NOTE Magnetic Resonance in Medicine 74:1103–1109 (2015)
Fast Spin Echo Imaging of Carotid Artery Dynamics Mari E. Boesen,1,2,3 Luis A. Souto Maior Neto,3 Alexandra Pulwicki,3 Jerome Yerly,2,3,4,5,6 R. Marc Lebel,3,7,8 and Richard Frayne1,2,3,4,7,9* Purpose: We propose the use of a retrospectively gated cine fast spin echo (FSE) sequence for characterization of carotid artery dynamics. The aim of this study was to compare cine FSE measures of carotid dynamics with measures obtained on prospectively gated FSE images. Methods: The common carotid arteries in 10 volunteers were imaged using two temporally resolved sequences: (i) cine FSE and (ii) prospectively gated FSE. Three raters manually traced a common carotid artery area for all cardiac phases on both sequences. Measured areas and systolic–diastolic area changes were calculated and compared. Inter- and intra-rater reliability were assessed for both sequences. Results: No significant difference between cine FSE and prospectively gated FSE areas were observed (P ¼ 0.36). Both sequences produced repeatable cross-sectional area measurements: inter-rater intraclass correlation coefficient (ICC) ¼ 0.88 on cine FSE images and 0.87 on prospectively gated FSE images. Minimum detectable difference (MDD) in systolic–diastolic area was 4.9 mm2 with cine FSE and 6.4 mm2 with prospectively gated FSE. Conclusion: This cine FSE method produced repeatable dynamic carotid artery measurements with less artifact and greater temporal efficiency compared with prospectively gated C 2014 Wiley FSE. Magn Reson Med 74:1103–1109, 2015. V Periodicals, Inc. Key words: carotid; distensibility; atherosclerosis; cine; cardiac gated; fast spin echo; FSE; cine FSE; black-blood; compressed sensing; constrained reconstruction
INTRODUCTION Atherosclerosis is characterized by a degenerative accumulation of lipid and other materials in the arterial walls and commonly affects vessels with disturbed blood flow, particularly in regions of change in vessel 1
Physics and Astronomy, University of Calgary, Calgary, Canada. Hotchkiss Brain Institute, University of Calgary, Calgary, Canada. 3 Seaman Family Centre, Foothills Medical Centre, AB Health Services, Calgary, Canada. 4 Electrical and Computer Engineering, University of Calgary, Calgary, Canada. 5 Radiology, University Hospital and University of Lausanne, Lausanne, Switzerland. 6 Centre for Biomedical Imaging, Lausanne, Switzerland. 7 Radiology, University of Calgary, Calgary, Canada. 8 General Electric Healthcare, Calgary, Calgary, Canada. 9 Clinical Neurosciences, University of Calgary, Calgary, Canada. 2
*Correspondence to: Richard Frayne, Ph.D., Seaman Family MR Research Centre, Foothills Medical Centre, 1403 29th Street NW, Calgary, Alberta T2N 2T9. E-mail: [email protected] Received 14 March 2014; revised 20 September 2014; accepted 22 September 2014 DOI 10.1002/mrm.25494 Published online 13 October 2014 in Wiley Online Library (wileyonlinelibrary. com). C 2014 Wiley Periodicals, Inc. V
diameter or direction (1). The carotid artery bifurcation is especially susceptible to the development of atherosclerotic lesions. Atherosclerosis increases the risk of stroke and is a major contributor to permanent disability and death, causing approximately 14% of all strokerelated deaths (2). Efficient imaging-based morphological and lesion composition assessments of atherosclerotic plaque may provide additional insight into the stroke risk posed by individual lesions (3–5). Additionally, an increase in vessel stiffness (or a decrease in vessel distensibility) is associated with both advancing age (6) and lesion burden (7), and can potentially provide further biomechanically based insight into impaired endothelial function and lesion rupture risk. The pulsatile nature of blood flow is modified by the major arteries; vessel distension plays an important role as it acts to (i) efficiently convert pulsatile flow originating at the heart to near constant flow in end organ tissues such as the brain (8), and (ii) increase diastolic flow rates (9). Changes in wall distention have been previously postulated as a cumulative index of stroke and cardiovascular risk. Ultrasound-based techniques predominate in determining carotid distensibility (10–12) mainly because of their high temporal resolution, ease of use, and relative low cost; however, these methods suffer from poor blood-wall contrast and image shadowing due to calcifications, and are difficult to use because the acquired vessel image is typically longitudinal rather than crosssectional. Bright-blood (gradient-recalled echo) cine magnetic resonance (MR) techniques (6) have been used to obtain distensibility estimates, although they produce vessel wall contrasts inferior to those produced by blackblood fast spin echo (FSE) based approaches (4,5). Prospectively gated black-blood methods (13) have been implemented to improve depiction of vessel wall composition and have improved blood-wall boundary depiction, but require a significant scan time if vessel motion is to be characterized over the entire cardiac cycle. An MR imaging approach that retains the image contrast of FSE and efficiently acquires images over the entire cardiac cycle is needed. There has been recent interest in cardiac phase resolved black-blood imaging of the carotid arteries (14,15). We further developed a cine FSE imaging approach that was first implemented by Mendes et al (16) and assessed its suitability for characterizing the common carotid artery over the cardiac cycle by comparing it to a prospectively gated FSE method. Cine FSE images acquire multiple cardiac phase images in roughly the same scan time required for a multiple signal average black-blood FSE technique that is currently used for wall composition assessment (4,5). With prospectively gated FSE approaches, only a single cardiac phase may be
1103
1104
acquired at a time. Furthermore, cardiac trigger delays must be estimated beforehand and therefore may not accurately capture the desired cardiac phase. In approximately the same amount of scan time as a single prospectively gated image, cine FSE generates multiple images across the entire cardiac cycle, enabling a later selection of representative systolic and diastolic phase images, both reducing scan time and potentially increasing accuracy. As this technique is retrospectively correlated to the cardiac cycle, it is compatible with the majority of imaging options available in standard FSE applications, such as multislice acquisition, fat saturation, inversion recovery preparation, and variable image contrast weightings. The purpose of this study was to assess the suitability of cine FSE for dynamic carotid imaging. We undertook a comparison of the retrospectively gated cine FSE technique against the more time-consuming prospectively gated FSE technique. We hypothesized that manually measured common carotid areas would not differ between prospectively gated and retrospectively gated cine FSE images of the same cardiac phases. We assessed, in healthy volunteers, the manually measured cross-sectional area of the carotid artery lumen at four prospectively gated cardiac phases and the corresponding four phases of a cine FSE acquisition. In addition, the systolic areas (As), diastolic areas (Ad), and the systolic-diastolic vessel distensions (DA ¼ As – Ad) were evaluated for both cine FSE and prospectively gated FSE. The inter- and intra-rater reliability and minimum detectable difference of repeated manual measurements of carotid area for both types of images were calculated.
Boesen et al.
odes are acquired during all cardiac phases. This variable density sampling is known to improve the results of constrained reconstructions in MR imaging (17). After the k-space data have been retrospectively sorted into cardiac phases, an appropriate parallel imaging reconstruction algorithm estimates any data not sampled in a specified cardiac phase. Given the cyclic nature of cardiac-gated cine FSE data, we have chosen to use a SENSE (18) reconstruction that takes advantage of the similarities between adjacent cardiac phases and of the coil sensitivity profile. We performed a nonlinear conjugate gradient minimization of the following temporally constrained sparse SENSE equation (19): min m kFSmyk22 þ l1 kTVt mk1 þ l2 kTVxy mk1
[1]
where y is the undersampled k-space coil data, m is the reconstructed coil-combined image, F is the undersampled Fourier transform operator, S is the coil sensitivity matrix, and l1 and l2 are regularization weighting parameters for the temporal and spatial constraint terms TVt and TVxy, respectively. The TVxy term is a penalty that promotes sparse differences in the spatial domain (20) while the TVt term acts as a high pass filter in the temporal direction (i.e., over cardiac phases). Our reconstructions, while similarly taking advantage of the inherent properties of the cine FSE dataset, differ from the original method of Mendes et al (16) in that they chose to overlap cardiac phase definitions to decrease acceleration factors and promote smooth variations over time by constraining their SENSE reconstruction with a temporal gradient. We opted for l1-norm regularization terms that promote sparse rather than smooth differences, thereby preserving sharp edges that may be present in both the spatial and temporal dimensions.
METHODS Imaging Method
Data Acquisition
Cine FSE imaging (16) uses an approach similar to a conventional multiple signal average FSE (4,5) and acquires MR k-space data asynchronously with the cardiac cycle. By recording the cardiac cycle during data acquisition, these data may be retrospectively converted into cardiacphase resolved images by grouping the k-space data into temporal bins of specified cardiac phases. This results in a series of undersampled k-space datasets, each representing a single cardiac phase; these datasets are then reconstructed using an appropriately constrained reconstruction algorithm. The data acquisition and image reconstruction processes are outlined in Figure 1. Unlike conventional FSE, cine FSE uses increased sampling density at the center of k-space and sparse, pseudo-random sampling in the periphery of k-space. Like conventional multiple signal average FSE, the total number of phase encode lines acquired is equal to the number of signal averages multiplied by the echo train length; however, in our implementation of cine FSE, the edges of k-space are sampled only once while the central lines are acquired as many as 20 times. The cine FSE sampling scheme used in this work is depicted in the online Supporting Figure S1. The intent of this sampling scheme is to ensure that most central k-space phase enc-
We scanned 10 volunteers (5 male/5 female, mean age of 21.1 6 2.2 years) with no history of cardiovascular disease. All subjects provided written informed consent before imaging using a protocol approved by our local Research Ethics Board. Imaging data were acquired on a 3 Tesla (T) MR scanner (Discovery 750; General Electric Healthcare, Waukesha, WI) using a commercially available 16-channel receive-only neurovascular coil (General Electric Healthcare). A pulse oximeter was placed on the right index finger and used to trigger and record the cardiac cycle during MR data acquisition. Additional purpose-written software recorded the time of acquisition of each FSE echo train relative to temporal position of the cardiac cycle. The common carotid arteries were imaged axially with (i) three repeated acquisitions of a cine FSE sequence and (ii) four prospectively gated FSE sequences acquired with varying, predetermined trigger delays. Key acquisition parameters are summarized in Supporting Table S1. Double inversion recovery blood suppression was applied on cine FSE acquisitions (inversion time ¼ 500 ms) but was not used for the prospectively gated FSE acquisitions due to the limitations it places on cardiac trigger delay selection. Also important to note is that to ensure pseudo-random sampling of all
FSE of Carotid Artery Dynamics
1105
FIG. 1. Schematic flow chart showing acquisition and reconstruction of cine FSE images. The cardiac signal is recorded during the fast spin echo acquisition of a nonuniformly sampled k-space (Supp. Fig. S1). The data are retrospectively gated into any number of cardiac phases (N) using a time-ordered phase encode table. The red lines represent data acquired during the first cardiac phase (1 of N) and the blue lines represent data acquired during the last defined cardiac phase (N of N). Each undersampled cardiac phase image was then jointly reconstructed using a temporally constrained sparse SENSE approach. In this work, the number of reconstructed cardiac phases was chosen as N ¼ 16.
cardiac phases, the cine FSE repetition time was set such that it was not a multiple of the cardiac cycle length. The cine FSE data reconstructed images at 16 evenly distributed phases across the cardiac cycle. The time points of the four prospectively gated FSE images were at trigger delays of 30 ms, 250 ms, 500 ms, and 700 ms. The on-line scanner software reconstructed the prospectively gated FSE images, while cine FSE images were reconstructed offline using purpose-written software (Matlab 2013a; Mathworks; Natick, MA) on a conventional workstation (running Mac OSX 10.7). Each cine FSE acquisition was reconstructed from its under-
sampled cardiac phase resolved k-space data by nonlinear conjugate gradient minimization of Eq. [1]. Regularization factors l1 and l2 were chosen empirically by evaluating all image sets for residual artifact, noise, and temporal motion of the carotid artery; they were chosen as l1 ¼ 0.1 and l2 ¼ 0.005 for all reconstructions. Image and Statistical Analysis Three raters with moderate image segmentation experience were asked to manually segment the left bloodvessel wall boundary on all images (16 cardiac phases
1106
Boesen et al.
FIG. 2. A single cardiac phase image of a cine fast spin echo (cine FSE) acquisition is shown (left) with a 1 cm line placed across the left common carotid artery of a healthy volunteer. Changes in lumen area across the line profile are shown for all 16 cardiac phases (systole to systole) of the same cine FSE acquisition (right).
per cine FSE acquisition and 4 prospectively gated cardiac phases) using image analysis software (Osirix; Geneva, Switzerland). In one subject, the right carotid artery was segmented as the selected image included the left carotid artery bifurcation. The datasets were presented to each rater in correct cardiac phase order but were randomized by volunteer and acquisition. To enable a direct comparison of measured crosssectional areas by technique, four cine FSE phases corresponding to the acquired prospectively gated FSE phases were identified by trigger time. A Bland-Altman analysis was used to compare the two methods (21). Additionally, systolic area (As) and diastolic area (Ad), taken as maximum and minimum cross-sectional areas, respectively, were determined for each rater and used to calculate the average systolic-to-diastolic area change (DA ¼ As Ad) in each subject. Paired t-tests were used
to assess the significance of any differences between sequences. To evaluate inter- and intra-rater reliability, all raters repeated their measurements three times on separate days. Inter- and intra-rater reliability coefficients were concurrently calculated by means of the co-variance-tovariance ratio method (22). This statistical approach accounts for differences in measurements between raters, as well as for the variation within each individual rater. Inter- and intra-rater intraclass correlation coefficients (ICC) were calculated. The inter- and intrarater standard errors of measurement (SEM) were determined for repeated systolic-diastolic area change measurements. The minimum detectable difference pffiffiffi (MDD) in pcross-sectional area change ðMDD¼Za=2 2 ffiffiffi SEM¼1:96 2SEMÞ was calculated from both the interand intra-rater standard errors of measurement and
FIG. 3. Comparison of a systolic (top) and diastolic (bottom) phase through the left common carotid artery in a normal subject. Good qualitative agreement is seen between cine fast spin echo (FSE, left) and prospectively gated FSE (right), with the exception of the artifactual signal inside the vessel lumen on prospectively gated FSE. Differences in fat signal are due to enhanced T1-weighting in the cine FSE images.
FSE of Carotid Artery Dynamics
1107
FIG. 4. a: Comparison of measured cross-sectional carotid areas on cine fast spin echo (FSE) and prospectively gated FSE images with cardiac phases matched by trigger delay. No significant difference in area measurement by technique was observed (P ¼ 0.36, paired ttest). b: Mean cross-sectional carotid area versus difference in measurements by technique (Bland-Altman plot). The mean difference was 0.69 6 4.7 mm2 and 95% confidence interval (CI) was 9.9 mm2 to þ8.5 mm2.
represents the minimum change in systolic-diastolic area that can be reliably detected at 95% confidence (22). Finally, Pearson’s correlation coefficient was used to assess the scan–rescan reliability of the cine FSE method by assessing all raters’ measures on three repeated cine FSE acquisitions. A critical value a ¼ 0.05 Table 1 Summary of Measured Carotid Areas, Area Changes, and Distensibility Coefficients Obtained in 10 Young, Normal Volunteers (19– 25 years)a
Systolic area, As (mm2) Diastolic area, Ad (mm2) Systolic-diastolic area change, DA (mm2) Mean area (mm2)b Inter-rater ICC Rater 1 Rater 2 Rater 3 Inter-rater SEM Rater 1 Rater 2 Rater 3 Inter-rater MDD Rater 1 Rater 2 Rater 3
Cine FSE (mean 6 SD)
Prospectively Gated FSE (mean 6 SD)
31.7 6 6.2
31.9 6 6.6
P ¼ 0.87
25.7 6 5.6
23.7 6 4.7
P ¼ 0.13
6.1 6 1.7
8.2 6 2.8
P ¼ 0.01
28.5 6 5.8 0.88 0.94 0.86 0.89 DA (mm2) 1.8 0.9 2.9 1.1 4.9 2.6 8.0 3.1
0.87 0.95 0.79 0.90 DA (mm2) 2.3 1.6 2.5 2.9 6.4 4.3 6.8 8.0
-
Inter- and intra-rater intraclass correlation coefficients (ICC) of repeated area measurements on all images are reported. Inter- and intra-rater standard errors of measurement (SEM) and minimum detectable difference (MDD) in repeated measures of systolic– diastolic area change (DA) are summarized. a Values are mean 6 standard deviation (SD) over all subjects. b Averaged over the cardiac cycle.
was used in this study. All data are reported as mean6 standard deviation (SD). RESULTS Raters were able to identify and manually outline the left common carotid artery in all subjects except one, whose positioning favored tracing the right carotid artery. The 16-phase cine FSE image series consistently depicted clear changes in carotid area over the cardiac cycle. Figure 2 shows the change in cross-sectional area of the left common carotid lumen over all cardiac phases in a representative cine FSE acquisition. The cine FSE images showed less flow artifact than the prospectively gated FSE images, especially after systolic phases when blood flow can reverse direction in the region of the carotid bifurcation (Fig. 3). A comparison of four cardiac phase matched images on each subject showed no significant difference between measured cross-sectional areas on cine FSE and prospectively gated FSE (P ¼ 0.36; Figure 4a). Figure 4b shows the Bland-Altman plot of mean values versus difference in measures. The mean difference was 0.69 6 4.7 mm2 and limits of agreement were 9.9 mm2 and þ8.5 mm2. Mean cross-sectional area over all subjects was 28.5 6 5.8 mm2 on the cine FSE images. Note that mean cross-sectional area was not calculated for the prospectively gated data because images were unevenly sampled across the cardiac cycle. Table 1 summarizes key carotid artery measurements: systolic area was 31.7 6 6.2 mm2 on cine FSE and 31.9 6 6.6 mm2 on prospectively gated FSE (P ¼ 0.87) and diastolic area was 25.7 6 5.6 mm2 on cine FSE and 23.7 6 4.7 mm2 on prospectively gated FSE (P ¼ 0.13). Measures of systolic-diastolic area change on cine FSE and prospectively gated FSE image series were significantly different (P ¼ 0.01). Table 1 summarizes the results of the inter- and intra-rater reliability assessment. For the cine FSE sequence, intra-rater reliability varied from an ICC of 0.86 to 0.94 across the three raters. The inter-rater reliabilities of the sequences were 0.88
1108
on cine FSE and 0.87 on prospectively gated FSE. The interrater minimum detectable difference in systolic-diastolic lumen area change was MDDinter ¼ 4.9 mm2 for the cine FSE approach and 6.4 mm2 for the prospectively gated FSE approach. Intra-rater MDD ranged from MDDintra,1 ¼ 2.6 mm2 to MDDintra,2 ¼ 8.0 mm2 on cine FSE and MDDintra,1 ¼ 4.3 mm2 to MDDintra,3 ¼ 8.0 mm2 on prospectively gated FSE. Scan–rescan reliability of the repeated cine FSE acquisitions had a Pearson’s correlation coefficient of r ¼ 0.898 (P < 0.01). DISCUSSION AND CONCLUSIONS The cine FSE sequence successfully imaged the common carotid arteries in all volunteers and produced mean cross-sectional areas (28.5 mm2 6 5.8 mm2; Table 1) within the confidence interval of previously reported common carotid measurements in a similar young population (29.2 6 0.2 mm2, calculated from reported diameter) (23). No significant difference was found in area measurements on cine FSE and prospectively gated FSE images. While both sequences were black-blood FSEbased techniques, there was noticeably reduced image artifact observed in the vessel lumen of the cine FSE data (Fig. 3). The observed suppression of artifact in the cine FSE images is attributed to the inversion recovery preparation and nonuniform sampling scheme (24), which resulted in easier manual segmentation of the vessel wall-blood boundary and, we hypothesize, more accurate lumen area estimates. Manual measurements on both types of images were highly repeatable, as evidenced by the inter-rater ICCs (0.88 cine FSE and 0.87 prospectively gated FSE). The average systolic–diastolic change in crosssectional lumen area in our group was 6.1 6 1.7 mm2 (cine FSE) and 8.2 6 2.8 mm2 (prospectively gated FSE). Differences in measured distension arise predominantly in diastolic area measurements (Table 1), where flow artifact, primarily in prospectively gated FSE images, can obscure the lumen-wall boundary. In addition, the preselected peripheral trigger delays of the prospectively gated technique may not correspond exactly to true systole and diastole. The cine FSE method, which collects data continuously through all cardiac phases, ensures that the entire cardiac cycle is sampled. The intra-rater MDDs indicate that our best rater is capable of reliably detecting area changes greater than 2.6 mm2 on cine FSE and 4.3 mm2 on prospectively gated FSE, both well below the observed distension in these healthy subjects. Ultrasound studies suggest 50% decreases in internal carotid artery distension with the presence of atherosclerosis (25), which translates to an expected systolic–diastolic area change of only 3 to 4 mm2, suggesting that MDD
Fast Spin Echo Imaging of Carotid Artery Dynamics Mari E. Boesen,1,2,3 Luis A. Souto Maior Neto,3 Alexandra Pulwicki,3 Jerome Yerly,2,3,4,5,6 R. Marc Lebel,3,7,8 and Richard Frayne1,2,3,4,7,9* Purpose: We propose the use of a retrospectively gated cine fast spin echo (FSE) sequence for characterization of carotid artery dynamics. The aim of this study was to compare cine FSE measures of carotid dynamics with measures obtained on prospectively gated FSE images. Methods: The common carotid arteries in 10 volunteers were imaged using two temporally resolved sequences: (i) cine FSE and (ii) prospectively gated FSE. Three raters manually traced a common carotid artery area for all cardiac phases on both sequences. Measured areas and systolic–diastolic area changes were calculated and compared. Inter- and intra-rater reliability were assessed for both sequences. Results: No significant difference between cine FSE and prospectively gated FSE areas were observed (P ¼ 0.36). Both sequences produced repeatable cross-sectional area measurements: inter-rater intraclass correlation coefficient (ICC) ¼ 0.88 on cine FSE images and 0.87 on prospectively gated FSE images. Minimum detectable difference (MDD) in systolic–diastolic area was 4.9 mm2 with cine FSE and 6.4 mm2 with prospectively gated FSE. Conclusion: This cine FSE method produced repeatable dynamic carotid artery measurements with less artifact and greater temporal efficiency compared with prospectively gated C 2014 Wiley FSE. Magn Reson Med 74:1103–1109, 2015. V Periodicals, Inc. Key words: carotid; distensibility; atherosclerosis; cine; cardiac gated; fast spin echo; FSE; cine FSE; black-blood; compressed sensing; constrained reconstruction
INTRODUCTION Atherosclerosis is characterized by a degenerative accumulation of lipid and other materials in the arterial walls and commonly affects vessels with disturbed blood flow, particularly in regions of change in vessel 1
Physics and Astronomy, University of Calgary, Calgary, Canada. Hotchkiss Brain Institute, University of Calgary, Calgary, Canada. 3 Seaman Family Centre, Foothills Medical Centre, AB Health Services, Calgary, Canada. 4 Electrical and Computer Engineering, University of Calgary, Calgary, Canada. 5 Radiology, University Hospital and University of Lausanne, Lausanne, Switzerland. 6 Centre for Biomedical Imaging, Lausanne, Switzerland. 7 Radiology, University of Calgary, Calgary, Canada. 8 General Electric Healthcare, Calgary, Calgary, Canada. 9 Clinical Neurosciences, University of Calgary, Calgary, Canada. 2
*Correspondence to: Richard Frayne, Ph.D., Seaman Family MR Research Centre, Foothills Medical Centre, 1403 29th Street NW, Calgary, Alberta T2N 2T9. E-mail: [email protected] Received 14 March 2014; revised 20 September 2014; accepted 22 September 2014 DOI 10.1002/mrm.25494 Published online 13 October 2014 in Wiley Online Library (wileyonlinelibrary. com). C 2014 Wiley Periodicals, Inc. V
diameter or direction (1). The carotid artery bifurcation is especially susceptible to the development of atherosclerotic lesions. Atherosclerosis increases the risk of stroke and is a major contributor to permanent disability and death, causing approximately 14% of all strokerelated deaths (2). Efficient imaging-based morphological and lesion composition assessments of atherosclerotic plaque may provide additional insight into the stroke risk posed by individual lesions (3–5). Additionally, an increase in vessel stiffness (or a decrease in vessel distensibility) is associated with both advancing age (6) and lesion burden (7), and can potentially provide further biomechanically based insight into impaired endothelial function and lesion rupture risk. The pulsatile nature of blood flow is modified by the major arteries; vessel distension plays an important role as it acts to (i) efficiently convert pulsatile flow originating at the heart to near constant flow in end organ tissues such as the brain (8), and (ii) increase diastolic flow rates (9). Changes in wall distention have been previously postulated as a cumulative index of stroke and cardiovascular risk. Ultrasound-based techniques predominate in determining carotid distensibility (10–12) mainly because of their high temporal resolution, ease of use, and relative low cost; however, these methods suffer from poor blood-wall contrast and image shadowing due to calcifications, and are difficult to use because the acquired vessel image is typically longitudinal rather than crosssectional. Bright-blood (gradient-recalled echo) cine magnetic resonance (MR) techniques (6) have been used to obtain distensibility estimates, although they produce vessel wall contrasts inferior to those produced by blackblood fast spin echo (FSE) based approaches (4,5). Prospectively gated black-blood methods (13) have been implemented to improve depiction of vessel wall composition and have improved blood-wall boundary depiction, but require a significant scan time if vessel motion is to be characterized over the entire cardiac cycle. An MR imaging approach that retains the image contrast of FSE and efficiently acquires images over the entire cardiac cycle is needed. There has been recent interest in cardiac phase resolved black-blood imaging of the carotid arteries (14,15). We further developed a cine FSE imaging approach that was first implemented by Mendes et al (16) and assessed its suitability for characterizing the common carotid artery over the cardiac cycle by comparing it to a prospectively gated FSE method. Cine FSE images acquire multiple cardiac phase images in roughly the same scan time required for a multiple signal average black-blood FSE technique that is currently used for wall composition assessment (4,5). With prospectively gated FSE approaches, only a single cardiac phase may be
1103
1104
acquired at a time. Furthermore, cardiac trigger delays must be estimated beforehand and therefore may not accurately capture the desired cardiac phase. In approximately the same amount of scan time as a single prospectively gated image, cine FSE generates multiple images across the entire cardiac cycle, enabling a later selection of representative systolic and diastolic phase images, both reducing scan time and potentially increasing accuracy. As this technique is retrospectively correlated to the cardiac cycle, it is compatible with the majority of imaging options available in standard FSE applications, such as multislice acquisition, fat saturation, inversion recovery preparation, and variable image contrast weightings. The purpose of this study was to assess the suitability of cine FSE for dynamic carotid imaging. We undertook a comparison of the retrospectively gated cine FSE technique against the more time-consuming prospectively gated FSE technique. We hypothesized that manually measured common carotid areas would not differ between prospectively gated and retrospectively gated cine FSE images of the same cardiac phases. We assessed, in healthy volunteers, the manually measured cross-sectional area of the carotid artery lumen at four prospectively gated cardiac phases and the corresponding four phases of a cine FSE acquisition. In addition, the systolic areas (As), diastolic areas (Ad), and the systolic-diastolic vessel distensions (DA ¼ As – Ad) were evaluated for both cine FSE and prospectively gated FSE. The inter- and intra-rater reliability and minimum detectable difference of repeated manual measurements of carotid area for both types of images were calculated.
Boesen et al.
odes are acquired during all cardiac phases. This variable density sampling is known to improve the results of constrained reconstructions in MR imaging (17). After the k-space data have been retrospectively sorted into cardiac phases, an appropriate parallel imaging reconstruction algorithm estimates any data not sampled in a specified cardiac phase. Given the cyclic nature of cardiac-gated cine FSE data, we have chosen to use a SENSE (18) reconstruction that takes advantage of the similarities between adjacent cardiac phases and of the coil sensitivity profile. We performed a nonlinear conjugate gradient minimization of the following temporally constrained sparse SENSE equation (19): min m kFSmyk22 þ l1 kTVt mk1 þ l2 kTVxy mk1
[1]
where y is the undersampled k-space coil data, m is the reconstructed coil-combined image, F is the undersampled Fourier transform operator, S is the coil sensitivity matrix, and l1 and l2 are regularization weighting parameters for the temporal and spatial constraint terms TVt and TVxy, respectively. The TVxy term is a penalty that promotes sparse differences in the spatial domain (20) while the TVt term acts as a high pass filter in the temporal direction (i.e., over cardiac phases). Our reconstructions, while similarly taking advantage of the inherent properties of the cine FSE dataset, differ from the original method of Mendes et al (16) in that they chose to overlap cardiac phase definitions to decrease acceleration factors and promote smooth variations over time by constraining their SENSE reconstruction with a temporal gradient. We opted for l1-norm regularization terms that promote sparse rather than smooth differences, thereby preserving sharp edges that may be present in both the spatial and temporal dimensions.
METHODS Imaging Method
Data Acquisition
Cine FSE imaging (16) uses an approach similar to a conventional multiple signal average FSE (4,5) and acquires MR k-space data asynchronously with the cardiac cycle. By recording the cardiac cycle during data acquisition, these data may be retrospectively converted into cardiacphase resolved images by grouping the k-space data into temporal bins of specified cardiac phases. This results in a series of undersampled k-space datasets, each representing a single cardiac phase; these datasets are then reconstructed using an appropriately constrained reconstruction algorithm. The data acquisition and image reconstruction processes are outlined in Figure 1. Unlike conventional FSE, cine FSE uses increased sampling density at the center of k-space and sparse, pseudo-random sampling in the periphery of k-space. Like conventional multiple signal average FSE, the total number of phase encode lines acquired is equal to the number of signal averages multiplied by the echo train length; however, in our implementation of cine FSE, the edges of k-space are sampled only once while the central lines are acquired as many as 20 times. The cine FSE sampling scheme used in this work is depicted in the online Supporting Figure S1. The intent of this sampling scheme is to ensure that most central k-space phase enc-
We scanned 10 volunteers (5 male/5 female, mean age of 21.1 6 2.2 years) with no history of cardiovascular disease. All subjects provided written informed consent before imaging using a protocol approved by our local Research Ethics Board. Imaging data were acquired on a 3 Tesla (T) MR scanner (Discovery 750; General Electric Healthcare, Waukesha, WI) using a commercially available 16-channel receive-only neurovascular coil (General Electric Healthcare). A pulse oximeter was placed on the right index finger and used to trigger and record the cardiac cycle during MR data acquisition. Additional purpose-written software recorded the time of acquisition of each FSE echo train relative to temporal position of the cardiac cycle. The common carotid arteries were imaged axially with (i) three repeated acquisitions of a cine FSE sequence and (ii) four prospectively gated FSE sequences acquired with varying, predetermined trigger delays. Key acquisition parameters are summarized in Supporting Table S1. Double inversion recovery blood suppression was applied on cine FSE acquisitions (inversion time ¼ 500 ms) but was not used for the prospectively gated FSE acquisitions due to the limitations it places on cardiac trigger delay selection. Also important to note is that to ensure pseudo-random sampling of all
FSE of Carotid Artery Dynamics
1105
FIG. 1. Schematic flow chart showing acquisition and reconstruction of cine FSE images. The cardiac signal is recorded during the fast spin echo acquisition of a nonuniformly sampled k-space (Supp. Fig. S1). The data are retrospectively gated into any number of cardiac phases (N) using a time-ordered phase encode table. The red lines represent data acquired during the first cardiac phase (1 of N) and the blue lines represent data acquired during the last defined cardiac phase (N of N). Each undersampled cardiac phase image was then jointly reconstructed using a temporally constrained sparse SENSE approach. In this work, the number of reconstructed cardiac phases was chosen as N ¼ 16.
cardiac phases, the cine FSE repetition time was set such that it was not a multiple of the cardiac cycle length. The cine FSE data reconstructed images at 16 evenly distributed phases across the cardiac cycle. The time points of the four prospectively gated FSE images were at trigger delays of 30 ms, 250 ms, 500 ms, and 700 ms. The on-line scanner software reconstructed the prospectively gated FSE images, while cine FSE images were reconstructed offline using purpose-written software (Matlab 2013a; Mathworks; Natick, MA) on a conventional workstation (running Mac OSX 10.7). Each cine FSE acquisition was reconstructed from its under-
sampled cardiac phase resolved k-space data by nonlinear conjugate gradient minimization of Eq. [1]. Regularization factors l1 and l2 were chosen empirically by evaluating all image sets for residual artifact, noise, and temporal motion of the carotid artery; they were chosen as l1 ¼ 0.1 and l2 ¼ 0.005 for all reconstructions. Image and Statistical Analysis Three raters with moderate image segmentation experience were asked to manually segment the left bloodvessel wall boundary on all images (16 cardiac phases
1106
Boesen et al.
FIG. 2. A single cardiac phase image of a cine fast spin echo (cine FSE) acquisition is shown (left) with a 1 cm line placed across the left common carotid artery of a healthy volunteer. Changes in lumen area across the line profile are shown for all 16 cardiac phases (systole to systole) of the same cine FSE acquisition (right).
per cine FSE acquisition and 4 prospectively gated cardiac phases) using image analysis software (Osirix; Geneva, Switzerland). In one subject, the right carotid artery was segmented as the selected image included the left carotid artery bifurcation. The datasets were presented to each rater in correct cardiac phase order but were randomized by volunteer and acquisition. To enable a direct comparison of measured crosssectional areas by technique, four cine FSE phases corresponding to the acquired prospectively gated FSE phases were identified by trigger time. A Bland-Altman analysis was used to compare the two methods (21). Additionally, systolic area (As) and diastolic area (Ad), taken as maximum and minimum cross-sectional areas, respectively, were determined for each rater and used to calculate the average systolic-to-diastolic area change (DA ¼ As Ad) in each subject. Paired t-tests were used
to assess the significance of any differences between sequences. To evaluate inter- and intra-rater reliability, all raters repeated their measurements three times on separate days. Inter- and intra-rater reliability coefficients were concurrently calculated by means of the co-variance-tovariance ratio method (22). This statistical approach accounts for differences in measurements between raters, as well as for the variation within each individual rater. Inter- and intra-rater intraclass correlation coefficients (ICC) were calculated. The inter- and intrarater standard errors of measurement (SEM) were determined for repeated systolic-diastolic area change measurements. The minimum detectable difference pffiffiffi (MDD) in pcross-sectional area change ðMDD¼Za=2 2 ffiffiffi SEM¼1:96 2SEMÞ was calculated from both the interand intra-rater standard errors of measurement and
FIG. 3. Comparison of a systolic (top) and diastolic (bottom) phase through the left common carotid artery in a normal subject. Good qualitative agreement is seen between cine fast spin echo (FSE, left) and prospectively gated FSE (right), with the exception of the artifactual signal inside the vessel lumen on prospectively gated FSE. Differences in fat signal are due to enhanced T1-weighting in the cine FSE images.
FSE of Carotid Artery Dynamics
1107
FIG. 4. a: Comparison of measured cross-sectional carotid areas on cine fast spin echo (FSE) and prospectively gated FSE images with cardiac phases matched by trigger delay. No significant difference in area measurement by technique was observed (P ¼ 0.36, paired ttest). b: Mean cross-sectional carotid area versus difference in measurements by technique (Bland-Altman plot). The mean difference was 0.69 6 4.7 mm2 and 95% confidence interval (CI) was 9.9 mm2 to þ8.5 mm2.
represents the minimum change in systolic-diastolic area that can be reliably detected at 95% confidence (22). Finally, Pearson’s correlation coefficient was used to assess the scan–rescan reliability of the cine FSE method by assessing all raters’ measures on three repeated cine FSE acquisitions. A critical value a ¼ 0.05 Table 1 Summary of Measured Carotid Areas, Area Changes, and Distensibility Coefficients Obtained in 10 Young, Normal Volunteers (19– 25 years)a
Systolic area, As (mm2) Diastolic area, Ad (mm2) Systolic-diastolic area change, DA (mm2) Mean area (mm2)b Inter-rater ICC Rater 1 Rater 2 Rater 3 Inter-rater SEM Rater 1 Rater 2 Rater 3 Inter-rater MDD Rater 1 Rater 2 Rater 3
Cine FSE (mean 6 SD)
Prospectively Gated FSE (mean 6 SD)
31.7 6 6.2
31.9 6 6.6
P ¼ 0.87
25.7 6 5.6
23.7 6 4.7
P ¼ 0.13
6.1 6 1.7
8.2 6 2.8
P ¼ 0.01
28.5 6 5.8 0.88 0.94 0.86 0.89 DA (mm2) 1.8 0.9 2.9 1.1 4.9 2.6 8.0 3.1
0.87 0.95 0.79 0.90 DA (mm2) 2.3 1.6 2.5 2.9 6.4 4.3 6.8 8.0
-
Inter- and intra-rater intraclass correlation coefficients (ICC) of repeated area measurements on all images are reported. Inter- and intra-rater standard errors of measurement (SEM) and minimum detectable difference (MDD) in repeated measures of systolic– diastolic area change (DA) are summarized. a Values are mean 6 standard deviation (SD) over all subjects. b Averaged over the cardiac cycle.
was used in this study. All data are reported as mean6 standard deviation (SD). RESULTS Raters were able to identify and manually outline the left common carotid artery in all subjects except one, whose positioning favored tracing the right carotid artery. The 16-phase cine FSE image series consistently depicted clear changes in carotid area over the cardiac cycle. Figure 2 shows the change in cross-sectional area of the left common carotid lumen over all cardiac phases in a representative cine FSE acquisition. The cine FSE images showed less flow artifact than the prospectively gated FSE images, especially after systolic phases when blood flow can reverse direction in the region of the carotid bifurcation (Fig. 3). A comparison of four cardiac phase matched images on each subject showed no significant difference between measured cross-sectional areas on cine FSE and prospectively gated FSE (P ¼ 0.36; Figure 4a). Figure 4b shows the Bland-Altman plot of mean values versus difference in measures. The mean difference was 0.69 6 4.7 mm2 and limits of agreement were 9.9 mm2 and þ8.5 mm2. Mean cross-sectional area over all subjects was 28.5 6 5.8 mm2 on the cine FSE images. Note that mean cross-sectional area was not calculated for the prospectively gated data because images were unevenly sampled across the cardiac cycle. Table 1 summarizes key carotid artery measurements: systolic area was 31.7 6 6.2 mm2 on cine FSE and 31.9 6 6.6 mm2 on prospectively gated FSE (P ¼ 0.87) and diastolic area was 25.7 6 5.6 mm2 on cine FSE and 23.7 6 4.7 mm2 on prospectively gated FSE (P ¼ 0.13). Measures of systolic-diastolic area change on cine FSE and prospectively gated FSE image series were significantly different (P ¼ 0.01). Table 1 summarizes the results of the inter- and intra-rater reliability assessment. For the cine FSE sequence, intra-rater reliability varied from an ICC of 0.86 to 0.94 across the three raters. The inter-rater reliabilities of the sequences were 0.88
1108
on cine FSE and 0.87 on prospectively gated FSE. The interrater minimum detectable difference in systolic-diastolic lumen area change was MDDinter ¼ 4.9 mm2 for the cine FSE approach and 6.4 mm2 for the prospectively gated FSE approach. Intra-rater MDD ranged from MDDintra,1 ¼ 2.6 mm2 to MDDintra,2 ¼ 8.0 mm2 on cine FSE and MDDintra,1 ¼ 4.3 mm2 to MDDintra,3 ¼ 8.0 mm2 on prospectively gated FSE. Scan–rescan reliability of the repeated cine FSE acquisitions had a Pearson’s correlation coefficient of r ¼ 0.898 (P < 0.01). DISCUSSION AND CONCLUSIONS The cine FSE sequence successfully imaged the common carotid arteries in all volunteers and produced mean cross-sectional areas (28.5 mm2 6 5.8 mm2; Table 1) within the confidence interval of previously reported common carotid measurements in a similar young population (29.2 6 0.2 mm2, calculated from reported diameter) (23). No significant difference was found in area measurements on cine FSE and prospectively gated FSE images. While both sequences were black-blood FSEbased techniques, there was noticeably reduced image artifact observed in the vessel lumen of the cine FSE data (Fig. 3). The observed suppression of artifact in the cine FSE images is attributed to the inversion recovery preparation and nonuniform sampling scheme (24), which resulted in easier manual segmentation of the vessel wall-blood boundary and, we hypothesize, more accurate lumen area estimates. Manual measurements on both types of images were highly repeatable, as evidenced by the inter-rater ICCs (0.88 cine FSE and 0.87 prospectively gated FSE). The average systolic–diastolic change in crosssectional lumen area in our group was 6.1 6 1.7 mm2 (cine FSE) and 8.2 6 2.8 mm2 (prospectively gated FSE). Differences in measured distension arise predominantly in diastolic area measurements (Table 1), where flow artifact, primarily in prospectively gated FSE images, can obscure the lumen-wall boundary. In addition, the preselected peripheral trigger delays of the prospectively gated technique may not correspond exactly to true systole and diastole. The cine FSE method, which collects data continuously through all cardiac phases, ensures that the entire cardiac cycle is sampled. The intra-rater MDDs indicate that our best rater is capable of reliably detecting area changes greater than 2.6 mm2 on cine FSE and 4.3 mm2 on prospectively gated FSE, both well below the observed distension in these healthy subjects. Ultrasound studies suggest 50% decreases in internal carotid artery distension with the presence of atherosclerosis (25), which translates to an expected systolic–diastolic area change of only 3 to 4 mm2, suggesting that MDD
