Rapid Left-to-Right Shunt Quantification in Children by Phase-Contrast Magnetic Resonance Imaging Combined With Sensitivity Encoding (SENSE)
Background— Parallel imaging by sensitivity encoding (SENSE) may considerably reduce scan time in MRI. For rapid flow quantification in children with congenital heart disease, we evaluated phase-contrast MRI (PC-MRI) techniques combined with SENSE.
Methods and Results— In 22 pediatric patients (mean age, 7.2±6.2 years) with cardiac left-to-right shunt, blood flow rate in the pulmonary artery (Qp) and ascending aorta (Qs) and flow ratio Qp/Qs were determined by PC-MRI with SENSE reduction-factor 2 and 3 (SF-2 and SF-3). Additionally, we used PC-MRI with higher spatial in-plane resolution (1.6×2.1 versus 2.3×3.1 mm) with and without SF-3. Results were compared with a recently validated standard PC-MRI protocol and tested in vitro using a pulsatile flow phantom. Reduction of signal averages from 2 to 1 and application of SENSE accelerated flow measurements by a factor of 3.5 (5.2) using PC-MRI with SF-2 (SF-3) compared with standard PC-MRI. For blood flow rate through the pulmonary artery and aorta, as well as for the Qp/Qs ratio we found negligible differences of ±3%, lower limits of agreement (mean±2 SD) of −7% to −18%, and upper limits of agreement (mean±2 SD) of +3 to +24%, demonstrating good agreement with standard PC-MRI. Mean Qp/Qs ratio by standard PC-MRI was 1.69±0.45 (range, 1.27 to 2.79). Interobserver variability was low, and high accuracy was confirmed in vitro for all protocols.
Conclusions— PC-MRI for flow quantitation may be combined with SENSE to achieve a substantive reduction of scanning time. In children with left-to-right shunt, Qp/Qs quantification is possible by PC-MRI+SF-3 in <60 seconds. Use of higher in-plane resolution did not improve measurement results.
Received January 27, 2003; revision received June 30, 2003; accepted July 2, 2003.
In pediatric patients, noninvasive assessment or exclusion of cardiac left-to-right shunts by use of phase-contrast cine MRI (PC-MRI) has been demonstrated to be reliable and clinically useful.1–3 Thus, hemodynamic information may be added to a morphological cardiac MR investigation. In the clinical setting, however, repeated flow measurements3 at multiple sites4 may be required. Because each measurement usually takes 2.5 to 5 minutes with standard PC-MRI protocols2,3 depending on heart rate and imaging parameter settings, a thorough hemodynamic MR investigation may be time consuming. Additionally, variations of heart rate5 and respiration rate6 may introduce artifacts. Therefore, a shorter acquisition time would be desirable. The sensitivity-encoding (SENSE) technique uses different single-coil sensitivities of a multi-element phased-array surface coil to spatially encode the MR signal independently from conventional gradient encoding. Thus, Fourier imaging may be accelerated by undersampling of lines in k-space. The resulting foldover is unfolded mathematically using the encoding information available from the surface-coil sensitivities.7 The purpose of this study was to evaluate the reliability of rapid PC-MRI techniques at different in-plane resolutions using the SENSE approach. In pediatric patients with cardiac left-to-right shunt, measurements in the ascending aorta and pulmonary artery were compared with a conventional retrospectively gated PC-MRI sequence that was recently validated in children.3
From July 2001 to February 2002, we enrolled into the study 25 consecutive children with atrial or ventricular level left-to-right shunt clinically referred for additional diagnostic evaluation by MRI. Of these, 3 patients had no shunt as revealed by MRI and were excluded from analysis. The remaining 22 children (mean age, 7.2±6.2 years; median age, 4.6 years; range, 1 to 17 years; 12 girls) formed the study population. Twelve had secundum atrial septal defect (ASD II), 2 had partial anomalous pulmonary venous return (PAPVR), 1 had ASD-II with PAPVR, 3 had superior sinus-venosus defect with PAPVR, 1 had primum ASD, and 3 had ventricular septal defect. All patients were in sinus rhythm. The study was approved by the institutional review committee and informed, written consent was obtained from parents or caretakers.
Each patient underwent MRI to measure through-plane flow in the ascending aorta and pulmonary artery (Figure 1). A standard PC-MRI measurement (reference PC-MRI, Table 1) was repeated twice in each location to serve as a clinical reference method3 and to determine repeatability (ie, precision, Figure 2). The mean value of both measurements was compared with 4 research PC-MRI pulse sequences (see below, Table 1) that were not repeated for the sake of total imaging time. In 6 randomly selected children, reference and research PC-MRI flow measurements were independently reanalyzed by 2 experienced observers to determine interobserver variability. Sedation was performed with midazolam and thiopental intravenously when necessary. Blood pressure, oxygen saturation, heart rate, and respiratory rate were monitored continuously.
All examinations were performed on a 1.5 Tesla whole-body MR scanner (Philips, ACS-NT, maximum gradient performance, 30-mT/m amplitude; slew rate, 150 T/m per s). After the survey scans, a SENSE reference scan was performed, as follows: fast field-echo, coronal orientation, repetition time 8.0 ms, echo time 0.6 ms, flip angle 7 degrees, voxel size 16.5×16.5×18 mm, 8 signal averages, water-fat-shift 0.06 pixels, free-breathing, and scan duration <1 minute. The conventional standard PC-MRI pulse sequence for both in vivo and in vitro flow measurements (reference PC-MRI) was based on a protocol recently validated in children with left-to-right shunt3 (Table 1) and served as reference method in our study. The 4 research PC-MRI protocols included a combination of standard PC-MRI with SENSE reduction-factor (SF) of 2 (PC-MRI+SF-2), SF of 3 (PC-MRI+SF-3), and a PC-MRI protocol with higher spatial resolution with SF of 3 (PC-MRIHR+SF-3) and without SENSE (PC-MRIHR). In-plane resolution was ≈2.3×3.1 mm2 for PC-MRI with or without SENSE and 1.6×2.1 mm2 for PC-MRIHR with or without SENSE, respectively. In PC-MRI techniques using SENSE, no rectangular field of view (FOV) was used to avoid backfolding because of the SENSE reconstruction algorithm.7 All PC-MRI measurements were obtained under free-breathing conditions and used retrospective vector-ECG8 gating to include end-diastole flow, repetition time 15 ms, ≈20 to 25 reconstructed heart phases, slice thickness 6 mm, flip-angle 30 degrees, and velocity-encoded value 200 cm/s. The body coil was used for signal transmission, and a 5-element phased-array surface (cardiac) coil was used for signal detection, with 3 elements placed in the back and 2 on the chest (patients positioned supine and head-first in the scanner). All flow measurements used a phase-correction algorithm provided by the manufacturer.
A polyvinylchloride tube with an inner diameter of 12.7 mm and a wall thickness of 2.4 mm was embedded in a glass tube (diameter 100 mm, length 340 mm) filled with copper sulfate solution (3 mmol). To approximate blood viscosity,9 the flow-phantom was filled with a mixture of distilled water and glycerol (60/40 vol/vol). The flow-phantom was placed parallel to the bore in the isocenter of the magnet. The cardiac surface coil was used for signal acquisition. The tubing was connected to a 60-mL artificial left ventricular system (Medos HIA VAD-System II) to deliver pulsatile flow. Flow pump and MR scanner were simultaneously triggered by an external ECG simulator. With a constant heart rate of 84 bpm, 6 different flow rates (ie, 497, 854, 1127, 1401, 1827, and 2219 mL/min) were achieved by appropriate adjustment of the systolic and diastolic pressures of the ventricular assist device. Flow rate was determined 3 times before and after the MR experiments by stopwatch/graduated cylinder to obtain the reference (true) values. All in vitro MR flow measurements were performed through-plane with velocity-encoding value of 150 cm/s and were repeated 3 times.
MR Image Analysis
Data analysis was performed offline on a computer workstation using a computer algorithm for semiautomatic vessel border detection developed by one of the authors (P.B.)3 to accelerate image analysis and optimize measurement reproducibility.10 In all standard and research PC-MRI flow measurements of 1 subject (5 years, ASD), we determined the signal-to-noise ratio (SNR) in magnitude images11 and the velocity-to-noise ratio (VNR) in phase images12 in the pulmonary artery and aorta (maximum flow, systole). The formula used to calculate SNR was as follows: equation
and the formula for VNR was as follows: equation
where SImean is signal intensity, SDnoise is standard deviation of noise, Venc is the velocity encoding value, and SDv is the standard deviation of velocity noise.
All data are expressed as mean±SD. In vitro results were analyzed by 2-variable linear regression analysis. The analysis of Bland and Altman13 was used to determine PC-MRI interobserver variability and to evaluate the agreement between the different PC-MRI techniques and the mean of both reference PC-MRI measurements for flow rate in the pulmonary artery (Qp) and aorta (Qs) and flow ratio Qp/Qs (Figure 3).
In Vivo Results
PC-MRI studies were completed within 15 to 20 minutes. Sedation for MRI was well tolerated. The mean heart rate was 101±20/min. Scan time was reduced to 28% with PC-MRI+SF-2 and to ≈19% with PC-MRI+SF-3 compared with the standard PC-MRI protocol and to 18% with PC-MRIHR+SF-3 compared with PC-MRIHR without SENSE, respectively (Table 1). Thus, scanning was accelerated by a factor of 3.5, 5.2, and 5.5 for PC-MRI+SF2, PC-MRI+SF-3, and PC-MRIHR+SF-3, respectively. This was achieved by reduction of signal averages and by application of SENSE (Table 1).
Comparison of Reference With Research PC-MRI in Children
In our pediatric study population with cardiac left-to-right shunt, the mean Qp/Qs was 1.69 (SD±0.45), as assessed by reference PC-MRI. The Bland-Altman analysis was applied to the log-transformed data,13 because differences increased linearly with mean stroke volumes and Qp/Qs values. Estimation of precision of the limits of agreement (defined as mean±2 SD) was based on calculation of 95% CI. A mean value of 1.0 after anti-log transformation (dimensionless ratio) is expected in the case of no difference between 2 tested methods. All calculated values are detailed in Table 2 for pulmonary artery (Qp), ascending aorta (Qs), and the flow ratio Qp/Qs. For through-plane flow measurements in the pulmonary artery (Figure 1) and ascending aorta and for the Qp/Qs ratio, we found a negligible difference of ±3% (mean, 0.97 to 1.03) between reference PC-MRI and all research PC-MRI techniques. Upper limits of agreement were in the range of 1.03 to 1.16, and lower limits of agreement were between 0.86 and 0.93, respectively. Thus, reference and research PC-MRI sequences assessing cardiac stroke volumes and Qp/Qs may differ by 3% to 16% above and 8% to 14% below in 95% of the cases, demonstrating good agreement between reference and research PC-MRI. With PC-MRI+SF-3 for assessing Qp/Qs, we observed slightly more scatter (Table 2), with upper and lower limits of agreement of 1.24 and 0.82 (95% CI, 1.14 to 1.35 and 0.75 to 0.89, respectively), but agreement was excellent (mean 1.01). Only little scatter was present with high-resolution PC-MRI+SF-3, where upper and lower limits of agreement were 1.12 and 0.89 (95% CI, 1.06 to 1.17 and 0.84 to 0.93, respectively).
Repeatability and Interobserver Variability in Children
Repeatability (precision or absence of random error) of the 2 reference PC-MRI flow measurements was excellent for the pulmonary artery (aorta), with a mean deviation of 2.3±1.7% (1.9±2.1%). Reevaluation of standard and research PC-MRI flow measurements by 2 observers (Bland-Altman analysis, data log-transformed) revealed a negligible mean difference of 0.99 to 1.01, with limits of agreement ranging from 0.93 to 1.10 (mean±2 SD), demonstrating low interobserver variability.
Accuracy and Precision In Vitro
Table 3 summarizes the in vitro flow measurements by timed collection (stopwatch/graduated cylinder=true values) and by reference PC-MRI, PC-MRI+SF-2, PC-MRI+SF-3, and PC-MRIHR without SENSE and PC-MRIHR+SF-3. The flow values in Table 3 represent mean values of repeated measurements, which had a maximum deviation of ±2%, indicating high repeatability in vitro (ie, measurement precision) for all PC-MRI techniques. Results of linear regression analysis also presented in Table 3 demonstrate good correlation between true values and reference PC-MRI, PC-MRI with SF-2, and PC-MRI with SF-3, indicating high accuracy. However, we observed a small but statistically significant underestimation of flow rate using PC-MRIHR, both with and without SENSE.
Signal-to-Noise Ratio and Velocity-to-Noise Ratio
SNR and VNR values11,12 are expressed as percentages of the standard (reference) PC-MRI protocol with cardiac surface coil for signal acquisition (100%). In the aorta (pulmonary artery), SNR was reduced to 47% (43%) with PC-MRI+SF-2, to 26% (26%) with PC-MRI+SF-3, and to 20% (15%) with PC-MRIHR+SF-3. When the standard PC-MRI was used with body coil instead of cardiac coil, SNR was reduced to 29% (37%).
In contrast, in the aorta (pulmonary artery), VNR was only reduced to 95% (88%) with PC-MRI+SF-2, to 95% (82%) with PC-MRI+SF-3, and to 88% (74%) with PC-MRIHR+SF-3. When the standard PC-MRI was used with body coil instead of cardiac coil, VNR was reduced to 95% (96%).
In the SENSE approach,7 an array of multiple, simultaneously operated receiver coils for signal acquisition (eg, cardiac 5-element phased-array surface coil) is used to spatially encode the MR signal as a function of the sensitivity of each coil element. This allows to reduce the number of phase-encoding steps by increasing the distance of readout lines in k-space (ie, parallel imaging) and thus accelerate acquisition, but at the expense of a lower SNR.7 Therefore, combination with SENSE is only recommended when SNR is not of critical importance for diagnostic image quality.
This study demonstrates that standard PC-MRI under free-breathing conditions may be favorably combined with SENSE. Quantitative flow measurements can be accelerated by a factor of 3.5 to 5 by reduction of signal averages (from 2 to 1) and application of SENSE. The corresponding SNR reduction was no limiting factor. Therefore, in children with congenital heart disease, left-to right shunt may be quantified reliably in less than 1 minute. The considerable time benefit is likely to be generally available in hemodynamic evaluation of congenital heart disease. Other possible applications include quantification of residual pulmonary valve regurgitation after surgery for Tetralogy of Fallot14 and the quantification of flows in the venae cavae and branch pulmonary arteries after Fontan palliation in patients with single-ventricle physiology.4 Thus, either an examination can be shortened (reduced duration of sedation/anesthesia in children or increased patient throughput) or more comprehensive flow information may be collected in a given amount of time (ie, multiple4 or repeated3 measurements to improve reliability). Additionally, in pediatric patients with higher heart rates, breath-hold scans during anesthesia have now become possible using a SENSE reduction factor of 3 (Table 1). This may additionally improve measurement accuracy, because artifacts from respiratory motion can thus be avoided.6 Although reliable, high in-plane resolution was not advantageous in our children but may be adequate for infants with smaller blood vessels to avoid possible flow overestimation resulting from partial volume effects.15
This 3- to 5-fold increase in imaging speed is achieved without changing imaging parameters designed to optimize measurement accuracy and precision.10,16 No drawbacks in measurement accuracy and precision were observed in vivo. This finding is surprising, because both reduction of numbers of signal averages and combination with SENSE result in lower SNR. This relates not only to SNR penalty bounded by a square-root function to the number of samples acquired but also contains noise from sensitivity maps as propagated in the SENSE reconstruction process,7 known to increase with higher SENSE-reduction factors. Our SNR estimations in PC-MRI magnitude images demonstrate considerable SNR improvement in vivo when a cardiac multi-element surface coil is used for signal acquisition instead of the body coil.3 This gain in signal can be invested to reduce scanning time. As apparent in Figure 1, SNR was sufficiently high even in PC-MRI+SF-3. Furthermore, our estimation of VNR in the phase images12 revealed no relevant decrease of VNR in research PC-MRI with reduced signal averages and SENSE compared with reference PC-MRI, although a decrease in SNR in the magnitude images11 was present. Useful phase-angle information, however, can still be extracted when SNR in magnitude images is reduced, because phase-angle images have an inherently better dynamic range compared with magnitude images reconstructed from the same raw data.17
Methodological Issues: PC-MRI With SENSE
A 1-minute scan is needed at the beginning to detect surface-coil sensitivities for all SENSE scans. In our experience, no breath-hold is required in children. To avoid severe deterioration of image quality from backfolding artifacts (attributable to the SENSE reconstruction process), no additional foldover should occur during planning of the desired FOV. Therefore, we chose a relatively large FOV of 380 mm to enable the use of a standardized PC-MRI+SENSE protocol at different body sizes or slice angulations with respect to geometrical properties and gradient performance. As a result, however, the impact of SENSE reconstruction may have been variable, depending on angulation and body size as related to the FOV. For instance, in an axial plane and a small child, the actual SENSE rate is lower than nominal because FOV exceeds the extent of the chest, and therefore only limited backfolding occurs that needs unfolding using the SENSE technique. On the other hand, in a larger subject with double-oblique slice orientation, this FOV may be even borderline small, so that compromises in spatial resolution and slice angulation may occasionally become necessary, especially the pulmonary artery. An anterior-posterior phase-encoding direction is most favorable when using SENSE. If necessary to achieve this direction at double-oblique slice angulation, some misalignment between flow-encoding gradient and blood flow direction may be allowed, with only minimal bias.16
Methodological Issues: Reference PC-MRI
Although there is no gold standard available for quantitative flow measurements in pediatric patients, PC-MRI protocols comparable to our reference method have been validated in vivo and in vitro.1–3,10,16 Some changes in the reference PC-MRI protocol compared with the recently validated method3 need consideration. Use of the cardiac surface coil (increased SNR, see the Results section) allowed to reduce signal averages in the reference PC-MRI protocol from 4 to 2, thus limiting overall scan time. Moreover, vector-ECG8 instead of conventional 3-lead-ECG allowed more reliable ECG-monitoring and hence assignment of heart phases. Therefore, it seems reasonable to assume that measurement conditions of the reference PC-MRI sequence were improved, as reflected by the high precision of flow measurements.
In Vitro Results
The flow phantom experiments were designed to test the same PC-MRI protocols as were used in vivo and demonstrated high accuracy at lower in-plane resolution. Apparently, no major artifact was introduced when using PC-MRI with SENSE in this model. The FOV, however, was too large to ensure that 2- and 3-fold undersampling would lead to actual foldover in the single coil data. Thus, no additional conclusions regarding the impact of SENSE acquisition and reconstruction on phase encoding of flow volume should be drawn. An adopted FOV with reduced matrix for a similar voxel size, however, would have been different from the PC-MRI protocols used in vivo.
High-resolution PC-MRI both with and without SENSE slightly underestimated flow volumes compared with values from timed collection (Table 3). Because stronger gradients are needed to achieve the higher in-plane resolution, we speculate that artificial effects from residual eddy currents and concomitant fields may be involved.18,19 Any possible explanation, however, suffers from the fact that this underestimation was not obvious in vivo (Table 2).
No conclusions from our data are applicable to children with arrhythmia, irregular breathing pattern, or valvular stenosis with turbulent blood flow. Only relative changes should be emphasized in the SNR and VNR measurements, because absolute values may be influenced by background filtering in the reconstructed images, reducing the noise component.
PC-MRI for flow quantitation may be combined with SENSE to achieve a substantive reduction of scanning time. SNR reduction was not a limiting factor. In children with left-to-right shunt, Qp/Qs quantification is possible by PC-MRI+SF-3 in <60 seconds. Use of higher in-plane resolution did not improve measurement results but may be adequate for infants with smaller blood vessels. The method has potential for rapid quantification of hemodynamics in congenital heart disease.
The study was supported in part by Philips Medical Systems, Best, the Netherlands. We thank Kerstin Berkemeier, RN, for data postprocessing and Hermann Esdorn, MD, and Andreas Peterschröder, MD, for data acquisition.
Presented at the 37th Annual Meeting of the Association for European Pediatric Cardiology, Porto, Portugal, May 16, 2002.
Beerbaum P, Korperich H, Barth P. Noninvasive quantification of left-to-right shunt in pediatric patients: phase-contrast cine magnetic resonance imaging compared with invasive oximetry. Circulation. 2001; 103: 2476–2482.
Houlind K, Stenbog EV, Sorensen KE. Pulmonary and caval flow dynamics after total cavopulmonary connection. Heart. 1999; 81: 67–72.
Wood ML. Fourier imaging. In: Stark DD, Bradley WG, eds. Magnetic Resonance Imaging. St Louis: Mosby Year Book; 1992: 21–67.
Kozerke S. Assessment of Heart Valve Function by Motion Adapted Magnetic Resonance Imaging [dissertation]. Zurich, Switzerland: Swiss Federal Institute of Technology Zurich; 2000.