Low-dose 3D time-resolved magnetic resonance angiography (MRA) of the supraaortic arteries: Correlation with high spatial resolution 3D contrast-enhanced MRA




To evaluate the feasibility of low-dose, 3D time-resolved contrast-enhanced magnetic resonance angiography (TR-CEMRA) in the assessment of the supraaortic vessel, and to compare the results with high-resolution contrast-enhanced MRA (HR-CEMRA).

Materials and Methods

This was an Institutional Review Board-approved retrospective study. Forty-five consecutive patients underwent contrast-enhanced 3D TR-CEMRA and 3D HR-CEMRA for evaluation of neurovascular disease at 3.0 T. Gadobutrol was administered at a constant dose of 1 mL for TR-CEMRA (independent of patient weight), and 0.1 mmol/kg for HR-CEMRA. Two readers evaluated image quality using a four-point scale (from 0 = excellent to 3 = nondiagnostic), and subsequently graded each stenosis into clinically relevant categories: normal (0%), mild stenosis (<50%), moderate to severe (>50%), and occlusion.


The overall image quality for low-dose TR-CEMRA was in the diagnostic range (median 0, range 0–3). On the grading of stenosis, TR-CEMRA using the TWIST sequence correlated with HR-CEMRA (r = 0.668, P < 0.001). In terms of the comparison of TR-CEMRA with HR-CEMRA, of the 675 supraaortic arterial segments evaluated for stenosis or occlusion, agreement occurred in 611 of 675 (90.5%), overestimation in 41 of 675 (6.1%), and underestimation 23 of 675 (3.4%).


TR-CEMRA achieved by administration of a small contrast dose (1 cc) yields rapid and important functional and anatomical information in the evaluation of supraaortic arteries. Due to limited spatial resolution, TR-CEMRA at the current parameters has a tendency to overestimate the stenosis of smaller intracranial arteries compared to HR-CEMRA. J. Magn. Reson. Imaging 2011;33:71–76. © 2010 Wiley-Liss, Inc.

DIGITAL SUBTRACTION ANGIOGRAPHY (DSA), with its excellent spatial and temporal resolution, remains the gold standard for the assessment of patients with clinical suspicion of a supraaortic steno-occlusive lesion. DSA, however, is not without small risks, including transient ischemic attack and permanent neurological deficit, thereby making it less ideal as a screening tool and for the longitudinal follow-up of patients with cerebrovascular disease (1–3). Recently, 3D contrast-enhanced magnetic resonance angiography (3D CE-MRA) (4) has become a widely clinically accepted technique for vascular imaging, and high-resolution, contrast-enhanced MRA (HR-CEMRA) has been increasingly used as an excellent alternative screening technique for morphologic imaging of cerebrovascular disease (5–7), while DSA is increasingly being performed when an intervention is being considered for extracranial and intracranial steno-occlusive disease (2).

3D time-resolved contrast-enhanced MRA (TR-CEMRA) has been previously described (8), and offers combined anatomic and hemodynamic information of the supraaortic vasculature with an extremely small dose of gadolinium chelate-based contrast agent (9). In this article we present additional results using the TWIST sequence (10, 11) for time-resolved imaging during the administration of a small contrast bolus of only 1 mL. The TWIST sequence is combined with parallel imaging GRAPPA (generalized autocalibrating partially parallel acquisition) to further improve the temporal resolution of the sequence (12). The tradeoff in using parallel imaging is a corresponding loss in signal-to-noise ratio (SNR). The use of a 1M contrast agent rather than the usual 0.5M concentration helped to compensate for the SNR losses of the parallel imaging. The 3D TR-CEMRA sequence offers combined anatomic and hemodynamic information while allowing one to retrospectively select pure arterial and venous phase images, consistently and rapidly, without a timing run.

There is a strong association between the administration of high-dose gadolinium-based paramagnetic contrast medium in patients with advanced renal failure and the recently reported complication of nephrogenic systemic fibrosis (NSF) (10, 12). Therefore, contrast agent doses as small as 0.03–0.07 mmol/kg body weight have been used successfully for TR-CEMRA at 1.5T (11, 13, 14) and 3T (15). To evaluate the feasibility and effectiveness of low-dose TR-CEMRA in the assessment of supraaortic vessel, acquired with the combination of parallel imaging (GRAPPA) and view-sharing technique (TWIST) at 3T, we individually compared TR-CEMRA with HR-CEMRA.


The study was Institutional Review Board-approved and a waiver of consent was obtained for a Health Insurance Portability and Accountability Act-compliant retrospective study. Forty-five consecutive patients (24 women, 21 men; age range 14–92 years, median age 69 years) who underwent supraaortic CE MRA on a 3T magnet (Verio; Siemens Medical Solutions, Erlangen, Germany) during a month in June 2009 were identified from the radiology database. Clinical indications for MRA included transient ischemic attack or stroke (n = 35), known extracranial or intracranial atherosclerotic disease (n = 9), and Moyamoya disease (n = 1).

Imaging protocol at our institute for the supraaortic MRA included 3D time of flight (TOF) MRA of the brain, 3D TOF MRA of the neck, and supraaortic 3D HR-CEMRA, followed by 3D TR-CEMRA (Fig. 1). For this study to define the potential for low-dose TR-CEMRA to contribute useful information of supraaortic steno-occlusive disease, we retrospectively reviewed 3D HR-CEMRA and 3D TR-CEMRA.

For HR-CEMRA, a standard automated bolus injection (Spectris, Medrad, Pittsburgh, PA) of 0.1 mmol/kg body weight of gadobutrol (Gadovist, Bayer Schering Pharma, Germany) was used at a flow rate of 1.5 cc/sec, followed by 20 cc of saline flush, at the same rate. The 3D MRA sequence with centric ordering of k-space was started manually as soon as the contrast agent was seen in the common carotid arteries on the 2D real-time fluoroscopy. Acquisition parameters were as follows: TR 3.23 msec; TE 1.22 msec; flip angle 25°; rectangular FOV (rFOV) 352 × 218; matrix 448 × 358; 224 partitions; voxel size after zero interpolation 1.0 × 0.8 × 0.6 mm3 (true voxel size 1.0 × 0.8 × 0.6 mm3); scan time 60 seconds (Table 1).

Figure 1.

A 73-year-old man with Lt. lower leg weakness. a: TR-CE MRA with subtracted coronal MIP images shows the clear visualization of arterial and venous phases. b: HR-CE MRA is obtained, with mild venous contamination mostly at the cranial vertex.

Figure 2.

TR-CEMRA (a) and HR-CEMRA (b) in a 70-year-old woman presenting with mild mental disturbance. TR-CEMRA demonstrates occlusion of the right MCA, with contrast staining at the right cerebral hemisphere on the late arterial phase, suggesting the leptomeningeal collaterals. Hypoplastic Lt. vertebral artery is not traced on early arterial phase, but is seen on the late arterial phase of TRCEMRA. Stenosis at the Lt. terminal ICA was overestimated on TR-CEMRA compared to HR-CEMRA.

Table 1. Imaging Parameters for TR-CEMRA and HR-CEMRA
Repetition time (msec) / echo time (msec)2.57/0.973.23/1.22
Flip angle (degrees)1925
Rectangular field of view (mm)420 × 341352 × 218
Matrix size448 × 318448 × 358
No. partitions36224
Slice thickness (mm)1.60.6
Voxel size (mm3)1.3 × 0.9 × 1.61.0 × 0.8 × 0.6
Parallel acquisitionGRAPPA acceleration factor 3GRAPPA acceleration factor 2

Subsequently, TR-CEMRA with TWIST was performed after intravenous injection of gadobutrol at a constant dose of 1 cc bolus at a flow rate of 1.5 cc/sec followed by 20 cc of saline flush at the same rate. Imaging parameters were as follows: TR 2.57 msec; TE 0.97 msec; flip angle 19°; rFOV 420 × 341; matrix 448 × 318; 36 partitions; voxel size after zero interpolation 1.3 × 0.9 × 1.6 mm3 (true voxel size 1.3 × 0.9 × 1.6 mm3); GRAPPA with acceleration factor 3; scan time 52 seconds; number of frames 2.2 sec/frame. On application of TWIST the user may select the percentage of center of k-space (A), compared with the entire k-space volume that is sampled with every acquisition (controlling image contrast), and the percentage of the sampling density in the periphery of k-space (B) sampled per acquisition. Due to the undersampling of the B-region, the center region A is scanned more frequently, and therefore has better temporal resolution. A preliminary study had been done with volunteers to determine optimal undersampling values of region A and B, and values of A = 8% and B = 20% of the TWIST sequence were used for TR-CEMRA in our series. This imaging protocol provides a temporal resolution of 2.2 seconds. Due to view-sharing in the image reconstruction process, the temporal footprint is 7.8 seconds. A total of 29 measurements were acquired. The baseline image was automatically subtracted and coronal and sagittal maximal intensity projection (MIP) images were scanner-generated for the TR-CEMRA sequence.

Two radiologists (K.B.S. and J.S.L., each with over 10 years experience) interpreted the coronal MIP images of HR-CEMRA and TR-CEMRA studies, in random order, on a dedicated PACS station (Marosis m-view; Infinitt, Seoul, Korea), blinded to patient history and identity. Readers were not blinded to the different MRA techniques. Separate image reading sessions were organized for both readers by the study coordinator (L.Y.J.), who attended all reading sessions. The readers were instructed to use the only coronal MIP postprocessed data. Both the HR-CEMRA and TR-CEMRA were evaluated qualitatively for image quality. Overall image quality was rated using a four-point scale with 3 = poor, nondiagnostic, 2 = fair with reservations about diagnostic content, 1 = good with confidence in diagnostic content, 0 = excellent with very high confidence in diagnostic content. HR-CEMRA and TR-CEMRA datasets were assessed for disease by viewing the MIP reconstruction image series on a workstation. Each stenosis was subsequently characterized into clinically relevant categories: normal (0%), mild stenosis (<50%), moderate to severe (>50%), and occlusion. Each HR-CEMRA and TR-CEMRA dataset was divided into 15 arterial segments—the right and left subclavian, common carotid, internal carotid, external carotid, anterior cerebral, middle cerebral, posterior cerebral and vertebral arteries, as well as the basilar artery. This resulted in a total of 675 arterial segments for analysis during both HR-CEMRA and TR-CEMRA.

A Wilcoxon rank-sum test was used to test for statistical differences between image quality ratings on HR-CEMRA and TR-CEMRA. Interobserver agreement for the image quality on each dataset, and for detection of findings on each dataset between the two readers, were determined by calculating the κ values, using a weighted kappa test (poor agreement κ = 0; slight agreement κ = 0.01–0.2, fair agreement κ = 0.21–0.4; moderate agreement κ = 0.41–0.6; good agreement κ = 0.61–0.8; excellent agreement κ = 0.81–1). Correlation between the HR-CEMRA and TR-CEMRA for the detection of stenosis was analyzed by Spearman correlation. All statistical tests were two-tailed and differences with P < 0.05 were regarded as statistically significant.


The overall image quality scores rated by both readers for the HR-CEMRA and low-dose TR-CEMRA were in the diagnostic range (median 0). Mean image quality scores for the TR-CEMRA and HR-CE MRA were 1.49 (median 0, range 1–4) and 1.29 (median 0, range 1–4) by reader 1, and 1.53 (median 0, range 1–4) and 1.40 (median 0, range 1–4) by reader 2. When assigned scores were compared with the Wilcoxon rank sum test, there was no statistically significant difference between the two observers in each imaging technique (P > 0.05 for both). In scoring of image quality, the kappa coefficient revealed good interobserver agreement for HR-CEMRA (κ = 0.66) and TR-CEMRA (κ = 0.62). There were two cases of nondiagnostic image quality for low-dose TR-CEMRA, graded by both readers, due to severe motion in patients with acute ICA occlusion. One of these two patients showed fair image quality and the other showed good image quality on HR-CEMRA. There was one case with nondiagnostic image quality for HR-CEMRA, showing severe venous contamination by the failure of bolus timing on 2D fluoroscopy. TR-CEMRA of the same patient showed excellent image quality by both readers.

For the evaluation of arterial stenosis, TR-CEMRA correlated with HR-CEMRA in both readers (r = 0.679 for reader 1 and r = 0.628 for reader 2, respectively, P < 0.001). Of the 675 supraaortic arterial segments evaluated for stenosis or occlusion, TR-CEMRA agreed with HR-CEMRA in 618 segments (91.6%) by reader 1 (K.B.S.) and 605 segments (89.6%) by reader 2 (J.S.L.) (Table 2). On TR-CEMRA, stenosis was overestimated in 39 segments (5.8%) and underestimated in 18 segments (2.7%) by reader 1, and in 40 segments (5.9%) and 30 segments (4.4%) by reader 2. In grading of stenosis, the kappa coefficient revealed moderate interobserver agreement (κ = 0.596) for TR-CEMRA and good (κ = 0.688) for HR-CEMRA. Readers demonstrated additional hemodynamic information on TR-CEMRA in three patients with severe steno-occlusive lesions.

Table 2. Evaluation of Stenoses by Two Readers at TR-CEMRA and HR-CEMRA
  1. Each HR-CEMRA and TR-CEMRA dataset was divided into 15 arterial segments. Each stenosis was subsequently characterized into clinically relevant categories: 1 = normal to mild stenosis (0–49% ), 2 = moderate to severe stenosis (50–99% ), 3 = occlusion or nonvisualization. Of the 675 supra aortic arterial segments evaluated for stenosis or occlusion, TR-CEMRA agreed with HR-CEMRA in 618 segments (91.6%) by reader 1 and 605 segments (89.6%) by reader 2. In grading of stenosis, the kappa coefficient revealed moderate inter-observer agreement (k = 0.596) for TR-CEMRA and good (k = 0.688) for HR-CEMRA.



In our series, acquisition of low-dose TR-CEMRA was feasible and could provide information about stenotic disease with slightly lower resolution compared to HR-CEMRA. Because the SNR in TR-CEMRA is proportional to the square root of data acquisition time and inversely proportional to the voxel size, low SNR is expected for high spatial resolution of images acquired with a high frame rate. Injection of contrast media with higher concentration of gadolinium compensates for the signal loss from parallel imaging and undersampling of peripheral k-space data (16). In the present study, injection of 1 cc of gadobutrol was sufficient to provide good image quality of TR-CEMRA at a frame rate up to 2.2 seconds and morphological information of supraaortic vessels as well. In the evaluation of peripheral arteries with TR-CEMRA, Kramer et al (17) reported that the decrease in SNR because of implementation of parallel imaging techniques can be absorbed by a high-concentration contrast agent. As well, they showed that a relatively slow injection rate (0.4 cc/sec) showed the best diagnostic results with a good arterial signal without disturbing venous overlay, utilizing its advantage in cases where good temporal resolution is needed without high spatial resolution. A faster injection scheme, on the contrary, results in fast signal increase with a great maximum. This is beneficial to accurately assess the degree of a high-grade stenosis and to differentiate between a pseudo-occlusion and an occlusion with distal retrograde vessel filling from collateral vessels. HR-CEMRA of carotid artery with elliptic centric k-space filling and submillimeter voxel sizes is a typical example of fast contrast injection (18).

The administration of high-dose gadolinium-based paramagnetic contrast medium has been recently reported in association with NSF, especially in patients with advanced renal failure (10, 11). The American College of Radiology recommends using gadolinium-based contrast agents in patients with a GFR <30 ml/min/1.73 m2 only if it is absolutely essential; in these cases, the lowest possible dose has to be administered (19). Low-dose TR-CEMRA thus has the potential for dose reduction of contrast media during acquisition of CEMRA. In our institute, CE MRA protocol for screening or diagnosis of supraaortic atherosclerotic disease includes injection of 0.1 mmol/kg of gadobutrol for HR-CEMRA and additional injection of 1 cc of gadobutrol for TR-CEMRA, which is much less than the current regulatory cumulative daily dose limit of 0.3 mmol/kg/day (20). For patients at risk of NSF in whom there will only be opportunity of single injection of limited amount of contrast media, a modified protocol with single acquisition of TR-CEMRA with slightly higher doses than 1 mL of contrast media can possibly be applied, although the parameter should be further optimized for proper vascular SNR and better spatial resolution.

Obtaining the low-dose TR-CEMRA can also be used as a test bolus technique to obtain information regarding the time between the injection of contrast media and arrival of contrast bolus in the vessel of interest. Although fluoroscopic detection of bolus arrival to trigger the HR-CEMRA is mostly performed by experienced MR technicians in our institute, predetermined bolus transit time by low-dose TR-CEMRA gives more confidence to the technician when they obtain the subsequent HR-CEMRA.

Lim et al (14) showed that TR-CEMRA with TWIST and parallel imaging had good accuracy (90.3%) for identifying stenosis involving the carotid artery. Lohan et al (9) reported that low-dose (1.5 or 3.0 cc) TR-CEMRA can reliably detect or rule out hemodynamically significant steno-occlusive disease in the carotid-vertebral arterial territory relative to HR-CEMRA. Furthermore, they showed that low-dose TR-CEMRA may be performed with 3.0 cc of gadolinium chelate with preservation of overall image quality and arterial segmental visualization relative to HR-CEMRA, whereas a 1.5 cc contrast dose is associated with more suboptimal studies. In our study with low-dose (1.0 cc) 3D TR-CEMRA in combination with GRAPPA and TWIST at 3T, there was a significant correlation and highly concordant results with HR-CEMRA in categorizing the stenosis of supraaortic arteries. With regard to the incorrect categorization compared to HR-CEMRA, our study showed that TR-CEMRA overestimated stenosis in 5.8% (39/618) and 5.9% (40/618) of evaluated arterial segments, and underestimated in 2.7% (18/618) and 4.4% (30/618) by two readers. A possible explanation of discordant results with larger numbers of overestimation would be a larger voxel size of TR-CEMRA compared to HR-CEMRA. To make the best of temporal resolution during acquisition of TR-CEMRA, our protocol was to use a relatively large slice thickness (1.6 mm) compared to HR-CEMRA (0.6 mm after interpolation). The fraction of undersampling at the central (A) and peripheral k-space region (B) may also contribute to the relatively poor accuracy of the sequence in the moderate stenosis range of smaller vessels (14). Further work to optimize the sampling proportion of k-space with respect to the caliber of the vessels of interest is needed.

Injection of contrast agent for a relatively short duration in our protocol may also have decreased the SNR of small-vessel conspicuity. Beranek-Chiu et al (21) reported that a low concentrated contrast agent flush, in comparison to a pure saline flush, during the acquisition of 4D CE MRA with CENTRA, SENSE factor of 3, and keyhole factor of 10% significantly improves contrast and delineation of the vessel in the TR-CEMRA technique. Beranek-Chiu et al suggested that prolongation of the intravascular contrast level by gadolinium-doped saline flush not only increases the vascular signal intensity during the late acquisition period, but also increases signal homogeneity, improves the point spread function, and leads to less blurring due to a reduced washout. The effect of changing both the contrast duration and the flushing method also needs to be further verified on TR-CEMRA sequence with TWIST.

With a combination of parallel imaging (GRAPPA) (22) and view-sharing technique (TWIST) (14, 23), 3D TR-CEMRA offers combined anatomic and hemodynamic information and obtains pure arterial and venous phase images consistently. In our study, three patients with severe steno-occlusive lesions demonstrated additional hemodynamic information, including collateral flow through the circle of Willis, or leptomeningeal collateral vessels on TR-CEMRA (Fig. 2).

Two cases in our study failed to demonstrate diagnostic image quality of the supraaortic vessels due to severe motion during the acquisition of TR-CEMRA. Although with high temporal resolution, as in our study (2.2 sec/frame), a keyhole imaging approach which would sample the center of k-space more frequently compared to high-spatial-frequency information peripherally still has limitations in patients with severe motion artifacts.

Our study has the limitation that low-dose TR-CEMRA was retrospectively compared with HR-CEMRA, and it was not compared with surgery or DSA. Because there were only two patients who had CEMRA followed by DSA in our series, a statistical comparison could not be made. Both TR-CEMRA and HR-CEMRA showed concordant findings of stenosis involving the cervical internal carotid artery and vertebral artery, respectively. Further study to compare it with DSA in larger number of patients would provide detailed information regarding the accuracy in diagnosis and grading of stenosis.

In conclusion, acquisition of low-dose TR-CEMRA by administration of a small contrast dose (1 cc of gadobutrol) is feasible, and shows relatively high concordant results in grading of stenosis. As well, it adds hemodynamic information to better evaluate atherosclerotic disease of the supraaortic arteries. Due to limited spatial resolution, however, TR-CEMRA at the current setting has a tendency to overestimate the stenosis compared to HR-CEMRA.