to posterior communicating artery) of posterior cerebral artery. Then, the arteries on the cross-sectional PC complex difference images were segmented using an active contour algorithm [21], followed by manual correction of the contours by a reviewer experienced in vascular imaging. Subsequently, PC volumetric flow rate (VFR) was calculated for each arterial segment by multiplying the mean velocity on the PC phase images and cross-sectional area. Whole brain 3D PC blood flow was obtained by adding up the VFR of RICA, LICA and basilar artery, while the hemispheric 3D PC blood flow was obtained by adding up the VFR of A2, M1 and P2.

Asymmetry index (AI) of the blood flow measurements was defined as the left side flow measurement subtracted from the right side flow measurement divided by the average of the two [22].

2.4. Statistics

Lilliefors test was used to test the normality of the continuous variables. The vessel length was compared between TOF and SNAP using paired t-test or Wilcoxon signed-rank test, depending on the normality of the variables. We then explored the correlation between the different brain blood flow measurements, including the vessel length on TOF and SNAP. To mitigate the influence of brain size on the flow measurement, the VFR obtained from 3D PC and the total vessel length obtained from 3D TOF or SNAP were normalized by the total brain volume and cube root of total brain volume, respectively. Pearson’s correlation analyses or Spearman’s correlation analyses, depending on the normality of the variables, were performed for the following pairs: (1) whole brain (or right- or left-hemisphere) ASL GM CBF and 3D PC blood flow; (2) whole brain (or right- or left-hemisphere) ASL GM CBF and 3D TOF (or SNAP) vessel length; (3) whole brain (or right- or left-hemisphere) 3D PC blood flow and 3D TOF (or SNAP) vessel length. Then correlation coefficients of the AI between the different blood flow measurements were calculated.

All above statistics were performed in MATLAB 2019b (MathWorks, Natick, MA). P values less than 0.05 were considered statistically significant. For simplicity, no adjustment of P value for multiplicity was performed.

3. Results

Thirty-one patients were enrolled and all successfully completed the MR imaging. Among these subjects, one was retrospectively excluded from the analyses due to the presence of arteriovenous malformation, which may have greatly altered the brain blood flow.

The demographics and clinical characteristics of the remaining 30 patients are shown in Table 2. The degree of stenosis was 29.85% ± 17.53% and 33.01% ± 20.09% for the right and left carotid arteries, respectively, and only 1 subject showed a stenosis >50% on both sides. The subjects had an average total brain volume of 1,073.2± 101.8 cm³. The mean values of the whole brain and hemispheric ASL CBF, 3D PC VFR, TOF vessel length and SNAP vessel length are reported in Table 3. The P values for the tests of normality of the ASL CBF, the 3D PC VFR after normalization, the TOF/SNAP vessel length before normalization, the TOF/SNAP vessel length after normalization, and the AIs were all > 0.05, except for the normalized 3D PC VFR in left hemisphere (P = 0.017). Therefore, for consistency, the Pearson’s correlation coefficient was used for all of the correlation analyses. The TOF vessel length in this study was found to be significantly larger than SNAP vessel length (P < 0.001).

Table 2.

Demographics and clinical characteristics.

	Mean ± SD or N (%)
Age (yr)	71.7 ± 9.8
Male	19 (63.3)
BMI (kg/m2)	27.3 ± 4.0
History of stroke or TIA	15 (50.0)
Antihypertensive drug use	21 (70.0)
Systolic blood pressure (mm Hg)	144.1 ± 13.1
Diastolic blood pressure (mm Hg)	77.9 ± 9.8
Diabetes mellitus	3 (10.0)
Smoking	11 (36.7)

SD: standard deviation; BMI: body mass index;

TIA: transient ischemic attach.

Table 3.

Mean blood flow measurement values (N = 30).

	Before normalization	After normalizationa
ASL GM CBF (ml/100g/min)
Whole brain	30.8 ± 5.1
Right hemisphere	30.8 ± 4.9
Left hemisphere	30.8 ± 5.6
3D PC VFR
Whole brain	7597.3 ± 1343.4 mm3/s	0.0071 ± 0.0010 s−1
Right hemisphere	3142.1 ± 556.3 mm3/s	0.0029 ± 0.0004 s−1
Left hemisphere	3146.3 ± 616.2 mm3/s	0.0029 ± 0.0005 s−1
TOF vessel length
Whole brain	3482.5 ± 749.1 mm	34.0 ± 7.0
Right hemisphere	1748.2 ± 378.8 mm	17.1 ± 3.6
Left hemisphere	1734.3 ± 396.4 mm	16.9 ± 3.7
SNAP vessel length
Whole brain	2792.6 ± 979.1 mm	27.2 ± 9.2
Right hemisphere	1403.7 ± 481.1 mm	13.7 ± 4.5
Left hemisphere	1388.9 ± 536.1 mm	13.5 ± 5.0

^aFor each subject, the 3D PC VFR was normalized with the total brain volume, while the TOFor SNAP vessel length was normalized with the cube root of the total brain volume. ASL: arterial spin labeling; GM: gray matter; CBF: cerebral blood flow; PC: phase contrast; VFR: volume flow rate; TOF: time of flight; SNAP: simultaneous non-contrast angiography and intraplaque hemorrhage

3.1. Correlations between different brain blood flow measurements

The correlation between the different brain blood flow measurements are shown in Table 4. The scatter plots and correlation results for the whole-brain measurements are shown in Fig. 3. As expected, the ASL GM CBF has statistically significant correlation with the PC blood flow measurement, although only moderate correlation was observed. Both TOF vessel length and SNAP vessel length had moderate to strong correlation with either ASL GM CBF or PC blood flow, except that only weak correlations between TOF vessel length and 3D PC blood flow were observed. Overall, in this study, the correlation coefficients were all higher for SNAP than TOF. Interestingly, the correlation between ASL GM CBF and SNAP vessel length was even higher than that between ASL GM CBF and PC blood flow. Fig. 4 shows the vessel tracings and ASL CBF maps of two representative subjects that had different blood flow.

Table 4.

Correlation between different brain blood flow measurements (N = 30)

	R	P
ASL GM CBF versus PC blood flow
Whole brain	0.636	<0.001
Right hemisphere	0.623	<0.001
Left hemisphere	0.629	<0.001
ASL GM CBF versus TOF vessel length
Whole brain	0.554	0.002
Right hemisphere	0.493	0.006
Left hemisphere	0.575	<0.001
ASL GM CBF versus SNAP vessel length
Whole brain	0.711	<0.001
Right hemisphere	0.672	<0.001
Left hemisphere	0.743	<0.001
PC blood flow versus TOF vessel length
Whole brain	0.358	0.052
Right hemisphere	0.383	0.037
Left hemisphere	0.483	0.007
PC blood flow versus SNAP vessel length
Whole brain	0.425	0.019
Right hemisphere	0.496	0.005
Left hemisphere	0.554	0.001

ASL: arterial spin labeling; GM: gray matter; CBF: cerebral blood flow; PC: phase contrast; TOF: time of flight; SNAP: simultaneous non-contrast angiography and intraplaque hemorrhage

Fig. 3.

Scatter plots and Pearson’s correlation of the different whole-brain blood flow measurements in this study. GM: gray matter; VFR: volume flow rate.

Fig. 4.

Comparison of vessel tracings for TOF (left) and SNAP (middle) with a CBF map from pCASL (right). The top row is an 82-year-old female, the bottom row is a 76-year-old female. It should be noted that the SNAP vessel length is greatly reduced in the patient with lower CBF.

3.2. Correlations between the AI of different brain blood flow measurements

The mean ± SD of the AI of ASL GM CBF, PC blood flow, TOF vessel length, and SNAP vessel length were 0.30% ± 8.03%, 0.04% ± 15.84%, 0.85% ± 10.95%, and 2.45% ± 21.85%, respectively. When using the absolute values, the mean ± SD of the AI of ASL GM CBF, PC blood flow, TOF vessel length, and SNAP vessel length were 5.6% ± 5.7%, 12.0% ± 10.0%, 8.1% ± 7.3%, and 15.7% ± 15.1%, respectively. The correlation between the AI of different brain blood flow measurements are shown in Fig. 5. The ASL GM CBF and SNAP vessel length showed a strong significant correlation, while the correlation between ASL GM CBF and TOF vessel length only had a P value of 0.09. Fig. 6 demonstrates the TOF and SNAP vessel tracings of a subject with overt hemispheric asymmetry on an ASL CBF map.

Fig. 5.

Correlation between the asymmetry index (AI) of different brain blood flow measurements (N = 30). The Pearson’s correlation coefficient R and P values are shown within the plots.

Fig. 6.

Demonstration of the blood flow asymmetry across the vessel tracings for TOF (left), SNAP (middle), and pCASL (right) in a 73-year-old male subject.

4. Discussion

This study analyzed the correlation between the vessel length on TOF and SNAP MRA images and conventional brain blood flow measurements in patients with carotid artery atherosclerotic disease. We found that both TOF and SNAP vessel lengths had moderate to strong correlation with ASL CBF, and weak to moderate correlation with 3D PC blood flow measurement. The SNAP vessel length was observed to have a stronger correlation with ASL CBF than 3D PC flow and vessel length on clinical routine TOF. The SNAP vessel length hemispheric asymmetry was also significantly correlated with the ASL CBF asymmetry. These results suggest that vessel length on TOF and SNAP can serve as a surrogate marker for brain blood flow. On the other hand, unlike ASL CBF which is a perfusion measurement on the parenchymal level, vessel length on TOF and SNAP images is mainly affected by the blood flow in the large to intermediate-sized arteries. Therefore, the vessel length on TOF and SNAP images may provide distinctive information regarding blood flow in arteries. It would be of further interest to explore the utility of such vessel length measurements in clinical practice and research.

The main contribution of this study is to use 3D PC and ASL CBF to validate the vessel length on TOF or SNAP MRA images as a surrogate marker of blood flow. Previous simulation studies [14,15] have suggested associations between the intracranial vasculature features on TOF or SNAP MRA images and brain blood flow. However, clinical studies that address the degree these vasculature features correlate with traditional brain blood flow measurements have been lacking. In clinical practice, it is well-accepted that the complete disappearance or significant signal drop of a large intracranial artery and the distal branches indicates a severe stenosis or occlusion, thus dramatically decreased blood flow. However, whether vessel visibility can serve as surrogate of blood flow in patients without occlusion of the major large intracranial arteries remains an unanswered question. This study addresses these issues by performing correlation analyses between the MRA vessel length and traditional blood flow measurements in patients without occlusion of the large intracranial arteries on the TOF-MRA images.

Overall, the vessel length on TOF or SNAP MRA was found to be correlated with ASL CBF and PC flow measurement (Table 4). This confirms the blood flow nature of vessel length on TOF and SNAP MRA images and suggests the feasibility for using it as surrogate of brain blood flow. Besides, vessel length on SNAP has stronger correlation with ASL CBF and PC flow than the vessel length on TOF. This suggests that the SNAP sequence used in this study might have a higher sensitivity to blood flow than clinical routine TOF, although this comparison may not be completely fair due to difference in the imaging parameters (e.g. spatial resolution). The difference in imaging mechanism may be the main factor contributing to this observation. In TOF, the blood signal is mainly affected by the magnetization saturation effect (as induced by repeating excitation), which in turn is dependent on the strength of the inflow effect (or the flow velocity). However, for routine TOF imaging, the Multiple Overlapping Thin Slab Acquisition (MOTSA [6]) and Tilt-Optimized Nonsaturated Excitation (TONE [23]) techniques are usually employed in a clinical setting, in order to reduce the saturation effect and maintain high vessel visibility even when the flow velocity is low. Therefore, the sensitivity to blood flow in a sense is deliberately decreased on routine TOF. When MOTSA is used, the vessel length or visibility of distal arteries on TOF is mainly affected by local flow velocity and less by velocity in the proximal segments that are in another slab. For SNAP, the blood signal is also affected by the magnetization saturation effect, but the effect may be greater because there is a much thicker slab (i.e. around half of the scanner’s maximum excitation range in head-feet direction, due to the use of global inversion) for the spins to travel through and not only the repeating excitation RF but also the repeating inversion RF will cause the saturation effect [14,15]. In this study, the subjects had carotid atherosclerosis and stenosis. Based on the above discussion, the potential flow alteration induced by the stenosis [24,25], is likely to have larger influence on visibility of distal arteries on SNAP than on TOF. It is worth mentioning that for both TOF and SNAP sequence, the sensitivity to blood flow may be further increased by adjusting some of the imaging parameters, such as slab thickness and flip angle.

Another finding from this study is that the vessel length from TOF and SNAP was more correlated to ASL CBF than 3D PC flow, and the correlation between SNAP vessel length and ASL CBF was even stronger than the correlation between ASL CBF and 3D PC flow (Table 4). This probably can be partially explained by the larger variability in PC-based flow measurement due to the partial volume effect and difficulty in accurate segmentation of the arteries. Dolui et al [26] showed that the PC blood flow measurement was more variable than pCASL CBF in a large sample of healthy middle-aged subjects, and the correlation between PC blood flow and pCASL CBF was 0.59, which is similar to our result (See Table 4). Peng et al showed that the accuracy of PC flow measurement is greatly affected by the spatial resolution and a 0.5 mm resolution is desirable for carotid arteries [27]. Intracranial arteries may need a higher resolution. The 3D PC resolution of 0.5×0.5×1 mm³ used in this study may not be optimal for intracranial arteries. Besides, the magnitude of the physiological variation of the blood flow in the different measurement locations may also contribute to this finding. The PC flow measurement in this study was made on proximal large arteries, while the vessel length measurement on TOF or SNAP was mainly made on the distal intermediate-size arteries and the ASL CBF measurement was made on the brain parenchyma. The latter two measurement locations may show smaller physiological blood flow variation, as compared to the large arteries. Nevertheless, it would be interesting to further explore whether the vessel length of TOF or SNAP can serve as an interrogating technique uniquely for the blood flow in the distal intermediate-sized arteries, given that the PC technique usually suffers insufficient spatial resolution and SNR for these arteries.

Hemispheric blood flow asymmetry is an important and straightforward image indicator in disease diagnosis [28–30], given that a normal brain usually shows no or very small hemispheric difference [22,31–33]. Therefore, it is important to examine the capacity of the TOF and SNAP vessel length in detecting the hemisphere asymmetry. In this study, SNAP vessel length AI was observed to be significantly strongly correlated with ASL CBF AI, while TOF vessel length AI only showed a trend of correlating with ASL CBF AI (Fig. 5). This suggests the potential of SNAP vessel length in identifying the hemispheric blood flow asymmetry. The weaker correlation for TOF probably can be explained by its lower sensitivity to blood flow, as discussed above. In this study, the PC flow AI was not correlated with all other 3 AIs, which may be attributed to the variability of PC flow measurement caused by partial volume effect in the A2, M1 and P2 segments, noise resulted from image postprocessing and cardiac pulsation-induced signal variation, as well as the difference in the imaging mechanisms. The small range of AI may also partially contribute to the lack of correlation.

Since the vessel length may be dependent on brain size, in this study a normalization of vessel length with respect to the cube root of total brain volume was performed before the correlation analyses. In clinical practice, it is suggested to take the brain size into account when using vessel length as a surrogate marker of brain blood flow. On the other hand, in the dataset of this study, we found that even without the normalization, the vessel length was found to be correlated with ASL CBF or normalized PC blood flow (See Supplementary Table S1). This may be because the patients’ brain size did not differ much in this dataset, or the vessel length variation induced by blood flow was much larger than that by the brain size.

It is worthwhile to mention that the mean ASL GM CBF (30.6 ml/100g/min) in our data was relatively low compared to previously reported values in an aged population [34]. This may be mainly due to the more advanced age of subjects in our study (the mean age is 71.4 years), who are expected to have a lower brain perfusion than younger subjects, and suboptimal labeling efficiency of the pCASL sequence. In this study, we have used the default parameters. for the pCASL labeling on the MR scanner aimed at reducing noise but which might be suboptimal for labeling efficiency. Nevertheless, this ASL CBF underestimation is consistent for all subjects, therefore will not affect the main findings of this study, which are based on the correlation analyses.

This study is not without limitations. First, the SNAP sequence did not cover the whole brain, and some very distal branches, especially that of the ACA, were not measurable. This could have potentially affected the correlations involving the iCafe vessel length. Second, the sample size was relatively small, given that the blood flow measurements are subject to normal physiological variation. Third, current vessel tracing with the tool iCafe is still time consuming (approximately 1.5 hours per case) due to the need for manual corrections. This will limit the clinical use of the technique. A future area of research would be to develop a deep learning-based algorithm to obtain the vessel length directly, without explicit vessel tracing. Fourth, this study was performed on a cohort of subjects who had evidence of carotid atherosclerosis, although our focus was the brain blood flow, thus limiting its generalizability. This population is likely to also have intracranial stenosis, which may influence the brain blood flow, including the vessel length on TOF and SNAP. We did not specifically assess these influences due to the limited sample size. As such, the findings of this study should be further tested on other populations, including those without documented carotid disease and intracranial artery disease.

5. Conclusion

In this study, we have shown that length of the visible intracranial arteries on SNAP MRA and TOF MRA can serve as a potential surrogate imaging marker for brain blood flow.