Splenic T1-mapping: a novel quantitative method for assessing adenosine stress adequacy for cardiovascular magnetic resonance

To establish the relationship between T1_spleen and splenic perfusion/switch-off, retrospective analysis was performed on CMR scans of 212 subjects; 104 subjects had CMR at 1.5 T (Magnetom Avanto, Siemens Healthcare, Erlangen, Germany) and 108 subjects had CMR at 3 T (Magnetom Trio a Tim system, Siemens Healthcare, Erlangen, Germany). The 1.5 T population had 75 patients (n = 36 known coronary artery disease [CAD], n = 39 Atrial Fibrillation [AF]) and 29 healthy controls; the 3 T population had 86 patients (n = 22 known CAD, n = 23 Type II Diabetes Mellitus [DM], n = 21 Severe Aortic Stenosis [AS], n = 20 Hypertrophic Cardiomyopathy [HCM]) and 22 healthy controls. Healthy controls had no history of cardiovascular disease, were not on regular medications, and had normal electrocardiograms.

All subjects avoided adenosine antagonizers (e.g. caffeine) for ≥24 h before CMR. T1-mapping was performed using the Shortened Modified Look-Locker Inversion recovery (ShMOLLI) prototype sequence (WIP 561 and 448C) with inline map generation, which uses 9-heartbeats breath-holds per T1-map acquisition and enables on-screen image reconstruction within 10 s [14].

Native T1-maps were acquired at rest and during peak adenosine stress (140 μg/kg/min, 4 min, IV) in short-axis (basal, mid-ventricular, apical) slices, followed immediately by first-pass perfusion imaging on matching slices during peak stress, with an IV bolus of GBCA (0.03 mmol/kg at 6 ml/s; Dotarem, Guerbet, Villepinte, France) and saline flush (15 ml at 6 ml/s) [15, 16]. Matching rest perfusion images were acquired >15 min after stress perfusion and adenosine discontinuation to allow sufficient time for contrast washout [15, 16].

Separate data files containing all T1-maps were created and anonymized before analysis by an observer (>3 years of T1-mapping analysis experience) blinded to perfusion images and clinical information. T1-maps were excluded from analysis if the spleen was not clearly visible (2%), had respiratory-motion artefacts on raw Inversion-Recovery-weighted images (3%) or had suboptimal goodness-of-fit R²-maps (2%) [17, 20]. Overall, 738 T1-maps were included in final analysis, using dedicated in-house software MC-ROI (programmed by S.K.P. in IDL, version 6.1, Exelis Visual Information Solutions, Boulder, Colorado) [14‐18, 20]. To estimate mean native T1_spleen, regions of interest (ROIs) were manually placed on T1-maps to include as much splenic tissue as possible, avoiding partial volume effects from large splenic blood vessels and borders with neighbouring tissues (Fig. 1). ROIs were quality checked against corresponding Inversion-Recovery-weighted images and R²-maps. To derive thresholds suitable for direct application on the CMR console, splenic T1-reactivity to adenosine stress (ΔT1_spleen) was expressed in absolute terms: ΔT1_spleen (ms) = StressT1_spleen – RestT1_spleen.

Inter-slice variability in resting T1_spleen, stress T1_spleen and ΔT1_spleen were assessed in cases where matching stress and rest T1-maps were performed in ≥2 different short-axis slice positions. To assess for intra-slice T1_spleen variability, we re-analyzed healthy-volunteers data from the original ShMOLLI methods paper, where T1-maps were repeated >15 min apart in the same short-axis slice within the same scan [14]. Intra-scan variability was calculated as the standard deviation of differences-from-the-mean in each individual.

Splenic first-pass perfusion was analysed by an observer (>4 years of perfusion imaging analysis experience) blinded to T1-maps and clinical information, using CMR⁴² software (Circle Cardiovascular Imaging Inc., Calgary, Canada). Splenic ROIs were placed on stress and rest perfusion images with frame-by-frame manual correction for artefacts and respiratory motion, to generate curves showing mean splenic signal intensity (SI, arbitrary units) changes over time (50–60s). Peak splenic perfusion SI (SI_spleen) was estimated as the numerical difference between baseline-SI and maximal-SI during splenic first-pass perfusion as previously described [6]. Adenosine-induced changes in SI_spleen compared to rest were expressed in percentages: ΔSI_spleen (%) = (StressSI_spleen – RestSI_spleen) ÷ RestSI_spleen × 100%.

Splenic switch-off on perfusion imaging was visually assessed by 2 independent observers (>3 years clinical CMR perfusion experience). In the 5/212 cases where the 2 observers disagreed, adjudication was sought from a 3^rd independent observer (Fig. 1). Perfusion images were graded as previously described [6]: either displaying splenic switch-off (the spleen on rest imaging is clearly brighter than on stress imaging), or no switch-off (the spleen on rest imaging is of similar brightness compared to stress imaging).

Data are reported as mean ± SD, tests are 2-tailed and parametric, based on Kolmogorov-Smirnov normality-checks. Differences in individual characteristics were tested using Student’s t-tests, paired within individuals (e.g. stress vs rest T1_spleen) and unpaired between groups (e.g. ∆T1_spleen in controls vs patients). Comparisons between ≥3 data groups were assessed using analysis of variance (ANOVA) with Bonferroni-corrected post-hoc method. Linear correlations were assessed using Pearson’s correlation coefficient (R) and non-linear correlations were assessed using Spearman’s rank correlation coefficient (rho). Intra-scan variability and inter-observer reproducibility of rest/stress T1_spleen and ΔT1_spleen were assessed using the Intra-class correlation coefficient (ICC), reporting 95% confidence intervals. The performance of ΔT1_spleen for replicating splenic switch-off was assessed using receiver-operating characteristics (ROC) curves [21], reporting area-under-the-curve (AUC ± SEM), and also sensitivity, specificity, diagnostic accuracy, positive predictive values (PPV) and negative predictive values (NPV), with 95% confidence intervals (CI). All analyses were performed on single measures per-subject, using MedCalc 12.7.8 (MedCalc Software, Ostend, Belgium). P < 0.05 denotes statistical significance.

Subject characteristics are summarised in Table 1. All subjects experienced at least one adenosine-related symptoms (e.g. chest-tightness, dyspnoea, flushing) [13], and >10 bpm increase in heart rate (HR) during adenosine stress, compared to rest. Significant blood pressure response (>10 mmHg SBP decrease during stress) was observed in 50% of subjects.

Mean stress HR was lower in 1.5 T patients compared to other subjects, despite similar resting HR, likely due to more frequent beta-blocker and non-dihydropyridine calcium channel antagonist administration in these patients (all AF/CAD, Table 1).

In healthy controls, mean resting T1_spleen values were 1102 ± 66 ms (1.5 T) and 1352 ± 114 ms (3 T), which decreased significantly during adenosine stress at 1.5 T (ΔT1_spleen: −40 ± 25 ms, p < 0.0001) and 3 T (ΔT1_spleen: −43 ± 31 ms, p < 0.0001). Patients had slightly lower resting T1_spleen compared to controls at 1.5 T (1083 ± 59 ms vs. 1102 ± 66 ms, p = 0.04), and this pattern was more pronounced at 3 T (1295 ± 105 ms vs. 1352 ± 114 ms, p = 0.01). Despite these observed resting T1_spleen differences, ΔT1_spleen was comparable between patients and controls, at 1.5 T (−44 ± 21 ms vs. −40 ± 25 ms, p = 0.43) and 3 T (−44 ± 26 ms vs. −43 ± 31 ms, p = 0.93; Table 2). In controls, there was a strong correlation between stress ΔT1_spleen (mean −4.1 ± 1.5%) and ΔT1_myocardium (mean 5.9 ± 1.8%), r = −0.72, p < 0.001. See Additional file 1 for more details.

In pooled analysis, ΔT1_spleen did not appear to be significantly affected by field strength (1.5 T vs. 3 T: −43 ± 22 ms vs. −42 ± 27 ms, p = 0.89), gender (male vs. female: −40 ± 23 ms vs. −47 ± 28 ms, p = 0.09), age (R = 0.10, p = 0.14, range 21–89 years) or the type of cardiovascular diseases (1.5 T CAD −42 ± 20 ms vs. 3 T CAD −40 ± 25 ms vs. AF −46 ± 22 ms vs. HCM −43 ± 28 ms vs. AS −43 ± 21 ms vs. DM −43 ± 32 ms, p = 0.54). In addition, ΔT1_spleen was not significantly affected by medication in patients (supplementary material in Additional file 2).

Inter-slice intra-scan variability (assessable in 96 subjects) was within ±19 ms for resting T1_spleen, ±18 ms for stress T1_spleen, and ±10 ms for ΔT1_spleen. Re-analysis of the original ShMOLLI cohort (spleen visible in 9/10 cases), revealed an intra-slice intra-scan repeat variability of T1_spleen of ±9 ms, ICC: 0.98 (95% confidence interval 0.93 to 0.99) [14]. ΔT1_spleen was derived by a second independent blinded observer in 45 subjects (20 controls; 25 patients: 5 CAD, 5 AF, 5 DM, 5 AS, 5 HCM), which yielded an ICC of 0.87 (95% confidence interval: 0.76 to 0.93). The Bland-Altman plot for inter-observer variability is shown in supplementary material (Additional file 3).

By semi-quantitative analysis, peak splenic perfusion SI (SI_spleen) decreased significantly with adenosine stress compared to rest, with no differences between controls and patients, or across field strengths (Table 3). ΔSI_spleen correlated strongly with ΔT1_spleen (rho = 0.70, p < 0.0001, Fig. 2). In contrast, ΔSI_spleen and ΔT1_spleen did not demonstrate significant correlations with stress-induced changes in RPP (R = 0.04, p = 0.60; R = 0.06, p = 0.38, respectively).

Subjects with visual splenic switch-off had greater stress ΔSI_spleen and ΔT1_spleen values compared to those with no switch-off (Table 4 and Fig. 3). In contrast, there were no significant differences in stress-related haemodynamic changes (HR, SBP, RPP) between subjects with splenic switch-off and no switch-off (Table 4 and Fig. 3).

ROC analysis using visual splenic switch-off as reference standard (true positives = splenic switch-off, true negatives = no switch-off) yielded an AUC of 0.90 ± 0.05 (p < 0.0001, Fig. 4). A ΔT1_spleen threshold of ≥ −30 ms during adenosine stress replicated visual splenic switch-off with a sensitivity of 90% (95% CI: 85–94%, p < 0.0001), specificity 88% (95% CI: 62–98%, p < 0.0001), diagnostic accuracy 90% (95% CI: 84–96%, p < 0.0001), PPV 98% (95% CI: 96–100%, p < 0.0001) and NPV 42% (95% CI: 26–61%, p < 0.0001).

Patients had lower resting native T1_spleen values compared to controls. This may be related to the presence of co-morbidities in patients, such as hypertension and peripheral vascular disease, which may induce peripheral vasoconstriction, with expected reductions in resting organ blood volumes and T1_spleen values. This observation deserves further investigation in larger future studies. Native T1-relaxation times of tissues are prolonged by increased blood volume (i.e. water content) [14, 15, 22]. Adenosine causes splenic artery vasoconstriction and reduced blood volume [6‐11], which shortens splenic T1-relaxation times. This is supported by our observation of significantly lower T1_spleen during adenosine stress compared to rest, in both controls and patients. The stress ΔT1_spleen was not significantly affected by different field strengths, age, gender and cardiovascular diseases, likely reflecting reproducible T1-estimations in this study [14, 15, 22].

The correlation between stress ΔT1_spleen and ΔT1_myocardium in normal controls suggests the vasoconstrictor effect of adenosine on the spleen is associated with vasodilatory effects in the myocardium. For the relationship between myocardial and splenic stress T1 in patients with cardiovascular disease, larger ongoing studies will offer reference ranges for ΔT1 in disease, and resolve the separate effects of regional myocardial differences and medication on stress T1 reactivity.

The observed strong correlation between ΔT1_spleen and ΔSI_spleen suggests that stress-induced changes in splenic blood volume are related to blood flow, which is regulated by alterations in the adenosine-mediated splenic arterial tone [10, 11]. The lack of significant correlation between ΔSI_spleen or ΔT1_spleen with rate pressure product is consistent with existing evidence showing dissociation between imaging and hemodynamic markers of stress response [5, 6], and further suggests that stress responses during clinical CMR cannot be reliably assessed using hemodynamic observations alone [5]. This deserves further investigation.

A threshold of ≥30 ms decrease in T1_spleen replicated complete splenic switch-off with a high positive predictive value of 98%. The intra-scan variability in T1_spleen (inter-slice: ±10 ms; intra-slice: ±9 ms) was 3-times less than this proposed threshold ≥30 ms drop, with excellent T1-fit as evident on quality control R²-maps, despite the lack of dedicated image optimization (e.g. shimming) over the spleen. For stress T1_spleen responses <30 ms, further work is needed to determine whether adenosine dose-increments or waiting longer with the same infusion rate may improve the confidence of stress responses, and impact on diagnostic performance of stress CMR for the diagnosis of ischemia.

This proof-of-concept study is based on ShMOLLI T1_spleen values derived from short-axis slices planned for myocardial perfusion CMR imaging; the spleen was not visible in a small proportion of T1-maps (~2%), and future applications of splenic T1-mapping may benefit from a dedicated image planned through the spleen. Rapid on-scanner T1-map reconstructions, with the immediate availability of goodness-of-fit measures (such as R²-maps), are imperative to enable practical “on-the-fly” repetition of reliable T1_spleen estimations to guide stress response optimization (Fig. 5). Given the overall excellent R²-maps over the spleen and the narrow T1_spleen ranges obtained, data in this study suggest that stress/rest splenic T1-mapping can be feasibly included in CMR protocols without major technical adjustments. Practical in-vivo T1-estimations are method-dependent, and demonstrate increasingly discrepant heart rate dependencies at longer T1-values [23]. Therefore, results achieved with ShMOLLI, in particular the splenic T1-thresholds replicating splenic switch-off, should be interpreted with care before directly translating to other T1-mapping techniques. Choosing methods that can withstand dynamic HR-variations and tachycardia without significant HR-dependencies is therefore paramount when performing stress-T1 studies. The gadolinium-based splenic switch-off sign is only seen with non-selective adenosine receptor agonists (dipyridamole and adenosine), but was absent with cardio-selective vasodilators (e.g. regadenoson) or inotropic agents (e.g. dobutamine) [6]. Further work is needed to elucidate stress T1_spleen responses using pharmacological agents other than adenosine and during physical exercise. Patients in this study were unselected for diseases known to affect splenic blood volumes, e.g. venous portal hypertension, hematological malignancies and systemic inflammation; thus, further studies to characterize the effects of these diseases on T1_spleen will help to determine the general applicability of this technique. While we identified a cut-off of ≥30 ms drop in T1_spleen during stress for replicating complete splenic switch-off, the clinical utility of this threshold for detecting true stress adequacy needs to be validated against false-negative perfusion scans, determined by comparison to invasive coronary angiography and pressure-wire based assessments of functional ischemia, such as fractional flow reserve. This is topic of ongoing work.