Intracranial Time-Resolved Contrast-Enhanced MR Angiography at 3T

T.A. Cashen; J.C. Carr; W. Shin; M.T. Walker; S.F. Futterer; A. Shaibani; R.M. McCarthy; T.J. Carroll

Abstract

BACKGROUND AND PURPOSE: A method is presented for high-temporal-resolution MR angiography (MRA) using a combination of undersampling strategies and a high-field (3T) scanner. Currently, the evaluation of cerebrovascular disorders involving arteriovenous shunting or retrograde flow is accomplished with conventional radiographic digital subtraction angiography, because of its high spatial and temporal resolutions. Multiphase MRA could potentially provide the same diagnostic information noninvasively, though this is technically challenging because of the inherent trade-off between signal intensity–to-noise ratio (S/N), spatial resolution, and temporal resolution in MR imaging.

METHODS: Numerical simulations addressed the choice of imaging parameters at 3T to maximize S/N and the data acquisition rate while staying within specific absorption rate limits. The increase in S/N at 3T was verified in vivo. An imaging protocol was developed with S/N, spatial resolution, and temporal resolution suitable for intracranial angiography. Partial Fourier imaging, parallel imaging, and the time-resolved echo-shared acquisition technique (TREAT) were all used to achieve sufficient undersampling.

RESULTS: In 40 volunteers and 10 patients exhibiting arteriovenous malformations or fistulas, intracranial time-resolved contrast-enhanced MRA with high acceleration at high field produced diagnostic-quality images suitable for assessment of pathologies involving arteriovenous shunting or retrograde flow. The technique provided spatial resolution of 1.1 × 1.1 × 2.5 mm and temporal resolution of 2.5 seconds/frame. The combination of several acceleration methods, each with modest acceleration, can provide a high overall acceleration without the artifacts of any one technique becoming too pronounced.

CONCLUSION: By taking advantage of the increased S/N provided by 3T magnets over conventional 1.5T magnets and converting this additional S/N into higher temporal resolution through acceleration strategies, intracranial time-resolved MRA becomes feasible.

Several approaches to MR angiography (MRA), using different contrast mechanisms, have been developed, including time-of-flight (TOF), phase contrast (PC), T2-preparation,¹ and contrast-enhanced (CE) MRA. Of these techniques, all can produce high-quality static images, but only first-pass multiphase CE-MRA is capable of capturing the dynamic filling of vessels, similar to conventional radiographic digital subtraction angiography (DSA). Although CE-MRA cannot yet match the spatial and temporal resolutions of conventional radiographic DSA (∼0.1 × 0.1 mm and as many as ∼10 frames/s), CE-MRA is noninvasive, does not involve ionizing radiation, has no risk of iatrogenic stroke, and provides a true 3D dataset. Time-resolved MRA techniques have been applied to various anatomic regions with success.²-⁵ Previously, however, the primary goal of time-resolved MRA was to eliminate error associated with timing the arrival of contrast. As hardware and software continue to improve, noninvasive evaluation of certain pathologies where the order, direction, and rapidity of vessel filling are of clinical significance—such as with intracranial arteriovenous malformations (AVMs) and fistulas (AVFs)—is potentially within reach.⁶

MR imaging is flexible in that signal intensity-to-noise ratio (S/N), spatial resolution, and temporal resolution may be traded to accommodate the application. Time-resolved MRA, for example, requires high temporal resolution while maintaining sufficient S/N and spatial resolution. In recent years, much effort in the field has been devoted to developing novel acceleration, or undersampling, techniques to increase spatial and temporal resolution at the expense of S/N. These techniques can be classified as temporal or spatial undersampling, where temporal undersampling involves the interpolation of k-space samples from past and future samples at the same spatial frequency and includes techniques such as sliding window,⁷ keyhole,⁸ and time-resolved imaging of contrast kinetics (TRICKS).² Spatial undersampling involves the interpolation of k-space samples from other acquired samples in the same timeframe and includes techniques such as partial Fourier⁹,¹⁰ and parallel imaging.¹¹-¹⁴

The standard superconducting MR scanner found in today’s clinic operates at 1.5T. A new generation of 3T scanners has been introduced in recent years to take advantage of the theoretical doubling of S/N at double the field strength.¹⁵ Several groups have found success with high-field timed CE-MRA in various anatomic regions, including the intracranial vasculature.¹⁶-¹⁹ High-field imaging does not come without technical challenges, however; the radio-frequency (RF) wavelength is sufficiently short that dielectric resonance and RF penetration effects become significant, resulting in signal intensity nonuniformity.²⁰ Off-resonance effects also become more pronounced for sequences like bSSFP (balanced steady-state free precession) that are particularly sensitive to this. Finally, the specific absorption rate (SAR) increases as the square of the increase in field strength. Even so, if the S/N boost of high-field imaging could be converted into improved temporal resolution via an undersampling technique, higher frame rate scans would be possible.

The goals of this study were to (1) determine suitable imaging parameters for time-resolved CE-MRA at 3T, (2) verify the increase in signal intensity at high field for CE-MRA, (3) develop a method for intracranial high spatial resolution timed MRA at 3T, and finally based on previous goals, (4) develop a method for undersampled time-resolved MRA at 3T. This report builds on previous reports²¹ by providing a pulse sequence with a higher data acquisition rate as well as further acceleration via temporal undersampling. Greater reconstructed data rates enable higher frame rates and higher spatial resolution, particularly in the through-plane dimension where intravoxel dephasing in thick partitions limits sensitivity to small vessels.

Materials and Methods

All images were acquired on 1.5- and 3T whole-body MR scanners (Sonata and Trio, Siemens Medical Solutions, Erlangen, Germany). Volunteer normal subjects and patients with either intracranial arteriovenous malformations or fistulas were recruited and consented according to the guidelines of our institutional review board.

Simulations

To find appropriate imaging parameters with respect to S/N for intracranial time-resolved MRA, simulations were performed to calculate the contrast-enhanced blood S/N, and the S/N difference of contrast-enhanced blood with respect to both nonenhanced blood and the adjacent cortical gray matter as a function of T_R and flip angle, α, by using technical computing software (MATLAB, MathWorks, Natick, Mass). For the Fast Low-Angle SHot (FLASH) sequence, signal intensity, S, was taken to be the steady-state transverse magnetization in the rotating reference frame immediately after RF excitation, M_x′y′^ss(t=0₊)²²: $\text{[math]}$ 1) where M_z°(B₀) is the equilibrium longitudinal magnetization as a function of main magnetic field strength. Because T_R ≪ T*₂, T*₂ decay at T_E is negligible. The equilibrium longitudinal magnetization was normalized to 1 at 1.5T. At 3T, this magnetization was quadrupled, because it is theoretically proportional to the square of field strength.¹⁵ With noise being linearly proportional to field strength, the resulting S/N is also linearly proportional to field strength. Because the difference in proton density between contrast-enhanced blood and both nonenhanced blood and cortical gray matter is negligible, these differences were ignored in calculating S/N differences.

T₁ values of nonenhanced blood and cortical gray matter were measured at 1.5T and 3T by using a nonselective inversion recovery gradient-echo sequence (TR = 10 seconds; TE = 3.6 milliseconds; TI = [30, 120, 300, 700, 1200, 1800, 2500, 3500] milliseconds; BW = 260 Hz/pixel; flip angle = 90°; field of view [FOV] = 250 × 250 × 5 mm; matrix = 128 × 128 × 1). A least-squares fit was applied to the longitudinal magnetization regrowth measured at each inversion time for each voxel within a region of interest placed in either the sagittal sinus or in a region of cortical gray matter. T₁ values were averaged in each region of interest and then averaged again between 5 healthy volunteers. The T₁ value of contrast-enhanced blood was calculated as follows²³,²⁴: $\text{[math]}$ 2) where T₁° is the T₁ of nonenhanced blood, α₁ is the longitudinal relaxivity of the contrast agent (4.3 mmol⁻¹ s⁻¹), and C is the concentration of the contrast agent. Intra-arterial contrast agent concentration, C_IA, was estimated based on the injection protocol²³,²⁴: $\text{[math]}$ 3) where C_IV is the injected intravenous contrast agent concentration (0.5 mol/L), C_inject is the injection volume flow rate (3 mL/s), and CO is the cardiac output, assumed to be a typical value of 5 L/min. Bolus dispersion effects were ignored.

A volunteer was imaged at 1.5 and 3T by using the FLASH sequence at combinations of T_R and flip angle ranging from 2 to 5 milliseconds and 5° to 40°, respectively, to experimentally verify the relationship between these parameters at the SAR limit. The empirical data were fit according to the following approximate relationship¹⁶: $\text{[math]}$ 4) where f_t represents transmitted RF bandwidth. At the SAR limit with constant f_t, the ratio of maximum allowable flip angle at 3 to 1.5T was calculated for each T_R and then compared with the value predicted by the model, which is 2. The SAR limit for the head is currently 3.2 W/kg in both normal and first level controlled modes according to FDA guidelines.

Imaging

The gadolinium-based contrast agent (Magnevist, Berlex, Wayne, NJ) was administered with a power injector (Spectris Solaris, MEDRAD, Indianola, Pa) in an antecubital vein.

CE-MRA: 1.5T versus 3T

To compare S/N and image quality at 1.5T and 3T for MR angiography, 5 volunteers (3 women and 2 men; mean age = 31 years; age range = 22–42 years) were scanned at 1.5T and again at 3T 7–14 days later. Single-channel head coils (Siemens Medical Solutions) with similar geometry were used on both 1.5T and 3T scanners to control for coil-related differences as best as possible. First, an axial dose timing scan was positioned just proximal to the carotid bifurcation to measure the time of contrast arrival to the intracranial circulation after injection (2D FLASH, TR/TE = 19.0/1.5 milliseconds; BW = 1500 Hz/pixel; flip angle = 15°; FOV = 294 × 239 mm; matrix = 128 × 64; spatial resolution = 2.3 × 3.7 mm; temporal resolution = 1 frame/s; 2 mL of contrast agent at 3 mL/s). The time of contrast arrival was subjectively defined as the peak in the signal intensity-time curve. A timed MR angiography scan was then set up in the coronal plane with coverage of the carotid arteries, circle of Willis, and cortical arteries (3D FLASH, TR/TE = 3.4/1.3 milliseconds; BW = 425 Hz/pixel; excitation pulse duration = 0.3 milliseconds; flip angle = 20°; FOV = 300 × 169 × 70 mm; matrix = 512 × 169 × 88; spatial resolution = 0.6 × 1.0 × 0.8 mm; linear ordering, 6/8 partial Fourier in all 3 dimensions; 18 mL of contrast agent at 3 mL/s). Imaging parameters and contrast dose were held fixed across examinations, independent of the size and weight of volunteers. On the basis of the time to contrast arrival, one precontrast image, serving as the subtraction mask, and 2 postcontrast images of the arterial and venous phases, respectively, were acquired.

S/N measurements were taken from the unsubtracted arterial phase images. Signal intensity was measured as the mean intensity of a region of interest placed in the cavernous segment of the internal carotid artery; noise was measured as the standard deviation of a region of interest placed in air in the same imaging section. A qualitative analysis of image quality was performed through a blinded reading by a board-certified radiologist. Images were scored on a 5-point scale (0–4) to evaluate vessel conspicuity of the Sylvian fissure (M2) branches and the more distal cortical (M4) branches of the middle cerebral arteries. In addition, images were scored for overall image quality. A statistical comparison was performed on the quantitative and qualitative analyses by using Wilcoxon signed-rank tests at a 5% significance level.

Timed CE-MRA: Partial Fourier and Parallel Imaging

As an intermediate step in the development of the time-resolved imaging technique, high-spatial-resolution images were acquired at 3T with an 8-channel phased array head coil (Siemens Medical Solutions). The 8-channel head coil provided greater localized sensitivity and enabled the use of parallel imaging. This addition of parallel imaging to the CE-MRA protocol was first validated before subsequent addition of temporal undersampling in the time-resolved imaging protocol. A dose timing scan was run first, as described previously. Angiograms were prescribed in both the coronal plane as above, and in the sagittal plane, with coverage of one whole hemisphere, including the sagittal sinus. Both triple partial Fourier (in all 3 dimensions) and in-plane generalized autocalibrating partially parallel acquisition (GRAPPA) were used to speed up acquisition (sagittal acquisition: 3D FLASH, TR/TE = 3.4/1.3 milliseconds; BW = 425 Hz/pixel; excitation pulse duration = 0.3 milliseconds; flip angle = 15°; FOV = 220 × 220 × 75 mm; matrix = 512 × 343 × 80; spatial resolution = 0.4 × 0.6 × 0.9 mm; linear ordering; 6/8 partial Fourier in all 3 dimensions; GRAPPA 2× acceleration/24 reference lines; 18 mL of contrast agent at 3 mL/s). In addition to scanning healthy volunteers (n = 5), anecdotal data were collected from patients (n = 4) exhibiting either arteriovenous malformations or fistulas.

Time-Resolved CE-MRA: Partial Fourier, Parallel Imaging, and TREAT

High-temporal resolution/moderate-spatial resolution images were acquired at 3T with the 8-channel head coil. No dose timing scan was necessary because the temporal resolution was greater than the normal intracranial arteriovenous transit time of approximately 3 seconds.²⁵ The angiogram was prescribed with the same coverage as the sagittal timed MRA, similar to a lateral view conventional radiographic DSA examination with selective injection into the internal carotid. Triple partial Fourier, in-plane GRAPPA, and the time-resolved echo-shared acquisition technique (TREAT)²⁶ were used to achieve sufficiently high temporal resolution. Because SAR limits restrict the maximum flip angle, a slightly longer excitation pulse was used to lower the transmitter bandwidth and consequently enable higher flip angles for a 0.2-millisecond time penalty in both T_R and T_E (3D FLASH; TR/TE = 3.2/1.3 milliseconds; BW = 600 Hz/pixel; excitation pulse duration = 0.5 milliseconds; flip = 15°; FOV = 220 × 220 × 75 mm; matrix = 192 × 192 × 30; spatial resolution = 1.1 × 1.1 × 2.5 mm; temporal resolution = 2.5 seconds/frame; 6/8 partial Fourier in all 3 dimensions; GRAPPA 2× acceleration/24 reference lines; 15 mL of contrast agent at 4 mL/s). The TREAT acquisition is diagrammed in Fig 1. For each timeframe, the central segment (A) and one of the remaining peripheral segments (B, C, D, etc.) are acquired. The acquired peripheral segment is alternated for each timeframe, such that for n segments, a given peripheral segment is acquired on every (n − 1)th frame. To fill missing data in a frame, data segments are copied from the nearest acquisition in time to the present frame. As a result, the central segment has a higher temporal sampling rate than the peripheral segments. The peripheral segments are interleaved so that each segment spans data from the boundary of the central segment to the edge of k-space. In addition, acquisition is ordered in such a way that acquisition within a segment begins and ends at the boundary between the central segment and the group of peripheral segments. Thus, large jumps between subsequent lines in k-space are avoided to suppress eddy currents and ringing artifacts.²⁷ In addition to scanning healthy volunteers (n = 40), anecdotal data were collected from patients (n = 10) exhibiting either arteriovenous malformations or fistulas, where high-temporal frequency information is necessary to characterize the abnormality. In one patient, the time-resolved MRA was compared with a previous conventional radiographic DSA examination.

Fig 1.

Comparison of the TRICKS and TREAT acquisition schemes. The segmentation methods are shown in panel A. In panel B, the distance of each sample from the center of k-space (k_r) is plotted as a function of time to demonstrate the different ordering methods. Notice that the TREAT scheme acquires data without discontinuities in k_r.