NOTE Magnetic Resonance in Medicine 75:831–838 (2016)
Joint Blood and Cerebrospinal Fluid Suppression for Intracranial Vessel Wall MRI Jinnan Wang,1* Michael Helle,2 Zechen Zhou,3 Peter B€ ornert,2 Thomas S Hatsukami,4 and Chun Yuan3,5 Purpose: To develop and evaluate a joint blood and cerebrospinal fluid (CSF) suppression technique for improved intracranial vessel wall MR imaging. Methods: The Delay Alternating with Nutation for Tailored Excitation (DANTE) prepulse was specifically optimized for CSF suppression to improve vessel wall and CSF contrast. It was evaluated on six patients and three healthy volunteers. CSF suppression efficiency, lumen signal to noise ratio, and wall-lumen contrast to noise ratio were compared between images with and without DANTE in major intercranial artery segments. Contrast changes in tissues were also compared with evaluate the technique’s compatibility with multicontrast imaging techniques. Results: The optimized DANTE images significantly improved intracranial vessel wall characterization on all images. Quantitatively, CSF to wall contrast improved by 28% (DANTE-VISTA 1.354 6 0.216 versus VISTA 1.057 6 0.13; P < 0.001). DANTE also significantly improved wall-lumen (10.55 6 3.79 versus 9.34 6 3.54; P < 0.001) and wall-CSF (4.62 6 3.19 versus 0.78 6 2.30; P < 0.001) contrast-to-noise ratios. DANTE prepared images were also found to make only minimal impact on static tissue contrast. Conclusion: DANTE prepared MR imaging can significantly improve contrast between the vessel wall and cerebral spinal fluid in major intracranial arteries, holding a good potential to be combined with multicontrast protocol for intracranial wall C 2015 Wiley imaging. Magn Reson Med 75:831–838, 2016. V Periodicals, Inc. Key words: DANTE; intracranial artery; vessel wall
INTRODUCTION Intracranial atherosclerotic disease (ICAD) causes 9–15% of all strokes in the United States (1), accounting for approximately 70,000 strokes annually (2). Due to its even higher prevalence in Asian and other non-Caucasian racial groups (3–6), ICAD is presumed to 1
Philips Research North America, Briarcliff Manor, New York, USA. Philips GmbH Innovative Technologies, Research Laboratories, Hamburg, Germany. 3 CBIR, Tsinghua University, Beijing, China. 4 Department of Surgery, University of Washington, Seattle, Washington, USA. 5 Department of Radiology, University of Washington, Seattle, Washington, USA. 2
*Correspondence to: Jinnan Wang, Ph.D., 850 Republican Street, Seattle, WA 98109. E-mail: [email protected] Received 25 September 2014; revised 2 February 2015; accepted 2 February 2015 DOI 10.1002/mrm.25667 Published online 13 March 2015 in Wiley Online Library (wileyonlinelibrary. com). C 2015 Wiley Periodicals, Inc. V
represent the most common subtype of ischemic stroke worldwide (7). Luminal imaging, such as CT or MR angiography, has traditionally been the only radiographic tool to evaluate intracranial vasculopathies. Studies of other vascular beds clearly demonstrate that angiography does not detect atherosclerotic lesions with enough sensitivity, and may underestimate plaque burden due to outward remodeling (8,9). In an autopsy study of over 300 subjects, 62% of the patients who died from ischemic stroke were found to have ICAD (10). Among all the subjects with ICAD, nearly 40% of all lesions presented with only minor or no luminal narrowing at all (10). A comprehensive evaluation of ICAD related risk factors for stroke can benefit from the direct evaluation of the vessel wall. The core technology for extracranial atherosclerotic wall imaging is multicontrast MR imaging, which has been used to measure in vivo wall size and plaque components (11,12). However, very few techniques are at the clinicians’ disposal for intracranial vessel wall characterization due to the technical challenge of simultaneous suppression of signals from flowing blood and the surrounding cerebrospinal fluid (CSF) (13). A few attempts have been made to address this challenge. Van der Kolk et al described a magnetization preparation inversion recovery (MPIR) based technique that is optimized so that both CSF and blood can be suppressed at the time of imaging (14). But this technique requires a long preparation time due to the very long T1 relaxation time of CSF. In the original report, it takes 12 min to image a thin slab of 13 mm (14). Additionally, the IR pulse, introduced to create a strong T1 weighting on the final images, renders the technique less compatible with other contrast weighting (T2/PD) schemes. Another ICAD imaging technique, a proton densityweighted Volumetric ISotropic Turbo spin echo Acquisition (VISTA) was proposed with much higher scan efficiency (15). This technique, however, still lacks a dedicated CSF suppression module and results in the CSF giving a confusing isointense signal when compared with the vessel wall. When the technique is tailored to T2 weighting, a strong hyperintense CSF signal reduces the visibility of the vessel wall, making the evaluation of size and components more challenging. As such, a clear technical gap in time-efficient CSF suppression with compatibility to multicontrast imaging techniques still exists. This study developed a fast joint blood and CSF (JBC) suppression technique. It was tested and optimized in a group of patients with intracranial atherosclerotic disease and three healthy volunteers. Its efficiency in suppressing CSF signal was compared with existing
831
832
FIG. 1. Joint blood and CSF suppression sequence diagram (a) and optimization plots (b,c). a: The sequence contains optimized DANTE prepulse, which is formed by a train of equidistant RF pulses repeated at an interpulse delay TD concomitant by gradient pulses to facilitate spatial selectivity, a fat saturation and a corresponding acquisition module. b: As shown in Optimization plot of wall and CSF magnetization difference, at the fixed DANTE sequence duration, shorter TD provided better contrasts. Contrast peaked when TD was 1 ms and FA is 12 (arrow). c: The vessel wall/CSF signal at the 1 ms TD plot shows that the corresponding wall and CSF magnetizations (Mz/M0) are 0.964 and 0.011, respectively.
image acquisition sequences, and its compatibility with multicontrast imaging sequences was evaluated. METHODS Pulse Sequence Optimization The JBC pulse sequence is shown in Figure 1a: it contains a delay alternating with nutation for tailored excitation (DANTE) prepulse, a fat saturation module and an acquisition module. The DANTE prepulse (16) has been
Wang et al.
proposed for tagging (17) and flow suppression (18) before, and is expected to help suppress signals from both the blood and CSF (19). When used for flow suppression, the DANTE inter pulse gradient is used as a phase introducer by providing constant phase increments for static tissue and quadratic phase increments for moving magnetizations (signal spoiling). Although perivascular fat is not expected around the intracranial arteries, the fat saturation module can help suppress the normally hyperintense fat signal from the skull, thereby avoiding the dynamic range compression from fat signals on digitized images. The acquisition sequence can be any MRI technique that fits the clinical application. In this study, both TSE and VISTA (20) sequences were evaluated because of their relatively lower sensitivity to susceptibility caused signal dropout compared with gradient echo based acquisition techniques. The DANTE pulse was optimized to achieve the highest contrast between the intracranial vessel wall and the surrounding CSF. By using the previously established DANTE signal model (18), signal evolution for both vessel wall and CSF can be simulated. This simulation assumes: the same unity initial magnetization (M0) for both vessel wall and CSF; complete magnetization spoiling in flowing spins (18); and, partial phase coherence among non-flowing spins (18). Longitudinal magnetizations were simulated at the temporal resolution of the delay time (TD). Based on our previous measurements on the same scanner, T1/T2 relaxations times for CSF and vessel wall were assumed to be 3120/160 ms and 1114/55 ms. All simulations were conducted in Matlab (Mathworks R2013a, Natick, MA), using custom designed simulation programs. Considering only the signal evolution in the DANTE pulse train, two key parameters (TD and flip angle of the elementary sequence in the DANTE train) were jointly optimized to achieve the highest signal difference (as measured by normalized magnetization Mz/M0) between vessel wall (MVW) and CSF (MCSF). TD ranges of 1–8 ms and FA ranges of 1 –50 were tendered for optimization. No TD of less than 1 ms was simulated because the minimum TD achievable on our scanner was 0.86 ms due to hardware limitations. Due to the overall small flip angles used in the DANTE prepulses, it usually takes a long time before the signal reaches its steady state, therefore, the transient signal patterns in each TD in the DANTE pulse sequence were calculated to ensure accurate simulation. For the same reason, a very long RF pulse train (over 500 ms for certain parameter sets) is required to achieve the steady state in the DANTE pulse, which can significantly prolong the scan time, making clinical implementation impractical. To avoid this, a total DANTE pulse duration of no more than 256 ms (rather than use the time required for the actual steady state to be achieved) was used as a cut-off point in this simulation. For each TD/FA parameter pair, the highest MVW and MCSF difference achieved within 256 ms was used as the optimal value for this pair. Among all the TD/FA pairs, the one that provided the highest MVW and MCSF difference was used in the following experiments. As optimal parameters were identified, vessel wall and CSF
JBC Suppression for Intracranial Wall MRI
signal changes were plotted to evaluate the vessel wall signal drop at that parameter setting. Patient Recruitment Three healthy volunteers and six patients with diagnosed or suspected intracranial artery disease, based on clinical angiography, were recruited for this study. Local institutional research board (IRB) approved the study and informed consent was obtained from all subjects. The images collected from the six patients were used to evaluate the CSF suppression efficiency provided by JBC imaging; the data from the healthy volunteers were used to evaluate the compatibility of JBC with multicontrast imaging. MR Scans All MR scans were conducted on a 3 Tesla (T) whole body scanner (Philips Achieva R3.21, Best, the Netherlands) using an eight-channel brain coil from the same vendor. After the scout scan and a whole-brain MRA used for localizing the intracranial arteries, the following JBC scans were performed. For CSF suppression efficiency evaluation, PDweighted VISTA scans with and without the DANTE prepulse were performed. The two scans used identical parameters with the exception of the DANTE prepulse, which used the optimized parameter sets identified in the previous section. The detailed VISTA scan parameters were configured similarly to the previous report (15), as follows: TR/TE 2000/19 ms, FOV 210 180 48 mm3, voxel size 0.6 0.6 0.6 mm3, echo train length 38, SENSE factor 2, and axial orientation. The total scan times were the same for both scans at 5 min 29 s. For compatibility with multicontrast imaging evaluation, standard T2-weighted TSE sequences with and without DANTE pulses were acquired. The scanning parameters were: repetition time/echo time (TR/TE) 4000/ 80 ms, field of view (FOV) 220 220 mm2, voxel size 0.6 0.6 mm2, slice thickness 4 mm, echo train length 12, and axial orientation. The scan time for 16 slices was 3 min 40 s. After inclusion of DANTE prepulses, the imaging time remained the same for both VISTA and T2 scans. In Vivo CSF Suppression Efficiency Evaluation To evaluate the CSF suppression efficiency in JBC imaging, the ratio between vessel wall and the neighboring CSF was used as a surrogate. For both VISTA and DANTE VISTA datasets, cross-sectional images of the intracranial vessel wall were first reformatted to evaluate three key arteries of the intracranial vascular tree: basilar, left and right middle cerebral arteries (MCAs; M1 segment). After registration, 5 consecutive slices from each segment (15 total for each subject) were selected from the DANTE VISTA images. When possible, intracranial artery portions with close contact to brain parenchyma were avoided to ensure the clear visualization of outer wall boundaries. On the cross-sectional images, inner and outer wall boundaries of the intracranial arteries were delineated and a neighboring CSF region was selected for signal measurements, using custom-
833
made plaque analysis software, CASCADE (21). As the two VISTA image sets were registered, the contours from DANTE VISTA were automatically transferred to the original VISTA image for measurements. The signal ratio is defined as, R ¼ SVW =SCSF
[1]
where SVW is the signal measured on the vessel wall and SCSF is the signal measured on the CSF. In this measurement, a higher R is desirable as it indicates a more efficient CSF signal suppression. To facilitate comparison of the results, CSF SNR and wall-CSF CNR were also measured. The inner and outer wall boundaries delineated in the above analysis were used in this comparison. The signal standard deviation, measured in the nasal cavity, was defined as the image noise. SNR and CNR were defined below, SNR ¼
S ; CNRwall-CSF ¼ SNRwall -SNRCSF snoise
[2]
where, S is the average signal of the region of interest, s is the standard deviation of the noise region. In this evaluation, a lower CSF SNR was desirable as it represents more efficient CSF suppression; a higher wall-CSF CNR was desirable as it means better wall characterization. In Vivo Flow Suppression Evaluation In addition to CSF suppression, blood suppression is crucial in quantifying atherosclerotic plaque burden for intracranial arteries. To evaluate JBC’s performance on blood suppression in intracranial arteries, lumen SNR and walllumen CNR were evaluated. The same SNR and CNR calculation method was used as described in the CSF suppression efficiency evaluation section. In this evaluation, a lower lumen SNR was preferable, as it results in a complete blood signal suppression, while a higher wall-lumen CNR means better vessel wall characterization. Image Contrast Change Evaluation As explained before, a desirable blood/CSF suppression module should be compatible with multicontrast imaging acquisitions so that established plaque component imaging techniques can be used. With this requirement, static tissue’s relative signal change on a standard contrast weighting image needed evaluation. A numerical simulation was first conducted to evaluate the static tissue signal change after the application of DANTE (as measured by the signal ratio between images with and without the DANTE prepulse). A wide T1 range of 200–1500 ms was used to cover most static tissue T1 relaxation times at 3T. The same simulation program used in the optimization section was performed. The signal ratios were plotted against the simulated T1 values. As the CSF signal can present hypo-, iso-, and hyperintensive signal patterns on in vivo T1w, PDw, and T2w images, evaluating JBC on T2w images represented the highest relevance. To rule out the impact from image scaling between different acquisitions, the average signal on the brain stem region was used as the reference base
834
Wang et al.
FIG. 2. DANTE PD VISTA allows for more effective intracranial artery wall delineation by suppressing CSF signal (a,c,e), as compared to images without DANTE prepulse (b,d,f). Arterial walls can be better visualized in MCA left (a,b), MCA right (c,d), and basilar segments (e,f), indicated by the arrows. A clear wall thickening can be visualized on the DANTE-VISTA image (a), indicated by the dotted arrow, but its appearance is overshadowed by the nearby hyperintense CSF flow artifact on VISTA image (dotted arrow, b).
for both DANTE and regular T2w TSE images. The brain stem was selected as the reference due to its size and a relatively homogenous signal on T2-weighted images. Normalized DANTE and regular TSE images were obtained after dividing the tissue signal by the reference signal (average signal on brain stem) for each pixel. To facilitate the evaluation of minor signal changes, false color images were also rendered using Matlab’s standard “Jet” colormap. Statistical Analysis All statistical comparisons were conducted using Excel software (Microsoft Corp, Redmond, WA). A paired Student’s t-test was used to compare the R difference between DANTE VISTA and VISTA images; unpaired unequal-variance t-tests were used to evaluate the R difference among different arteries. Paired Student’s t-test was used to compare all SNR and CNR measurements between DANTE VISTA and VISTA images. A P-value of equal or less than 0.05 was considered significant. RESULTS Pulse Sequence Optimization The numerical simulation indicated that the contrast between vessel wall and CSF signal rapidly peaks and then gradually decreases as the FA increases for all TD
values. Overall a shorter TD promised better contrast while given the same total duration of the DANTE prepulse train. As shown in Figure 1b, maximal contrast was achieved when TD was 1 ms and FA was 12 , while the total number of TD was 256. Because primary flow directions in basilar and MCA arteries are foot–head and left–right directions, gradients in both directions were turned on in the DANTE prepulse. With our hardware settings, the gradient parameters used in DANTE were: gradient strength 30 mT/m and gradient duration 0.72 ms. At these parameter settings, vessel wall and CSF signal changes at different flip angles were also plotted (Fig. 1c). As indicated by the arrows, optimized DANTE parameters caused only 3.6% signal loss on the vessel wall. These parameters were used as the optimal parameters in the following in vivo experiments. In Vivo CSF Suppression Efficiency Evaluation A total of 90 locations were obtained from six patients— five slices from each major artery (basilar, left MCA and right MCA) of each subject. Qualitatively, DANTE prepared PD VISTA sequence was found to provide improved intracranial artery wall delineation by effectively suppressing CSF signal, as compared to the sequence without DANTE prepulse (Fig. 2). In both MCA and basilar arteries, outer wall boundaries were better
JBC Suppression for Intracranial Wall MRI
835
Suppression efficiency by JBC at different vascular beds showed more efficiency on basilar arteries, as compared to MCAs. However, there was no statistical difference among different pairs of vascular beds (P > 0.05 for all). CSF SNR and wall-CSF CNR comparisons yielded that DANTE-VISTA presented significantly lower CSF SNR and higher wall-CSF CNR in all vascular beds (P < 0.001 for all), as shown in Table 1. This result confirmed that DANTE-VISTA provided effective CSF suppression and improved the characterization of intracranial vessel walls. FIG. 3. Signal ratio comparison between DANTE and non-DANTE images. DANTE provides a significantly higher signal ratio across all arteries (P < 0.001). The error bar represents the standard error of each measurement. *Signal ratio is significantly lower than the corresponding JBC image (P < 0.001)
delineated on DANTE VISTA images (Fig. 2a,c,e), as compared to regular VISTA images (Fig. 2b,d,f). Apparent wall thickening was visualized on the DANTEVISTA (Fig. 2a), as indicated by the dotted arrow, but its appearance was overshadowed by a nearby hyperintense CSF flow artifact on the VISTA image (Figure 2b, dotted arrow). Flow artifacts were also present in the same region. Quantitatively, a significantly higher signal ratio between vessel wall and CSF was obtained on DANTE prepared images in all major vascular beds (Fig. 3). On average, the vessel wall signal was 35.4% higher than the neighboring CSF region on DANTE VISTA images (R ¼ 1.354 6 0.216), while only 5.7% higher on regular VISTA images (R ¼ 1.057 6 0.13). This difference was statistically significant (P < 0.001). The same pattern of R maintains when each individual artery bed was examined, as detailed in Figure 3.
In Vivo Flow Suppression Evaluation DANTE prepared VISTA was also found to help improve the lumen boundary definition by further suppressing blood signal. As shown in Table 1, intracranial artery luminal signals on JBC images were significantly lower than those of regular VISTA images (DANTE-VISTA 7.87 6 3.03 versus VISTA 9.69 6 3.77; P < 0.001), indicating a more complete blood suppression because of the DANTE magnetization preparation. At the same time, wall-lumen CNR was significantly higher on DANTE images (DANTE-VISTA 10.55 6 3.79 versus VISTA 9.34 6 3.54; P < 0.001), indicating a more favorable lumen-wall contrast. It worth mentioning that the wall SNR of DANTE images is slightly (3.2%) lower than that of VISTA images, which was not found to be statistically significant (DANTE-VISTA 18.43 6 6.44 versus VISTA 19.03 6 6.22; P ¼ 0.339). The difference was comparable to the predicted simulation (3.6% in Figure 1c). Multicontrast Compatibility Evaluation The simulation (Sup. Fig. S1, which is available online) found that DANTE makes nearly homogeneous impact on tissues with a wide range of T1. The maximal
Table 1 Blood and CSF suppression efficiency comparisons between DANTE VISTA and regular VISTA images DANTE VISTA Basilar
MCA Left
MCA right
All
Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR
16.59 6.76 9.82 12.17 4.43 19.44 8.13 11.31 14.26 5.18 19.25 8.71 10.53 14.98 4.27 18.43 7.87 10.55 13.80 4.62
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6.82 3.71 3.35 5.41 2.65 7.66 2.84 5.01 4.30 4.37 4.12 2.06 2.57 3.59 2.15 6.44 3.03 3.79 4.61 3.19
VISTA 18.27 9.01 9.25 16.59 1.67 18.83 9.49 9.34 18.03 0.80 19.99 10.56 9.43 20.13 0.14 19.03 9.69 9.34 18.25 0.78
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6.12 3.31 3.92 5.69 2.64 5.84 3.72 3.17 5.98 1.92 6.75 4.19 3.62 6.61 1.97 6.22 3.77 3.54 6.21 2.30
P-value 0.077 0.002 0.307
Joint Blood and Cerebrospinal Fluid Suppression for Intracranial Vessel Wall MRI Jinnan Wang,1* Michael Helle,2 Zechen Zhou,3 Peter B€ ornert,2 Thomas S Hatsukami,4 and Chun Yuan3,5 Purpose: To develop and evaluate a joint blood and cerebrospinal fluid (CSF) suppression technique for improved intracranial vessel wall MR imaging. Methods: The Delay Alternating with Nutation for Tailored Excitation (DANTE) prepulse was specifically optimized for CSF suppression to improve vessel wall and CSF contrast. It was evaluated on six patients and three healthy volunteers. CSF suppression efficiency, lumen signal to noise ratio, and wall-lumen contrast to noise ratio were compared between images with and without DANTE in major intercranial artery segments. Contrast changes in tissues were also compared with evaluate the technique’s compatibility with multicontrast imaging techniques. Results: The optimized DANTE images significantly improved intracranial vessel wall characterization on all images. Quantitatively, CSF to wall contrast improved by 28% (DANTE-VISTA 1.354 6 0.216 versus VISTA 1.057 6 0.13; P < 0.001). DANTE also significantly improved wall-lumen (10.55 6 3.79 versus 9.34 6 3.54; P < 0.001) and wall-CSF (4.62 6 3.19 versus 0.78 6 2.30; P < 0.001) contrast-to-noise ratios. DANTE prepared images were also found to make only minimal impact on static tissue contrast. Conclusion: DANTE prepared MR imaging can significantly improve contrast between the vessel wall and cerebral spinal fluid in major intracranial arteries, holding a good potential to be combined with multicontrast protocol for intracranial wall C 2015 Wiley imaging. Magn Reson Med 75:831–838, 2016. V Periodicals, Inc. Key words: DANTE; intracranial artery; vessel wall
INTRODUCTION Intracranial atherosclerotic disease (ICAD) causes 9–15% of all strokes in the United States (1), accounting for approximately 70,000 strokes annually (2). Due to its even higher prevalence in Asian and other non-Caucasian racial groups (3–6), ICAD is presumed to 1
Philips Research North America, Briarcliff Manor, New York, USA. Philips GmbH Innovative Technologies, Research Laboratories, Hamburg, Germany. 3 CBIR, Tsinghua University, Beijing, China. 4 Department of Surgery, University of Washington, Seattle, Washington, USA. 5 Department of Radiology, University of Washington, Seattle, Washington, USA. 2
*Correspondence to: Jinnan Wang, Ph.D., 850 Republican Street, Seattle, WA 98109. E-mail: [email protected] Received 25 September 2014; revised 2 February 2015; accepted 2 February 2015 DOI 10.1002/mrm.25667 Published online 13 March 2015 in Wiley Online Library (wileyonlinelibrary. com). C 2015 Wiley Periodicals, Inc. V
represent the most common subtype of ischemic stroke worldwide (7). Luminal imaging, such as CT or MR angiography, has traditionally been the only radiographic tool to evaluate intracranial vasculopathies. Studies of other vascular beds clearly demonstrate that angiography does not detect atherosclerotic lesions with enough sensitivity, and may underestimate plaque burden due to outward remodeling (8,9). In an autopsy study of over 300 subjects, 62% of the patients who died from ischemic stroke were found to have ICAD (10). Among all the subjects with ICAD, nearly 40% of all lesions presented with only minor or no luminal narrowing at all (10). A comprehensive evaluation of ICAD related risk factors for stroke can benefit from the direct evaluation of the vessel wall. The core technology for extracranial atherosclerotic wall imaging is multicontrast MR imaging, which has been used to measure in vivo wall size and plaque components (11,12). However, very few techniques are at the clinicians’ disposal for intracranial vessel wall characterization due to the technical challenge of simultaneous suppression of signals from flowing blood and the surrounding cerebrospinal fluid (CSF) (13). A few attempts have been made to address this challenge. Van der Kolk et al described a magnetization preparation inversion recovery (MPIR) based technique that is optimized so that both CSF and blood can be suppressed at the time of imaging (14). But this technique requires a long preparation time due to the very long T1 relaxation time of CSF. In the original report, it takes 12 min to image a thin slab of 13 mm (14). Additionally, the IR pulse, introduced to create a strong T1 weighting on the final images, renders the technique less compatible with other contrast weighting (T2/PD) schemes. Another ICAD imaging technique, a proton densityweighted Volumetric ISotropic Turbo spin echo Acquisition (VISTA) was proposed with much higher scan efficiency (15). This technique, however, still lacks a dedicated CSF suppression module and results in the CSF giving a confusing isointense signal when compared with the vessel wall. When the technique is tailored to T2 weighting, a strong hyperintense CSF signal reduces the visibility of the vessel wall, making the evaluation of size and components more challenging. As such, a clear technical gap in time-efficient CSF suppression with compatibility to multicontrast imaging techniques still exists. This study developed a fast joint blood and CSF (JBC) suppression technique. It was tested and optimized in a group of patients with intracranial atherosclerotic disease and three healthy volunteers. Its efficiency in suppressing CSF signal was compared with existing
831
832
FIG. 1. Joint blood and CSF suppression sequence diagram (a) and optimization plots (b,c). a: The sequence contains optimized DANTE prepulse, which is formed by a train of equidistant RF pulses repeated at an interpulse delay TD concomitant by gradient pulses to facilitate spatial selectivity, a fat saturation and a corresponding acquisition module. b: As shown in Optimization plot of wall and CSF magnetization difference, at the fixed DANTE sequence duration, shorter TD provided better contrasts. Contrast peaked when TD was 1 ms and FA is 12 (arrow). c: The vessel wall/CSF signal at the 1 ms TD plot shows that the corresponding wall and CSF magnetizations (Mz/M0) are 0.964 and 0.011, respectively.
image acquisition sequences, and its compatibility with multicontrast imaging sequences was evaluated. METHODS Pulse Sequence Optimization The JBC pulse sequence is shown in Figure 1a: it contains a delay alternating with nutation for tailored excitation (DANTE) prepulse, a fat saturation module and an acquisition module. The DANTE prepulse (16) has been
Wang et al.
proposed for tagging (17) and flow suppression (18) before, and is expected to help suppress signals from both the blood and CSF (19). When used for flow suppression, the DANTE inter pulse gradient is used as a phase introducer by providing constant phase increments for static tissue and quadratic phase increments for moving magnetizations (signal spoiling). Although perivascular fat is not expected around the intracranial arteries, the fat saturation module can help suppress the normally hyperintense fat signal from the skull, thereby avoiding the dynamic range compression from fat signals on digitized images. The acquisition sequence can be any MRI technique that fits the clinical application. In this study, both TSE and VISTA (20) sequences were evaluated because of their relatively lower sensitivity to susceptibility caused signal dropout compared with gradient echo based acquisition techniques. The DANTE pulse was optimized to achieve the highest contrast between the intracranial vessel wall and the surrounding CSF. By using the previously established DANTE signal model (18), signal evolution for both vessel wall and CSF can be simulated. This simulation assumes: the same unity initial magnetization (M0) for both vessel wall and CSF; complete magnetization spoiling in flowing spins (18); and, partial phase coherence among non-flowing spins (18). Longitudinal magnetizations were simulated at the temporal resolution of the delay time (TD). Based on our previous measurements on the same scanner, T1/T2 relaxations times for CSF and vessel wall were assumed to be 3120/160 ms and 1114/55 ms. All simulations were conducted in Matlab (Mathworks R2013a, Natick, MA), using custom designed simulation programs. Considering only the signal evolution in the DANTE pulse train, two key parameters (TD and flip angle of the elementary sequence in the DANTE train) were jointly optimized to achieve the highest signal difference (as measured by normalized magnetization Mz/M0) between vessel wall (MVW) and CSF (MCSF). TD ranges of 1–8 ms and FA ranges of 1 –50 were tendered for optimization. No TD of less than 1 ms was simulated because the minimum TD achievable on our scanner was 0.86 ms due to hardware limitations. Due to the overall small flip angles used in the DANTE prepulses, it usually takes a long time before the signal reaches its steady state, therefore, the transient signal patterns in each TD in the DANTE pulse sequence were calculated to ensure accurate simulation. For the same reason, a very long RF pulse train (over 500 ms for certain parameter sets) is required to achieve the steady state in the DANTE pulse, which can significantly prolong the scan time, making clinical implementation impractical. To avoid this, a total DANTE pulse duration of no more than 256 ms (rather than use the time required for the actual steady state to be achieved) was used as a cut-off point in this simulation. For each TD/FA parameter pair, the highest MVW and MCSF difference achieved within 256 ms was used as the optimal value for this pair. Among all the TD/FA pairs, the one that provided the highest MVW and MCSF difference was used in the following experiments. As optimal parameters were identified, vessel wall and CSF
JBC Suppression for Intracranial Wall MRI
signal changes were plotted to evaluate the vessel wall signal drop at that parameter setting. Patient Recruitment Three healthy volunteers and six patients with diagnosed or suspected intracranial artery disease, based on clinical angiography, were recruited for this study. Local institutional research board (IRB) approved the study and informed consent was obtained from all subjects. The images collected from the six patients were used to evaluate the CSF suppression efficiency provided by JBC imaging; the data from the healthy volunteers were used to evaluate the compatibility of JBC with multicontrast imaging. MR Scans All MR scans were conducted on a 3 Tesla (T) whole body scanner (Philips Achieva R3.21, Best, the Netherlands) using an eight-channel brain coil from the same vendor. After the scout scan and a whole-brain MRA used for localizing the intracranial arteries, the following JBC scans were performed. For CSF suppression efficiency evaluation, PDweighted VISTA scans with and without the DANTE prepulse were performed. The two scans used identical parameters with the exception of the DANTE prepulse, which used the optimized parameter sets identified in the previous section. The detailed VISTA scan parameters were configured similarly to the previous report (15), as follows: TR/TE 2000/19 ms, FOV 210 180 48 mm3, voxel size 0.6 0.6 0.6 mm3, echo train length 38, SENSE factor 2, and axial orientation. The total scan times were the same for both scans at 5 min 29 s. For compatibility with multicontrast imaging evaluation, standard T2-weighted TSE sequences with and without DANTE pulses were acquired. The scanning parameters were: repetition time/echo time (TR/TE) 4000/ 80 ms, field of view (FOV) 220 220 mm2, voxel size 0.6 0.6 mm2, slice thickness 4 mm, echo train length 12, and axial orientation. The scan time for 16 slices was 3 min 40 s. After inclusion of DANTE prepulses, the imaging time remained the same for both VISTA and T2 scans. In Vivo CSF Suppression Efficiency Evaluation To evaluate the CSF suppression efficiency in JBC imaging, the ratio between vessel wall and the neighboring CSF was used as a surrogate. For both VISTA and DANTE VISTA datasets, cross-sectional images of the intracranial vessel wall were first reformatted to evaluate three key arteries of the intracranial vascular tree: basilar, left and right middle cerebral arteries (MCAs; M1 segment). After registration, 5 consecutive slices from each segment (15 total for each subject) were selected from the DANTE VISTA images. When possible, intracranial artery portions with close contact to brain parenchyma were avoided to ensure the clear visualization of outer wall boundaries. On the cross-sectional images, inner and outer wall boundaries of the intracranial arteries were delineated and a neighboring CSF region was selected for signal measurements, using custom-
833
made plaque analysis software, CASCADE (21). As the two VISTA image sets were registered, the contours from DANTE VISTA were automatically transferred to the original VISTA image for measurements. The signal ratio is defined as, R ¼ SVW =SCSF
[1]
where SVW is the signal measured on the vessel wall and SCSF is the signal measured on the CSF. In this measurement, a higher R is desirable as it indicates a more efficient CSF signal suppression. To facilitate comparison of the results, CSF SNR and wall-CSF CNR were also measured. The inner and outer wall boundaries delineated in the above analysis were used in this comparison. The signal standard deviation, measured in the nasal cavity, was defined as the image noise. SNR and CNR were defined below, SNR ¼
S ; CNRwall-CSF ¼ SNRwall -SNRCSF snoise
[2]
where, S is the average signal of the region of interest, s is the standard deviation of the noise region. In this evaluation, a lower CSF SNR was desirable as it represents more efficient CSF suppression; a higher wall-CSF CNR was desirable as it means better wall characterization. In Vivo Flow Suppression Evaluation In addition to CSF suppression, blood suppression is crucial in quantifying atherosclerotic plaque burden for intracranial arteries. To evaluate JBC’s performance on blood suppression in intracranial arteries, lumen SNR and walllumen CNR were evaluated. The same SNR and CNR calculation method was used as described in the CSF suppression efficiency evaluation section. In this evaluation, a lower lumen SNR was preferable, as it results in a complete blood signal suppression, while a higher wall-lumen CNR means better vessel wall characterization. Image Contrast Change Evaluation As explained before, a desirable blood/CSF suppression module should be compatible with multicontrast imaging acquisitions so that established plaque component imaging techniques can be used. With this requirement, static tissue’s relative signal change on a standard contrast weighting image needed evaluation. A numerical simulation was first conducted to evaluate the static tissue signal change after the application of DANTE (as measured by the signal ratio between images with and without the DANTE prepulse). A wide T1 range of 200–1500 ms was used to cover most static tissue T1 relaxation times at 3T. The same simulation program used in the optimization section was performed. The signal ratios were plotted against the simulated T1 values. As the CSF signal can present hypo-, iso-, and hyperintensive signal patterns on in vivo T1w, PDw, and T2w images, evaluating JBC on T2w images represented the highest relevance. To rule out the impact from image scaling between different acquisitions, the average signal on the brain stem region was used as the reference base
834
Wang et al.
FIG. 2. DANTE PD VISTA allows for more effective intracranial artery wall delineation by suppressing CSF signal (a,c,e), as compared to images without DANTE prepulse (b,d,f). Arterial walls can be better visualized in MCA left (a,b), MCA right (c,d), and basilar segments (e,f), indicated by the arrows. A clear wall thickening can be visualized on the DANTE-VISTA image (a), indicated by the dotted arrow, but its appearance is overshadowed by the nearby hyperintense CSF flow artifact on VISTA image (dotted arrow, b).
for both DANTE and regular T2w TSE images. The brain stem was selected as the reference due to its size and a relatively homogenous signal on T2-weighted images. Normalized DANTE and regular TSE images were obtained after dividing the tissue signal by the reference signal (average signal on brain stem) for each pixel. To facilitate the evaluation of minor signal changes, false color images were also rendered using Matlab’s standard “Jet” colormap. Statistical Analysis All statistical comparisons were conducted using Excel software (Microsoft Corp, Redmond, WA). A paired Student’s t-test was used to compare the R difference between DANTE VISTA and VISTA images; unpaired unequal-variance t-tests were used to evaluate the R difference among different arteries. Paired Student’s t-test was used to compare all SNR and CNR measurements between DANTE VISTA and VISTA images. A P-value of equal or less than 0.05 was considered significant. RESULTS Pulse Sequence Optimization The numerical simulation indicated that the contrast between vessel wall and CSF signal rapidly peaks and then gradually decreases as the FA increases for all TD
values. Overall a shorter TD promised better contrast while given the same total duration of the DANTE prepulse train. As shown in Figure 1b, maximal contrast was achieved when TD was 1 ms and FA was 12 , while the total number of TD was 256. Because primary flow directions in basilar and MCA arteries are foot–head and left–right directions, gradients in both directions were turned on in the DANTE prepulse. With our hardware settings, the gradient parameters used in DANTE were: gradient strength 30 mT/m and gradient duration 0.72 ms. At these parameter settings, vessel wall and CSF signal changes at different flip angles were also plotted (Fig. 1c). As indicated by the arrows, optimized DANTE parameters caused only 3.6% signal loss on the vessel wall. These parameters were used as the optimal parameters in the following in vivo experiments. In Vivo CSF Suppression Efficiency Evaluation A total of 90 locations were obtained from six patients— five slices from each major artery (basilar, left MCA and right MCA) of each subject. Qualitatively, DANTE prepared PD VISTA sequence was found to provide improved intracranial artery wall delineation by effectively suppressing CSF signal, as compared to the sequence without DANTE prepulse (Fig. 2). In both MCA and basilar arteries, outer wall boundaries were better
JBC Suppression for Intracranial Wall MRI
835
Suppression efficiency by JBC at different vascular beds showed more efficiency on basilar arteries, as compared to MCAs. However, there was no statistical difference among different pairs of vascular beds (P > 0.05 for all). CSF SNR and wall-CSF CNR comparisons yielded that DANTE-VISTA presented significantly lower CSF SNR and higher wall-CSF CNR in all vascular beds (P < 0.001 for all), as shown in Table 1. This result confirmed that DANTE-VISTA provided effective CSF suppression and improved the characterization of intracranial vessel walls. FIG. 3. Signal ratio comparison between DANTE and non-DANTE images. DANTE provides a significantly higher signal ratio across all arteries (P < 0.001). The error bar represents the standard error of each measurement. *Signal ratio is significantly lower than the corresponding JBC image (P < 0.001)
delineated on DANTE VISTA images (Fig. 2a,c,e), as compared to regular VISTA images (Fig. 2b,d,f). Apparent wall thickening was visualized on the DANTEVISTA (Fig. 2a), as indicated by the dotted arrow, but its appearance was overshadowed by a nearby hyperintense CSF flow artifact on the VISTA image (Figure 2b, dotted arrow). Flow artifacts were also present in the same region. Quantitatively, a significantly higher signal ratio between vessel wall and CSF was obtained on DANTE prepared images in all major vascular beds (Fig. 3). On average, the vessel wall signal was 35.4% higher than the neighboring CSF region on DANTE VISTA images (R ¼ 1.354 6 0.216), while only 5.7% higher on regular VISTA images (R ¼ 1.057 6 0.13). This difference was statistically significant (P < 0.001). The same pattern of R maintains when each individual artery bed was examined, as detailed in Figure 3.
In Vivo Flow Suppression Evaluation DANTE prepared VISTA was also found to help improve the lumen boundary definition by further suppressing blood signal. As shown in Table 1, intracranial artery luminal signals on JBC images were significantly lower than those of regular VISTA images (DANTE-VISTA 7.87 6 3.03 versus VISTA 9.69 6 3.77; P < 0.001), indicating a more complete blood suppression because of the DANTE magnetization preparation. At the same time, wall-lumen CNR was significantly higher on DANTE images (DANTE-VISTA 10.55 6 3.79 versus VISTA 9.34 6 3.54; P < 0.001), indicating a more favorable lumen-wall contrast. It worth mentioning that the wall SNR of DANTE images is slightly (3.2%) lower than that of VISTA images, which was not found to be statistically significant (DANTE-VISTA 18.43 6 6.44 versus VISTA 19.03 6 6.22; P ¼ 0.339). The difference was comparable to the predicted simulation (3.6% in Figure 1c). Multicontrast Compatibility Evaluation The simulation (Sup. Fig. S1, which is available online) found that DANTE makes nearly homogeneous impact on tissues with a wide range of T1. The maximal
Table 1 Blood and CSF suppression efficiency comparisons between DANTE VISTA and regular VISTA images DANTE VISTA Basilar
MCA Left
MCA right
All
Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR Wall SNR Lumen SNR Wall-Lumen CNR CSF SNR Wall-CSF CNR
16.59 6.76 9.82 12.17 4.43 19.44 8.13 11.31 14.26 5.18 19.25 8.71 10.53 14.98 4.27 18.43 7.87 10.55 13.80 4.62
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6.82 3.71 3.35 5.41 2.65 7.66 2.84 5.01 4.30 4.37 4.12 2.06 2.57 3.59 2.15 6.44 3.03 3.79 4.61 3.19
VISTA 18.27 9.01 9.25 16.59 1.67 18.83 9.49 9.34 18.03 0.80 19.99 10.56 9.43 20.13 0.14 19.03 9.69 9.34 18.25 0.78
6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
6.12 3.31 3.92 5.69 2.64 5.84 3.72 3.17 5.98 1.92 6.75 4.19 3.62 6.61 1.97 6.22 3.77 3.54 6.21 2.30
P-value 0.077 0.002 0.307
