Introduction
Differential blood oxygenation between the left and right heart (ΔSaO
2) is an established index of cardiac performance; left ventricular (LV) dysfunction results in stagnant blood flow – resulting in increased time for organ extraction of oxygen from blood and delivery of a greater fraction of deoxygenated blood to the right heart. Increased ΔSaO
2 has been shown to predict adverse prognosis in patients with heart failure with and without pulmonary hypertension [
1‐
3] for whom it is commonly used to guide management [
4,
5]. However, in current clinical practice, oxygen saturation is measured by invasive catheterization (cath). Non-invasive imaging methods to measure oxygenation in the heart are limited, prohibiting non-invasive quantification of cardiac blood pool oxygenation as part of routine clinical evaluation [
6‐
13]. Given the fact that invasive catheterization entails procedural risks and can be challenging in critically ill patients [
14‐
16], a non-invasive imaging method to accurately measure cardiac oxygenation would be of substantial clinical utility.
Quantitative susceptibility mapping (QSM) is an emerging cardiovascular magnetic resonance (CMR) technique that enables quantification of diamagnetic and paramagnetic materials [
17‐
23]. Iron is a magnetically active element contained in hemoglobin that is central to oxygen transport - it is weakly diamagnetic when bound to oxygen, and paramagnetic when deoxygenated [
24]. This change in magnetic susceptibility by deoxyheme [
25,
26], provides a metric by which QSM can measure blood oxygen saturation. Prior work has validated QSM tissue characterization, including liver and brain iron content [
22,
27‐
29]. Regarding blood oxygenation, a pilot study by our group showed cardiac QSM to be feasible in healthy subjects [
30]. However, a 2D acquisition strategy was employed, which is suboptimal for imaging patients in whom breath-holding is often compromised. To address this, a free-breathing 3D QSM approach was developed that uses diaphragmatic navigator gating to track respiratory position. This study compared 3D free-breathing QSM (3D
NAVQSM) to 2D breath held QSM (2D
BHQSM) in controls, as well as to the reference of ΔSaO
2 measured in patients undergoing invasive cardiac catheterization.
Discussion
This is the first study to test the free breathing 3D cardiac QSM for non-invasive measurement of blood oxygenation, inclusive of validation data provided by cine-CMR quantified cardiac remodeling and invasively quantified oxygen saturation in cardiac patients. Key findings are as follows: first, among a normative test cohort, 3DNAVQSM had a greater success for interpretable results than did 2DBHQSM (100% vs. 60%) and did so within shorter scan time. Second, 3DNAVQSM performed robustly in a subsequent cohort of 39 patients with established cardiovascular disease, among whom results demonstrated differential LV/RV susceptibility in 87% (34/39) of cases: magnitude of ΔSaO2 differed in relation to LV systolic dysfunction as quantified on cine-CMR, as evidenced by greater ΔSaO2 on QSM among patients with impaired LVEF (< 50%) compared to those with preserved LVEF, and similar results when QSM results were compared in relation to decreased LV function as stratified based on stroke volume. Third, among a subgroup of patients undergoing invasive catheterization, 3DNAVQSM yielded good correlation with invasively quantified ΔSaO2, corresponding to reasonably small bias and limits of agreement (− 0.1 and ± 8.6%, respectively) between approaches.
While our data validate cardiac QSM for differential chamber oxygenation, it is important to recognize that prior research has applied different CMR approaches for this purpose. Most previous approaches are based on measurement of blood CMR relaxation times (T2, T2*, and T1) [
6‐
10,
12,
13,
42]. However, conventional methods based on longitudinal (T1) or transverse (T2) relaxation properties can be challenging to apply clinically. For example, the dependence of spin echo T2 on oxygenation is well understood, but in practice requires measuring several model parameters in addition to oxygenation, thereby potentially limiting accuracy or complicating clinical implementation. Recently, an oxygen saturation measurement based on acquiring multiple T2 maps using a 2D T2 prepared bSSFP sequence designed to overcome these limitations was shown to provide good agreement with invasive catheterization based measurement in an animal study. A comparison between this promising approach and our proposed QSM approach is warranted in a future study [
11]. An alternative approach consists of quantifying the magnetic susceptibility of blood. The physical model relating blood susceptibility to oxygen saturation is simpler than that for T2 as it is linear with the slope a known physical constant. The magnetic susceptibility of venous blood can be computed from the CMR image phase by geometric modeling [
43‐
47]. Mapping of magnetic susceptibility throughout the 3D field of view, as is done in QSM, enables measurement of oxygen saturation of any vascular structure (including the heart) by simple ROI analysis [
25,
26,
48]. Our current data extends on prior work by our group that has shown QSM to provide an index of hemorrhage [
26,
28], as well as an index of metabolism and oxygen utilization in the brain [
49,
50].
Our QSM results regarding differential LV and RV blood oxygen saturation are consistent with values reported in prior literature as well as expected differences between subjects with and without cardiovascular disease. Regarding control data, ΔSaO
2 measured from 3D
NAVQSM (17.5 ± 3.1%) was in agreement with a prior study that reported ΔSaO
2 in healthy subjects undergoing invasive cardiac catheterization [
51], in which a mean difference of 18.8% was reported (arterial: 97.3%, venous: 78.5%). ΔSaO
2 as measured in controls were also lower than that in patients with cardiovascular disease (17.5 ± 3.1% vs. 24.9 ± 6.1%,
p < 0.001), consistent with expected physiological differences between the two groups. Among our subgroup of patients with cath validation (
n = 15), QSM derived ΔSaO
2 demonstrated a linear relationship with invasive measurements, and bias between the two oxygenation measurement approaches was small. While our data cannot be interpreted in context of an exact partition value with which to guide clinical decision making, it should be noted that an array of studies have demonstrated prognosis to vary in relation to magnitude of differential oxygen saturations between the left and right heart [
1‐
5]. In this context, our findings suggest promise for development of QSM derived ΔSaO
2 as a non-invasive risk stratification tool, such that clinically stable patients with normal values would be screened out as low risk, whereas those with elevated QSM derived ΔSaO
2 are referred for confirmatory invasive testing.
Note that, in this work, QSM was performed during non-contrast imaging in healthy subjects, and at the end of contrast-enhanced exams (~ 30 min post gadolinium infusion) in patients. Whereas gadolinium changes blood susceptibility, prior data has shown gadolinium to be near completely cleared from myocardium, and to be well mixed in the intravascular space at our imaging time point [
52,
53]. Accordingly, QSM calculations were based on the premise that contrast blood pool concentration was constant across cardiac chambers, such that gadolinium contributions to susceptibility in the LV and RV cancel out when computing differential oxygenation. Indeed, our data demonstrate that cardiac QSM derived ΔSaO
2 measurements agree well between pre- and post-contrast.
One key technical innovation in our current study concerns use of navigator technology for free breathing 3D QSM. The conventional cardiac 2DBHQSM approach generates data by imaging LV (short axis) slices individually via sequential breath-holds in order to reconstruct a 3D field map. When one or more of these breath-holds is acquired at a different respiratory position than the others, the resulting field map will not be a true 3D volume, and the QSM map will contain artifacts as the reconstruction model assumes a continuous 3D dataset as input. The susceptibility inversion problem is inherently 3D, given that susceptibility changes within one region of a given structure (e.g. RV blood pool) affects the field in the surrounding areas in all three (x, y, z) spatial directions: this is because the field is a 3D convolution of the underlying 3D susceptibility distribution with the dipole kernel. The 3DNAVQSM approach has no inherent mis-registration limitation, and the success of 3DNAVQSM acquisition depends on navigator accuracy in motion tracking. These aspects of QSM reconstruction may explain our seemingly discordant finding regarding GRE image quality and diagnostic performance of the two QSM pulse sequences tested in our study: even though GRE images from the 2DBHQSM sequence were assigned higher image scores than images from the 3DNAVQSM sequence, 2DBHQSM failed to generate interpretable QSM more often than did 3DNAVQSM and this failure was primarily attributable to slice mis-registration.
There are several limitations of our study. First, scan time of 3D
NAVQSM sequence is long (~ 4–7 min) compared to other routine clinical sequences. One potential approach to shorten scan time is to increase parallel acceleration factor, use compressed sensing [
54,
55], and/or apply data acquisition strategies such as echo planar readout [
56] or echo sharing methods [
57]. Non-Cartesian acquisitions that allow for self-gating [
58] and multi-phase reconstruction [
59] are also alternatives to shorten cardiac QSM. A second limitation to the current 3D
NAVQSM approach is that the respiratory motion is tracked by a diaphragmatic navigator, which is known to occasionally fail [
60]. More advanced navigator techniques such as a fat navigator can be used to improve success rates [
31,
32,
61‐
63]. A third issue to consider is that the QSM validation in this study was derived from cardiac remodeling indices in 39 cardiac patients of whom 15 had invasive catheterization for evaluation of known or suspected heart failure; further validation in a larger clinical cohort including patients undergoing catheterization for different indications is warranted. Finally, whereas the interval between QSM and invasive catheterization was relatively short (mean 4 ± 3 days, median 2 days [IQR 1–5 days]), cardiac chamber oxygenation could have varied during this time, thus resulting in discordance between tests. Future research with simultaneous invasive and CMR oxygenation measurements in animal models could be of utility for further validation of QSM, as well as comparisons with alternative pulse sequences for oxygenation quantification (i.e. T2). Nevertheless, our current findings that 3D navigator QSM was feasible among a clinical cohort in whom it generally agreed with differential oxygenation saturation as measured invasively is of substantial importance with respect to translational application of this technique, and provides initial proof of concept with respect to clinical implementation.
Heart rate could have impacted our QSM results. This has been shown to be the case for oxygenation methods such as BOLD, which relies on the magnitude of the CMR signal. On the other hand, QSM, which relies on the phase of the CMR signal, is expected to be relatively insensitive to heart rate because the contributions to the phase (field) induced by the difference in LV and RV magnetic susceptibility (oxygenation) are not affected by heart rate. Further research is warranted to specifically assess physiologic factors impacting QSM, as well as to validate this pulse sequence in a larger clinical cohort including patients undergoing catheterization for different indications.
In conclusion, we provide validation of free breathing 3D cardiac QSM as an index of cine-CMR evidenced LV dysfunction and differential LV/RV oxygen saturation as measured by invasive catheterization. Future research is necessary to test accelerated free-breathing QSM strategies, refine cardiac QSM for myocardial tissue characterization, and validate QSM-derived blood oxygenation for non-invasive stratification of heart failure symptoms and prognostic outcomes.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.