Introduction
Cardiac magnetic resonance imaging (MRI) and contrast-enhanced MR angiography (CEMRA) have become mainstream clinical tools during the past decade with applications in pediatric and adult congenital heart disease (CHD) [
1,
2]. Whereas the vast majority of reports on cardiac MRI/MRA have focused on 1.5 T, market data as far back as 2004 indicate that 3.0-T systems make up 25% of new MRI purchases in the United States [
3]. Since its approval for whole-body imaging by the United States Food and Drug Administration, 3.0-T systems have become the standard for neurological and musculoskeletal imaging at many institutions [
4]. Despite high signal-to-noise ratio and increasing availability, adoption of cardiac imaging at 3.0 T has been slow, largely due to the more troublesome off-resonance artifacts on steady-state free precession (SSFP) cine images [
5,
6]. Whereas in adults, several studies have addressed the advantages of 3.0 T for high spatial CEMRA [
5‐
7], to the best of our knowledge, there are no published reports specifically addressing imaging of pediatric CHD at 3.0 T compared to 1.5 T.
The purpose of our study, therefore, was to determine whether both high-resolution (HR) CEMRA and diagnostic-quality SSFP cardiac cine can be acquired reliably in pediatric CHD patients at 3.0 T, and to compare objective and subjective measures of image quality to similar techniques at 1.5 T.
Discussion
Our results affirm that both high-resolution CEMRA and diagnostic SSFP cardiac cine images were obtained reliably at 3.0 T in neonates and small children (ages 3 days to 8 years, weighing 1.4–25.0 kg). With patient-specific frequency tuning [
22], off-resonance SSFP artifacts were manageable and the cine images were generally diagnostic. Further, signal-to-noise ratio and contrast-to-noise ratio are significantly increased at 3.0 T relative to 1.5 T and CEMRA with very high-resolution matrices was routinely achievable.
The clinical value of MR is well established for a variety of cardiovascular disorders [
2,
23‐
26]. Whereas in adults, the manifestations of cardiac disease can generally be defined by focusing on the heart, children with CHD typically have both intracardiac and extracardiac anomalies. Therefore, the ability to visualize extracardiac vascular anomalies in addition to intracardiac morphological abnormalities is crucial. Although echocardiography is readily available, it may leave several morphological and hemodynamic questions unresolved. Cardiac CT with either single- or dual-source technology can yield high spatial resolution images with increasingly small radiation doses. However, sub-milliSievert doses are achievable only in the most ideal situations (low body mass index, sinus rhythm with excellent beta blockade). Exposure to radiation, however low the dose, should be considered only when alternative methods are not available. Furthermore, neonates and infants may have heart rates that challenge the temporal resolution of even dual-source CT scanners. With MRI, the temporal resolution of cine imaging can be adjusted to cope with even the fastest heart rates [
27]. MRI has been shown to be a versatile, noninvasive and non-radiating technique for the evaluation of neonates, infants and children, including those who are critically ill [
2,
28,
29].
In young children, the small caliber of the heart and blood vessels requires higher spatial resolution than in adults [
23,
25] for comparable detail. In Cartesian MRI, spatial resolution is expressed in the three logical dimensions (frequency encoding, in plane-phase encoding and through-plane phase encoding) that define the volume element, or voxel. In 3-D imaging, all three orthogonal directions are equally relevant in determining voxel size. Each voxel dimension may be expressed as the ratio of the field of view (FOV) to the matrix size. For CEMRA of the thorax in adults at 1.5 T, voxel volumes on the order of 2–2.5 mm
3 are not atypical even for acquisitions regarded as high resolution [
30,
31]. For CEMRA of the thorax and abdomen in children at 1.5 T, voxel volumes of 2.7 mm
3 have been described [
32] and in neonates, to our knowledge, the smallest voxels previously described have been 1.6 mm
3 [
33]. Whereas for adults and larger children, spatial resolution as previously reported may be entirely adequate, for small children with diminutive central vessels, the requirements for high spatial resolution are much more stringent. Earlier reports of CEMRA in the abdomen with limited spatial resolution in children found disappointing results relative to CT [
34]. We therefore sought to maximize spatial resolution in small children by performing CEMRA at 3.0 T with fine matrices and aggressive parallel acquisition factors. Our highest spatial resolution in the smallest children at 3.0 T is 8 times greater than previously reported in the literature and 2.5 times greater than what we achieved at 1.5 T. As spatial resolution is increased, signal-to-noise ratio decreases and this generally limits the available spatial resolution at 1.5 T. We were able to realize unprecedented spatial resolution for CEMRA in small children because the higher available signal-to-noise ratio at 3.0 T can be used to maintain adequate signal-to-noise ratio even as the voxel size is diminished.
For SSFP cine imaging, we aimed for less aggressive increments in spatial resolution relative to 1.5 T, because increasing spatial resolution generally involves the use of a longer time repetition (TR) and a corresponding increased sensitivity to off resonance banding artifact. We used the maximum allowable flip angle in all cases to achieve the shortest TR possible. Because imaging of very small children requires high spatial resolution and small FOV, the short TR in SSFP imaging is often in conflict, which may be a limiting factor for SSFP imaging at 3.0 T. Despite these challenges, the smallest achievable SSFP voxel volume in our study was 2.4 mm3 at 3.0 T and 2.8 mm3 at 1.5 T.
At 3.0 T, the boost in baseline signal-to-noise ratio can be exploited to support higher resolution [
17,
18]. Our signal-to-noise ratio measurements showed several interesting features. Theoretically, signal-to-noise ratio increases linearly with field strength. In practice, signal-to-noise ratio is affected by voxel size, receiver bandwidth, RF coil sensitivity profile, and sequential vs. parallel image acquisition [
5,
6]. Despite these trade-offs, we observed an increase in myocardial and blood signal-to-noise ratio by a factor of 2.4 and an increase in contrast-to-noise ratio by a factor of 3.3. Although consistent with published literature on SSFP cine imaging at 3.0 T in the adult population [
35‐
37], it should be noted that our signal-to-noise ratio/contrast-to-noise ratio measurements were made at end-diastole, where off-resonance is typically not prevalent at either field strength. With this caveat in mind, the higher prevalence of off-resonance artifact at the base of the heart may well tip this balance more in favor of 1.5 T. In our study, the acquisition matrix for SSFP cine at 3.0 T was slightly coarser than that typically used at 1.5 T in order to minimize the TR per line and mitigate off-resonance effects. Despite the higher prevalence of artifacts on images acquired at 3.0 T, the image quality was comparable between 3.0 T and 1.5 T as reflected by well-defined borders and high myocardium to blood pool contrast ratio on 3.0-T images.
Another factor that may contribute to the high measured signal-to-noise ratio at 3.0 T relates to the frequency response function of the SSFP signal [
38]. As the frequency of the spins drifts away from resonance, the signal amplitude initially increases before decreasing dramatically at +/− (1/2TR) to form a dark band. The signal from blood in the neighborhood of this off-resonance condition, therefore, may actually increase and in doing so may yield a higher value for signal-to-noise ratio and blood-myocardial contrast-to-noise ratio. However, this is not a desirable condition because it represents an unstable state from which any increase in resonance offset will result in dark band artifact. Price et al. [
22] recently addressed the importance of stabilizing the frequency for cardiac SSFP imaging in neonates at 3.0 T. Although the authors used a different scanner platform, they noted a significant drift in the resonant frequency in normal, spontaneously breathing neonates at 3.0 T and emphasized the importance of local shimming and adjustment of resonance frequency offsets. In our study population, dark band artifacts occurred more frequently in the LAX and basal SAX views. We did not observe a particular relationship between subject size and the presence of dark band artifacts.
Conventional wisdom suggests that SSFP imaging is better at 1.5 T than at 3.0 T. This is routinely the case. However, in our experience, local frequency tuning provided diagnostic SSFP MR imaging at 3.0 T in a wide range of patient size, age and complex cardiac as well as vascular abnormalities with some increase (additional 2–3 min) in overall exam time. At 3.0 T, frequency tuning was performed about 30% of the time; whereas, at 1.5 T, only 4% of the studies required frequency tuning. At 3.0 T (as compared to 1.5 T), the higher signal-to-noise ratio as manifested by well-defined myocardial borders and contrast is impacted negatively by the increased presence of artifacts.
Several limitations in our study merit discussion. We performed a retrospective analysis with a cross-sectional study design. Ideally, our patients would undergo the same sequence protocol at 1.5 T so that intra-individual comparisons can be performed. However, ethical considerations relating to the study of infants and neonates under anesthesia preclude non-indicated MR scans to be performed purely for head-to-head comparative research purposes. Further, due to the unique nature of the underlying spectrum of pathologies, each exam was tailored to the specific indication and therefore corresponding views and imaging sequences were not always available for comparison in all patients. Importantly, our criteria for assigning patients either to 3.0 T or 1.5 T results in clear selection bias. We assigned very small children to 3.0 T, if available, to achieve the highest resolution CEMRA. In doing so, we imposed asymmetry in the 3.0-T and 1.5-T patient groups, such that the 1.5-T group is on average older and larger. However, there is substantial overlap in the older and larger patients of both 1.5-T and 3.0-T cohorts, making head-to-head comparisons reasonable. Additionally, we recognize that with parallel imaging [
17,
18], measurement of signal-to-noise ratio becomes complicated. However, the design of our study made it difficult to implement the signal-to-noise ratio measurement methods described by Kellman et al. [
39] and Reeder et al. [
40]. Although the method used in our paper to estimate signal-to-noise ratio is less than ideal, others have compared our practical alternative approach to other methods available and have found it to produce reasonable estimates [
41,
42].