Breath-hold and free-breathing quantitative assessment of biventricular volume and function using compressed SENSE: a clinical validation in children and young adults

The assessment of cardiac volumetric indices is important for the diagnosis and follow-up of both congenital and acquired heart disease [1‐6]. Cardiovascular magnetic resonance (CMR) imaging is an accurate and reproducible modality that is the clinical reference standard for quantitative evaluation of ventricular chamber size, function and myocardial mass [6‐11]. Currently, retrospectively cardiac gated two-dimensional segmented k-space cine balanced steady state free precession (bSSFP) is the preferred CMR sequence for the quantitative assessment of cardiac function. The bSSFP sequence has high intrinsic blood pool to myocardium contrast and high signal-to-noise ratio (SNR) that results in well-defined endocardial boundaries throughout the cardiac cycle [12‐14]. Short-axis (Sax) cine bSSFP images are routinely acquired during breath-holds, because the bSSFP sequence is susceptible to artifacts from respiratory motion and disruption of magnetization steady state [14]. In routine clinical practice, one to three cine SAx slices are acquired in a breath-hold (BH) of 5 to15 cardiac cycles by trading the intrinsic high bSSFP signal-to-noise (SNR) for imaging speed using parallel imaging techniques that employ regular k-space undersampling in the spatial dimensions e.g. sensitivity encoding (SENSE), without compromising the blood to myocardial contrast and providing adequate spatio-temporal resolution [15, 16]. Although the accuracy and reproducibility of CMR bSSFP for the measurement of ventricular volumes, function, and cardiac mass is well established [17‐19], the requirement for repeated BHs remains a limitation, especially in children and sedated patients. Accelerated cine CMR techniques (such as k-t BLAST, TPAT, TSENSE, and compressed sensing) have reported bias in left ventricular (LV) volumes, function, and LV mass due to spatiotemporal blurring and temporal filtering [20‐26]. Free-breathing (FB) respiratory triggered retrospectively cardiac gated cine bSSFP sequences have been reported to provide biventricular volumes, function, and LV mass comparable to BH acquisitions with a SENSE acceleration factor of 2 in adults and children [27, 28]. With the goal of accelerating SENSE, a compressed sensitivity encoding (C-SENSE) algorithm was developed that employs a pseudorandom undersampling of k-space in the spatial domain. C-SENSE provides diagnostic CMR image quality with acceleration factors greater than 2, allowing for significantly reduced BH times [29]. However, there is little data validating quantitative ventricular assessment using C-SENSE either with BH or FB acquisitions [30].

The purpose of our study was to test the hypothesis that the use of C-SENSE acceleration in BH and FB respiratory-gated retrospectively cardiac gated cine bSSFP sequences produces diagnostic quality images and accurate ventricular volumetric indices with decreased BH times.

This HIPAA-compliant, retrospective study was approved by the institutional review board (IRB) at our institution. The requirement for informed consent was waived. The free-breathing cine bSSFP sequence in its current form was implemented within our institution, and all the data and information were always under the control of our institution.

There were 26 patients (26 males; 16.7 ± 6.4 years (range: 9–35), body surface area (BSA) 1.6 ± 0.4 (range: 0.94–2.4 m²) and heart rate 83 ± 7.2 beats/min (range: 50–115)) who underwent all three SAx cine scans during the study period. Table 1 summarizes the patient characteristics. Indications for CMR included Duchenne muscular dystrophy (n = 12), pectus excavatum (n = 10), Becker muscular dystrophy (n = 1), Marfan syndrome (n = 1), Turner syndrome (n = 1), and chemotherapy induced cardiomyopathy (n = 1). All SAx cine images were obtained without technical failure or significant artifact. A total of 15 ± 2 (range: 12–18) SAx slices were acquired per patient in 8 ± 1 breath-holds (2 slices per breath-hold) (range: 6–9). BH durations of 8 ± 1 (range: 7–10) sec with C-SENSE = 3 were significantly shorter (p < 0.001) by 33% compared to 12 ± 1 s (range: 10–14) with SENSE = 2. Actual image acquisition time, including BH instructions and time between BHs, for the SAx stack with BH SENSE = 2 (4.9 ± 1.2 min (range: 2.6–7)) was significantly longer (p < 0.05) compared to that with BH C-SENSE = 3 (4.3 ± 1.5 min (range: 2.1–7)) and comparable to that with FB C-SENSE = 3 (5.2 ± 1.5 min (range: 2.3–7)). Figure 1a depicts the comparison of scan time per slice.

Table 2 demonstrates the data comparing LV and RV volumetric indices measured with the three sequences. LV and RV stroke volume and ejection fraction, and LV mass computed were comparable. There were no significant differences in RV volumes, and LV and RV stroke volumes, ejection fractions, or myocardial mass between BH SENSE and BH C-SENSE sequences. There was a small but statistically significant difference in LV and RV ventricular end diastolic and end systolic volumes between BH C-SENSE and FB C-SENSE sequences. Figure 2 depicts linear regression and Bland-Altman plots comparing LV stroke volume computed using aortic quantitative flow with those computed using three cine bSSFP SAx acquisitions. The LV stroke volumes from the three sequences had excellent correlations with aortic flow, regression slopes ranged from 0.98 to 1.02. All three sequences overestimated the LV stroke volume by 2.7 (5%) to 4.6 (8%) ml/beat compared with the aortic phase contrast data. The limits of agreement of LV stroke volumes for all three sequences was less than 24%.

BMC normalized to myocardial signal for FB C-SENSE was significantly lower (p < 0.05) than with both BH SENSE and BH C-SENSE acquisitions (Fig. 1b). Image quality scores were comparable between the three readers across all criteria in all sequences. The mean of three observers’ scores in each criterion were used for further analysis. Mean rank scores for BMC, EED, artifacts, and total image quality score with FB C-SENSE acquisitions were significantly lower (p < 0.005) than with the BH SENSE acquisitions (Fig. 1c). Mean rank scores for EED and combined total image quality scores with BH C-SENSE were significantly lower (p < 0.05) than BH SENSE. Mean rank scores in each scoring criterion were comparable between BH SENSE and BH C-SENSE sequences. Figure 3 depicts the image quality scores for each scoring criteria. The combined image quality score was excellent (16 of 26) to good (10 of 26) with BH SENSE, excellent (13 of 26) to good (13 of 26) with BH C-SENSE, and excellent (3 of 26) to good (21 of 26) to adequate (2 of 26) with FB C-SENSE (Fig. 3). For the combined image quality score, the difference between BH SENSE and BH C-SENSE was (0.08 ± 0.18 (range: − 0.50 - 0.39) and the difference between BH SENSE and FB C-SENSE was (0.42 ± 0.34 (range: − 0.26 - 1.17). Figure 4 shows representative images with excellent, good, and adequate combined image quality scores using all three acquisition techniques.

The results of this retrospective study demonstrated that image quality and biventricular volumetric indices using cine bSSFP acquisition with C-SENSE acceleration factor of 3, either during BH or FB, are comparable to the standard of care BH cine bSSFP acquisition with SENSE acceleration factor of 2. Both BH and FB cine bSSFP sequences with C-SENSE = 3 had spatio-temporal resolution nominally identical to the standard of care BH sequence and provided diagnostic image quality in all 26 patients encompassing a wide range of body sizes and heart rates. The BH duration was reduced by 33% using C-SENSE = 3 compared to SENSE = 2. The total imaging time for the FB acquisition with C-SENSE = 3 was comparable to SENSE = 2.

Although the standard BH cine bSSFP sequences are well established, accelerated acquisition with shorter BH times or without BHs helps with imaging children and severely ill or sedated patients. Numerous strategies for undersampling k-space in the time domain using either regular or irregular patterns in combination with either prospective cardiac gating or real-time cine imaging have been reported to provide diagnostic image quality and comparable ventricular volumetric assessment [20, 21, 23‐26, 39‐43]. However, studies with temporal undersampling schemes have reported underestimation of the LV mass and bias in both stroke volume and ejection fraction [23‐26]. Some of the difference in volumetric indices can be attributed to experimental and physiologic variations. However, bias in volumetric measurements is partly due to the decrease in effective temporal resolution when using temporal undersampling. Systematic evaluation of incremental increases in temporal undersampling using different reconstruction approaches on fully sampled cine bSSFP data has shown that these biases worsen with increasing acceleration factors and can be larger than physiologic variations [44]. In our study, bias and standard deviations for volumetric indices of both BH and FB C-SENSE = 3 sequences compared to standard of care BH SENSE = 2 sequence are comparable to the inter- and intra-observer values reported in the literature [19, 26, 35, 45‐47]. The overestimation of LV stroke volume by SAx measurements compared to aortic flow is 4 to 7% of the LV mass, which is comparable to reported LV trabeculation mass value of 8.2% seen in healthy subjects [48]. A consistent, small, non-zero velocity offset in aortic phase contrast data due to double-oblique slice orientation at the sinotubular junction may also have contributed to this difference. Overall, the statistically significant but small absolute difference seen in ventricular volumes between BH and FB is not clinically significant.

Although biventricular volumetric indices and LV mass were comparable across the three acquisitions, the endocardial delineation of both the BH and FB C-SENSE acquisitions were slightly inferior to BH SENSE. This suggests that spatial blurring caused by spatial domain pseudorandom sampling likely contributes to the differences in RV volumetric indices, exacerbated by the irregular RV contour shape. Decrease in blood-to-myocardial contrast can be attributed to diminished blood pool signal due to increased undersampling. The differences in BMC image quality between BH and FB C-SENSE suggest inferior attainment of steady state in FB sequence due to longitudinal through plane motion during the cardiac cycle, particularly in ventricular basal slices. The variability in depth of breathing may also contribute to the decrease in EED and artifact scores for FB compared to BH C-SENSE. This is noted frequently in our patients with muscular dystrophy and pectus excavatum. Another artifact noticed in larger patients was residual parallel lines next to bSSFP black bands at the edges of the field of view in images with C-SENSE factor of 3.

In addition to the noise penalty associated with the undersampling, the noise is further amplified in parallel imaging due to the nonorthogonality of the coil sensitivity profiles reflected by geometry or g-factors [49]. In 2D cartesian parallel imaging techniques, the image degradation due to noise increases exponentially around critical reduction factor of 3 to 4 [50]. A recent study in healthy adults using C-SENSE factor of 4 for cine bSSFP imaging showed adequate image quality; however, the acquired voxel size of 2.8 mm in phase encode direction was significantly larger than 1.6–1.7 mm used in our pediatric population [30]. Additionally, that study depended on shallow breathing while our study used explicit respiratory gating. In order to minimize the spatial blurring that occurs with iterative reconstruction from pseudo randomly sampled k-space data while trying to accelerate image acquisition, we conservatively tested a C-SENSE factor of 3 in this study. This study showed that the C-SENSE factor of 3 shortened the breath-hold time for SAx cine bSSFP acquisitions significantly while maintaining adequate LV and RV myocardial border definition. We showed good agreement of volumetric indices with those acquired with SENSE factor of 2. This reduction in acquisition time can be traded for either faster patient throughput by acquiring additional slices per BH or to accommodate patients with impaired BH capacity.

Furthermore, the study demonstrated that C-SENSE acquisitions can be used for quantitative ventricular assessment along with FB cardiorespiratory synchronized cine bSSFP. Although study population consisted of children and young adults with structurally normal hearts, the sequence has potential utility for FB cardiorespiratory synchronized cine bSSFP in combination with C-SENSE in older adults as well, especially patients with impaired BH capacity or capability. There is nothing about the sequence or acquisition scheme that would preclude large or older patients.

There are few limitations to this study. First, we did not acquire the cine bSSFP data with full k-space sampling. Quantitative aortic flow was used as an internal reference to partially address this limitation. Second, the study population with structurally normal hearts mostly consisted of pectus and muscular dystrophy patients, both of which have male predominance. Third, the data acquisition for this retrospective study was performed as part of clinical scan sessions, so actual scan time comparisons between BH acquisitions is confounded by the irregular gaps between consecutive BHs. Lastly, SNR between different acquisitions was not compared quantitatively given the complex spatial distribution of noise in C-SENSE due to both data pseudorandom acquisition and iterative reconstruction. Differences in SNR were partially incorporated in the BMC measurements.

Cine bSSFP imaging with compressed SENSE acceleration factor of 3 reduces BH times by 33% compared to a clinically established SENSE acceleration factor of 2 with nominally identical spatio-temporal resolution while providing comparable LV and RV volumetric and functional indices and image quality. Also, C-SENSE combined with cardiorespiratory synchronized free-breathing cine bSSFP imaging provides comparable LV and RV volumetric and functional indices to that with clinically established BH acquisition with SENSE factor of 2 in comparable time. This reduction in acquisition time can be traded for either faster scan sessions with more slices per breath-hold or for reduction or elimination of BHs to accommodate patients with impaired BH capacity.