1　INTRODUCTION

       As has been well-described^[1], susceptibility weighted imaging (SWI) is a 3D gradient echo magnetic resonance imaging (MRI) sequence that is fully flow compensated in all directions and exploits both the magnitude and phase of MR images. More specifically, SWI venography takes advantage of the enhanced susceptibility of deoxygenated paramagnetic hemoglobin as compared with oxygenated hemoglobin to enhance the presence of venous vascular structures via special processing that merges magnitude and phase information^[2]. Multiple studies have reported on the varied potential applications of SWI for neuroimaging, ranging from improved visualization of normal neurological vascular anatomy to better depiction of vascular malformations and to enhanced delineation of the vascular make-up of tumors^[3,4,5,6].

       In the past two decades, as the demand for improved image quality and shorter acquisition time has propelled MR performance, high-field MR has become increasingly sought after in clinical imaging. Because the susceptibility effects of deoxyhemoglobin are directly proportional to field strength and echo time, higher magnetic field strengths are predicted to allow for a decrease in echo time with a presumed improved visualization of venous structures in the brain^[7,8]. Recently, the applicability of ultra-high field MR systems (defined as systems with field strengths higher than 3 Tesla) has expanded to human brain and body imaging, offering new horizons in image resolution and contrast. One of the most prominent changes in image contrast at ultra-high field strengths is the significant boost in susceptibility or T2^* effects which offers important benefits for SWI venography.

       The primary goal of this study is to demonstrate the improved image quality of MR venography using SWI at 7 Tesla (7T) as compared to 3 Tesla (3T). The second goal is to evaluate the imaging parameters (TE, TR, bandwidth and flip angle) that yield optimal visualization of the venous structures at 7T and to establish basic principles for SWI venographic imaging at ultra-high field strengths. Finally, we describe a novel segmentation method for the visualization and quantification of the venous structures of the brain based on the improved resolution and contrast offered at 7T.

2　MATERIALS AND METHODS

2.1　Subjects

       Ten healthy adult volunteers (mean age, 27.6 years, range 23-36 years, 7 men, 3 women) were recruited for this study after meeting the following inclusion criteria: no history of cardiovascular disease and diabetes, no history of stroke or hypertension, no lesions found on conventional MRI, and no history of neurologic disease. After receiving an explanation of the study procedure, participants provided written informed consent approved by our Institutional Review Board.

2.2　Theory

       SWI is a fully velocity compensated high-resolution 3D gradient-echo sequence that uses both magnitude and phase information. The raw images of both magnitude and phase from each SWI scan were used to create new sources of contrast for venous blood deoxyhemoglobin in SWI venographic images. The standard deviation in the phase image depends on the signal in the magnitude image and varies significantly between tissues; the phase CNR is thus calculated as the contrast divided by the sum of the variances:

       where and are the standard deviations of the phase signal for the parenchyma and venous blood respectively. The standard deviation in the phase image is equal to the standard deviation in the magnitude image divided by the signal in the magnitude image. If the noise in the magnitude image is invariant across the image one can write:

       where Sp and Sb are the magnitude signal for parenchyma and blood respectively and σmag is the standard deviation in the magnitude image. This expression ignores the nonlinear effects of the magnitude operation on the noise distribution since we are interested in regions with high SNR (>3)^[9].

       Assuming a flip angle equal to the Ernst angle of the parenchyma, and also assuming TR<<T1p and TR<<T1b, the magnitude signal for each tissue is:

       These expressions were used with equation (2) for the simulation of CNR in phase images.

       The FA simulations used the standard formula for steady state incoherent imaging.

       T1 values from the literature and T2^* values measured in this study were used for these simulations (T1/T2^* for GM: 2132 ms/33.3 ms, WM: 1220 ms/26.9 ms, CSF: 4000 ms/2000 ms, venous blood: 2587 ms/16 ms)^[10].

       Contrast in the phase image is generated according to the formula:

       where γ is the gyromagnetic ratio for protons. The phase varies only with the difference in tissue susceptibility Δχ and the product B0·TE. The product B0·TE implies that if the echo time is adjusted appropriately according to field strength the phase should be invariant. This linear relationship allows much shorter echo times at higher field strengths to generate a comparable phase contrast as the long echo times at lower fields and will allow for either more rapid data acquisition or better coverage of the brain.

点击查看大图

点击查看大图

点击查看大图

点击查看大图

点击查看大图

2.3　MRI Acquisition

       All examinations were performed on both 3T and 7T (Siemens Magnetom, Erlangen, Germany) whole-body human MR systems in the same day. For the 7T scans, a 24 element coil array (Nova Medical Inc., Massachusetts) was used. For the 3T scans, a 12 element head coil array was used. SWI images were acquired with a 3D gradient-echo sequence with flow compensation in all three directions^[1]. In order to generate a protocol specifically tailored for SWI venography on both 7T and 3T, the role of repetition time (TR), echo time (TE), flip angle (FA), slice thickness, matrix, and bandwidth (BW) were all considered and were varied as follows: TR from 30-45 ms, TE from 12-32 ms, BW from 80-140 Hz/pixel and FA from 10-25°. All SWI scans used parallel imaging (GRAPPA, iPat factor of 2) at both 3T and 7T. Slice thickness ranged from 600μm to 2mm for 7T scans and from 1mm to 2mm for 3T scans with in-plane resolution ranging from 210 μm×210 μm at 7T and 420 μm×420 μm at 3T. In addition, a whole brain conventional fast spin-echo T2-weighted scan (TR/TE=6000 ms/68 ms) was also obtained as an anatomical reference. The total scanning time per volunteer was approximately 45 minutes on each magnet.

2.4　Image Post Processing and Analysis

       As described elsewhere^[1], phase images were reconstructed and filtered offline with a high pass homodyne filter using in-house image-processing software (SPIN, Detroit, Michigan). A filter size of 64×64 was used for both the 3T and the 7T images. A negative phase mask was generated and multiplied into the original magnitude image to create susceptibility-weighted (SW) images^[1,2] with the same phase multiplication factor of 4 on both 3T and 7T data. Minimum intensity projections (mIP) over 4 to 32 SWI partitions were created to visualize the connected course of venous structures. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated from selected volumetric regions-of-interest (ROI) for gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), and venous blood on magnitude, phase, and SW images in all 10 volunteers for the purpose of imaging optimization on 3T and 7T images at equivalent slice positions. Noise was measured by computing the standard deviation of a region of interest in air at the same location across all images.

       Subsequent evaluation and analysis of 3T and 7T SWI data were based on SW images, filtered phase images, and mIPs. The following steps were performed for imaging optimization.

       T2* measurements:

       T2^* measurements were performed on 2 of the 10 volunteers in order to determine venous blood transverse relaxation (T2^*) characteristics at 7T. T2^* measurement of venous blood was achieved by carefully placing a region of interest (ROI) over predefined areas of straight sinus and a large central vein. For parenchyma, an ROI in adjacent cortical GM and subcortical WM at the level of the lateral ventricles was used. Care was taken to avoid inclusion of CSF by using both the magnitude and the phase images. In order to calculate T2^*, linear regression was performed on the natural log of the 3 TE data points from 16 to 24 ms. The mean and standard deviation of T2^* values were documented for these tissues.

       Comparison between 3T and 7T SWI:

       For comparison between 3T and 7T SWI data, two different comparisons were made on data acquired from 6 volunteers. The first comparison, with data from four volunteers, used optimized parameters based on the analysis for image optimization at each field strength. This comparison included SNR, CNR between venous blood and other brain tissues, and qualitative visual analysis. Second, for a single volunteer the quantitative segmentation results of venous blood volume from the optimized sequence at 7T and 3T were also compared. The optimal parameters (i.e. TE, FA, and BW) were analyzed and determined separately for 7T and 3T data based on their SNR, CNR, and T2^* measurements.

       Quantification of Venous Vasculature:

       Surface rendered SWI venographic images and quantitative venous volume measurements were obtained through segmentation of the venous structures of the brain based on the SWI data. Segmentation results were obtained using a statistical thresholding algorithm with several stages as illustrated in Figure 1. First, a thresholding algorithm that uses both magnitude and phase was used to remove the background noise and skull as described elsewhere^[11]. Second, a statistical local thresholding algorithm was applied to single SW images to mark the veins. Finally, a shape filtering noise removal algorithm was used to remove false positives. The local thresholding algorithm used is similar to local thresholding algorithms that have been proposed for segmenting arteries^[12]. A voxel is marked as a vein if it lies 2.3 standard deviations below the mean of a local ROI (40×40×3 voxels) centered on that voxel (2.3 was empirically determined to give good results on a wide variety of SWI datasets). After all voxels have been examined, clusters of connected voxels are discarded if they are below a certain size to remove false positives. Clusters of connected voxels are further filtered through a shape analysis using the compactness^[13] and relative anisotropy (RA) of each cluster. RA is calculated based on the eigenvalues and eigenvectors of the covariance matrix for a connected object. The covariance matrix is similar to the moment of inertia tensor calculated for the center of mass. The eigenvalues and vectors describe the principal axes of the object. False positives tend to be roughly spherical and so have a high compactness and low RA while vessels are cylindrical so they have low compactness and high RA. The threshold size, compactness, and RA are set manually after visual inspection of the results.

       To compare field strengths the datasets were first co-registered with automated image registration (AIR) and the lower resolution 3T dataset was interpolated with AIR using the chirp-z algorithm to reduce blurring^[14]. A large ROI was selected to include a representative area of the brain without including regions that might introduce errors (such as the edge of the brain). The vessel segmentation was then run and the results manually reviewed to verify that the segmentation was accurate.

点击查看大图

Figure 1.  Image processing procedure and results of venous vasculature segmentation through vascular tracking using high resolution magnitude (A) and filtered phase (B) images (TR/TE/FA=45 ms/16 ms/15°; BW=100 Hz/Pixel, voxel size=0.21 mm×0.21 mm×1 mm). Processing steps involve background removal and vascular tracking on mIP images (C) to generate color-coded venous vasculature maps (D) with each color representing a single collecting venous system. The number of voxels corresponding to each venous system is used for quantification (E).

3　RESULTS

3.1　Optimization of 7T SWI Venography

       For optimization of TE, measured T2^* values were used to simulate the TE with optimal venous contrast and then images were acquired for a range of TEs suggested by the simulation (TR/TE/FA=35 ms/12-24 ms/10° and 45 ms/20-32 ms/15° at 7T and 3T respectively). The CNR in the acquired images was calculated at each TE for the phase and SW images to determine the best possible TE for venous contrast.

       The average T2^* values at 7T were 15.5±2.3 ms for venous blood in large central veins; 26.9 ± 0.4 ms for WM, and 33.4±1.2 ms for GM. As seen in Figure 2 the phase CNR simulation using these values showed an optimal echo time of 23 ms at 7T with a very broad peak. The simulation of CNR per unit imaging time and bandwidth, however, had an optimal echo time of 16 ms. The sharp decrease in CNR for TE values below 12 ms represents the point at which BW must be increased to accommodate the shorter TE leading to an increase in noise. At 7T in the filtered phase images (Figure 2 and Figure 3), the CNR between venous blood and all tissues types increases with echo time up to 24ms as predicted by the simulation. However, in the SW images (Figure 3), which represent the final output and which contain information from both phase and magnitude, CNR of venous-parenchymal tissue decreases at higher TEs due to T2^* signal losses for GM and WM, bringing their signal close to the venous signal. A reasonable compromise in CNR between venous blood and all tissues in the SW images may be achieved using a TE of 14 to 18ms. The shorter echo times provide the additional benefit of decreasing motion-related and air-tissue-interface-related artifacts as compared with longer TEs. At 3T, the TE yielding the highest CNR in the SW image was 28ms, which is close to the T2^* value of venous blood from this and other published studies^[15,16].

       The CNR in phase and SW images is also affected by FA in that the FA will change the available SNR. According to the simulations (Figure 4), FAs that give high SNR across tissue types are 10°-15° for 7T and around 12° for 3T (data not shown). For experimental data, the optimal flip angle was ascertained by assessing the CNR of the filtered phase and SW images over a range of flip angles (10°-25°) at both 7T and 3T. As shown in Figure 5, for 7T a 15° FA resulted in good CNR for all tissues, approaching the maximum available venous-GM and venous-WM CNR without decreasing the venous-CSF CNR excessively. For better visualization of the periventricular venous structures (venous-WM contrast), a higher flip angle can be used; however, since higher flip angles also increase specific absorption rates (SAR) and over suppress the CSF, the relatively low flip angle of 15° was preferred at 7T. At the same time, the best CNR at 3T also occurred at a flip angle of 15° for the SW images.

点击查看大图

Figure 2.  Top row shows simulated CNR between brain parenchyma and venous blood in phase images, to determine optimal TE at 7T. The dashed line shows a more realistic simulation of CNR divided by the square root of imaging time and BW. The shortest echo times (<12 ms) are not optimal as BW must be increased in order to achieve them, thereby lowering the SNR. The longest echo times are also not optimal, as they increase imaging time which could be used to acquire more partitions and boost SNR. This suggests a broad optimal range from 14-22 ms. The bottom row shows phase images acquired at 7T (TR/FA=35ms/10°, BW=120Hz/pixel) with different echo times. Note the increased contrast at longer echo times but also the increase in noise, blooming artifacts, and air tissue interface artifacts (frontal lobe).

点击查看大图

Figure 3.  CNR measurements to determine the optimal contrast between venous blood and other brain tissues (GM, WM, and CSF) on filtered phase (top) and post-processed SW (bottom) images with different TEs (12-24 ms) at 7T (TR/FA=30 ms/15°, BW=120 Hz/Pixel). A TE between 14 ms and 18 ms represents a good balance of contrast between venous blood and the various brain tissues, particularly in the end-result SW images.

点击查看大图

Figure 4.  Simulation curves (top row) for gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), and venous blood to determine optimal SNR with varying FAs at 7T. The bottom row shows magnitude images acquired at 7T (TR/TE=30ms/16ms, BW=100Hz/Pixel) with different flip angles (FA) (10 to 25°). An FA of 15° represents a good balance of SNR in the simulations and both SNR and contrast in the images.

点击查看大图

Figure 5.  Contrast-to-noise (CNR) measurements to determine the optimal contrast between venous blood and other brain tissues (GM, WM, and CSF) on filtered phase and post-processed SW images with different flip angles (10 to 25°) at 7T. Once again, an FA of 15° represents a good balance of contrast between venous blood and a various brain tissues, particularly in the end-result SW images.

3.2　Comparison of 3T and 7T SWI Venography

       Using optimal parameters for 3T scans (TR/TE/FA=45 ms/26 ms/20°) and 7T scans (TR/TE/FA=30 ms/16 ms/15°), we compared SNR and CNR for venous blood and surrounding brain tissues. For the SNR comparison identical resolutions were used (0.43 mm×0.43 mm×2.00 mm), then to take advantage of the increased SNR and contrast at 7T a higher resolution scan was compared (0.21 mm×0.21 mm×1.00 mm). At the same resolution the 7T scan was found to have twice the SNR as the 3T scan (20 versus 10). Figure 6 compares mIP (8 mm slab) images acquired at 3T and the high-resolution images acquired at 7T. Despite the larger matrix size (1024×1024 versus 512×512) and increased number of partitions (64 versus 32) at 7T versus 3T, the potential for a decreased TR (30 ms versus 45 ms) and decreased number of averages (1 versus 2) allowed for only minimally increased scan time (12 min 46 sec versus 9 min 49 sec, respectively). In addition, the substantially increased SNR at 7T allows higher spatial resolution, which is beneficial for visualization of small veins and venules in the cortical regions (Figure 7).

       The quantification of venous structures was based on the datasets acquired at 3T and 7T (TR/TE/FA/resolution 45 ms/35 ms/20°/0.42 mm×0.42 mm×2.00 mm and 35 ms/18 ms/15°/0.26 mm×0.26 mm×1.00 mm, respectively). The 3T scan showed a venous volume of 2.19% for the selected ROI while the 7T scan showed a 2.67% venous volume, a 22% increase in venous volume that can be visualized. Since the newly visible veins have a much smaller volume this represents a substantial increase in the number of veins that are visualized. The results of segmentation of venous vasculature using 7T SWI, which was based on high resolution (0.21 mm×0.21 mm×1.00 mm) images, is shown in Figure 8. To our knowledge, this is the first time such high quality noninvasive quantitative SWI venography at 7T has been reported. Although these higher resolutions are possible at 3T, they are not practical due to the long scan times required to recover the lost SNR.

       Figure 9 shows that the high resolution SWI venogram yields information comparable to that obtained from dye-injected roentgenograms in published postmortem data^[17]. In addition, 3D venography movie (not demonstrated here) can be generated to shows surface rendering of the segmented vasculature for better visualization of the medullary veins in the brain. Surface rendering was implemented using the visualization tool kit (VTK).

点击查看大图

Figure 6.  SWI mIP images acquired at 3T (A) and 7T (B) with the same brain coverage (8 mm). The image parameters are TR/TE/FA=45 ms/25 ms/20°, resolution 0.42 mm×0.42 mm×2.00 mm for 3T, and TR/TE/FA=30 ms/16 ms/15°, resolution 0.21 mm×0.21 mm×1.00 mm for 7T. Note the substantially improved resolution and small vessel visibility at 7T as compared to 3T. Compared to 3T, numerous additional small veins and venules are depicted at 7T.

点击查看大图

Figure 7.  High resolution 7T filtered phase images from one subject show small venules in the cortical regions, which are not visible at lower field strength.

点击查看大图

Figure 8.  7T SWI venography: 3D high resolution SWI mIP image (left), color-coded vasculature map (middle), and 3D surface-rendered venogram (right) in a healthy volunteer. Different colors in the middle image represent connected venous structures.

点击查看大图

Figure 9.  Two views of the medullary venous system in the cerebral hemisphere: A) Roentgenogram of a 10 mm thick histological dye brain slice in a "normal" postmortem brain (ref. 16); B) one 12 mm thick slice from an in vivo 7T SWI venography study in a healthy volunteer. Our noninvasive 7T data approaches the level of histologic examination regarding the detection of small deep and superficial medullary veins, arcuate veins, and subcortical veins.

4　DISCUSSION

       Our results indicate that the substantial increase in SNR and susceptibility contrast achievable at 7T as compared with 3T greatly enhances the capability of SWI venography to detect and distinguish small veins and venules in the brain, both qualitatively and quantitatively. In fact, the 7T venographic data not only provides high spatial resolution (0.21 mm×0.21 mm×1.0 mm, about 8 times smaller voxel size than at 3T, from the increased SNR and increased ability to scan faster) but also allows direct visualization of venules previously unseen in vivo (which are about 50 μm in diameter). This study also offers an initial survey of imaging parameters which may be used to achieve high-quality venography at ultra-high-field strength.

       The CNR in SWI venography is dependent on T2^* values of venous blood and brain tissues as this determines the available signal at a given echo time. It is, therefore, important to optimize the SWI acquisition parameters based on the tissue T2^* values at different field strengths. The values for GM and WM at 7T measured in this study are in good agreement with published values (33.2 and 26.8 ms respectively^[18]). At 7T, there has been little consensus on the T2^* value of venous blood in the literature, with published values ranging from T2^*=7.4 ms ^[19] and T2=6.8 ms (which implies T2^*<6.8 ms ^[20]) up to T2^*=16 ms ^[19]. The low value measured by Yacoub (T2 = 6.8±0.4 ms) was measured ex vivo with a low oxygen saturation (Y=38%). In the same study another volunteer with a more normal oxygen saturation (Y=59%) yielded T2=13.1±0.2 ms. The low value measured by Koopmans (T2^*=7.4±1.4 ms) ^[21] was measured in the sagittal sinus using a sequence that was not flow compensated. This will give artificially low values as the flow in the sagittal sinus will cause T2^*-like decay. Li ^[20] measured T2^* in 10 volunteers, finding values ranging from 11-16 ms with an average of 13±2 ms using a 2D flow compensated sequence. Our measurement of approximately 15.0±3.9 ms is in line with these values, and with the values reported by Yacoub ^[19] for normal oxygen saturation of 59%. If the normal oxygen saturation is higher than this, as some recent work has suggested ^[22,23], this lends increased weight to our values. Still, much variability is seen in measured values which might be due in part to changing oxygen saturation or errors from flow. For example, the measurements made in the sagittal sinus are less reliable because it is susceptible to flow errors due to its flow direction perpendicular to the slice-selection direction.

       Other studies have shown that the optimal TE is usually in the vicinity of the T2^* of venous blood for SWI venography ^[16]. The optimal TE of 14 to 18ms determined in this work (Figure 2 and Figure 3) is consistent with such observations. One of the driving forces for the application of SWI at higher field strengths is an increased capability for shorter TE, potentially reducing TE-related motion artifacts and allowing for faster scanning. Shorter echo times also allow increased coverage in the same amount of time with similar contrast, since the boost in SNR at short echo times from scanning faster and collecting more partitions mostly compensates for the loss of contrast. Shorter TE may also allow the SWI scan to be used for MR angiography.

       SNR is commonly used to evaluate the overall image quality, and SNR is expected to increase with B0^[24]. In our study for optimized parameters, SNR was found to be 2 times higher for SWI data at 7T than at 3T. This higher SNR and the ability to use shorter TRs (due to shorter optimal TE) allow marked increases in spatial resolutions. We have used this advantage to increase our resolution by a factor of 8. To fully exploit the advantages of the shorter TE at higher field strength, TR should be set to the minimal or near minimal value. The shorter TR can lead to substantially shorter scan times^[21] or be used to scan at a higher resolution in the same amount of time.

       The optimal flip angle was found to be the same (15°) for 3T and 7T, with a broad range of flip angles from 10-20° that give acceptable results. This range allows use of a lower flip angle to reduce SAR and modify CSF/GM/WM contrast, or else a higher flip angle to increase the time-of-flight effect and enhance arteries in thin slabs. Our data have also shown that the CSF appears darker at 3T than at 7T for the same flip angle, which makes over-suppressing the CSF at higher flip angles less problematic at 7T.

       As we have demonstrated in this work, the use of higher spatial resolution, increased SNR, and increased susceptibility contrast at 7T enables substantial improvements in the quality of SWI venography. SWI has already proved clinically useful for characterization of a number of diseases in both children and adults^[4]. Applications include, but are not limited to, identifying vascular malformations, visualizing tumoral draining veins, studying vascular changes in Sturge Weber Syndrome^[25,26], and detailing perivenous lesions in multiple sclerosis^[4]. 7T high field SWI venography offers still greater potential for increased visualization and enumeration of the small vessels involved in these pathologies. In addition, the detailed segmentation maps possible at ultra-high field strength not only allow the creation of venous maps of the brain for better understanding of venous architecture, but also enable the quantification of even small changes in the number of venules. This has potential use for evaluation and follow-up of venous disease states.

       There are several limitations to our study. The head coil arrays we used at 3T versus 7T are not identical, complicating the comparison of 3T and 7T SWI venographic data. Some differences in design are unavoidable given the differing requirements of each respective field strength. That said, one noteworthy difference in coil design is that a 12-element array was used at 3T, whereas a 24-element array was used at 7T. This might reasonably be expected to bias SNR and CNR results in favor of 7T. However, our SNR comparisons were performed near the center of the brain, reducing somewhat the impact of the increased number of surface coil elements at 7T.

       It should also be acknowledged that the use of parallel imaging in our SWI sequences complicates our ROI-based SNR and CNR analyses. A full accounting of SNR and CNR must of course include all details of both GRAPPA and SWI reconstruction and processing, and this accounting might represent a worthy topic of study in itself. However, given the comparatively large number of array elements and the comparatively low acceleration factor used in this work, inhomogeneities in the noise background are expected to be small and affect both 3T and 7T images in a similar way. We chose to adopt a simple empirical basis of comparison, with the understanding that the values reported should be taken as relative rather than absolute. Indeed, the principal focus of this paper was to demonstrate the feasibility of obtaining high quality SWI venography at 7T.

       Finally, our data is based on a small sample size and on healthy volunteers. Further studies are needed to evaluate the alteration of imaging parameters in the context of various pathologies, although we do not expect substantial changes even with pathology unless the oxygen saturation is dramatically reduced (at which point the blood oxygenation level dependent or BOLD effect becomes even more pronounced). Despite these limitations, the enhanced characteristics of SWI at 7T yield a view of veins previously documented only in histologic studies (Figure 9) ^[17].

       In summary, SWI has already been shown to offer improved visualization of a range of pathologies from hemorrhagic lesions to venous anomalies. High-quality ultra-high field venography therefore has clinical relevance in its potential not only for better detailing normal neurovascular anatomic structures, but also for discriminating the subtle abnormalities of various vascular diseases of the central nervous system.