Quantitative cardiovascular magnetic resonance: extracellular volume, native T1 and 18F-FDG PET/CMR imaging in patients after revascularized myocardial infarction and association with markers of myocardial damage and systemic inflammation

Patients that were retrospectively enrolled for this investigation (n = 25) represent a large subgroup of a cohort from a previously published study [15], which has focused on global measures of ¹⁸F-FDG uptake and LGE and their relationship to functional outcome. All patients underwent examination on a clinical 3T PET/MRI scanner (Biograph mMR, Siemens Healthineers, Erlangen, Germany) 5 ± 1 days after myocardial infarction and subsequent, successful revascularization (Thrombolysis in Myocardial Infarction (TIMI) grade ≥ 2, average: 2.9). The study was approved by the local ethics committee, performed in agreement with the Declaration of Helsinki, and all participants gave written and informed consent. Criteria for retrospective enrollment were membership in the final study cohort reported in [15] and availability of native and post contrast T₁ data. Additionally, patients showing signs of microvascular obstruction (MVO) in LGE images were excluded from quantitative analysis, as the necessary contrast agent equilibration for ECV mapping is not attainable. Also, segmentation of MVO border zones was deemed not practicable in these cases due to the differences in spatial resolution between PET and CMR images, and quantitative values would not have been comparable to results from transmural segmentation. A detailed description of the criteria for retrospective enrollment similar to the corresponding statements in [15] is given in Table 1.

¹⁸F-FDG PET was performed with patients receiving a low-carbohydrate diet the day before imaging, followed by a 12-h fasting period in order to suppress physiological myocardial FDG uptake. 30 min before ¹⁸F-FDG injection, patients received unfractionated heparin (50 UI/kg body weight intravenously) to further suppress physiological myocardial FDG uptake [15]. A list-mode PET scan in 3D mode was started 144 ± 39 min after intravenous injection of 311 ± 72 MBq of ¹⁸F-FDG. Correction of emission data was performed for randoms, scatter, dead time and attenuation. Attenuation correction was accomplished using 2-point Dixon sequence as previously described [19]. Parts of the body truncated in the attenuation map due to the limited field-of-view were recovered from PET emission data using the maximum likelihood reconstruction of attenuation and activity (MLAA) technique [20]. For reconstruction, a 3D attenuation-weighted ordered-subsets expectation maximization iterative reconstruction algorithm (AW-OSEM 3D) was used with three iterations and 21 subsets, Gaussian smoothing at 4 mm full width at half maximum, matrix size 344 × 344, zoom 1 and a resulting spatial resolution of 5 mm. For quantitative analysis, ¹⁸F-FDG image signals were expressed as standardized uptake values based on lean body mass (SUV LBM) or - for comparison to blood markers - also as tissue-to-background ratio (TBR) normalized to blood signal taken from an LV-centric region of interest.

As part of a comprehensive resting-state CMR exam, native and post-contrast T₁ maps were acquired using a MOLLI prototype sequence in three short axis slices per patient. Acquisition (3(3)3(3)6) scheme [21]), retrospective motion correction and registration of native and post-contrast MOLLI data was performed as previously described [22]. If necessary, additional manual motion correction was applied to the T₁ image series as it was acquired in shallow breathing. After registration, ECV maps were calculated from the resulting T₁ maps using individually measured hematocrit. LGE images covering the complete left-ventricular (LV) myocardium were acquired directly following post contrast T₁ maps after a cumulative dose of 0.2 mmol/kg Gd-DTPA (Magnevist, Bayer Healthcare, Berlin, Germany).

For PET/CMR image registration, the MunichHeart software [23] was used to fuse ¹⁸F-FDG PET images with ECV and native T₁ maps. Manual alignment was performed between ungated PET data and individually motion-corrected CMR data if necessary. Manual segmentation on the basis of ECV maps was performed for the most infarct-centric of the three acquired slices, which was determined on the basis of LGE images covering the whole left ventricle. Basal and mid-ventricular slices were segmented into 32 sectors (16 for apical slices), equally spaced along a centerline between endo and-epicardial borders without additional morphological reference. This segmentation was applied to the registered ECV, native T₁ and ¹⁸F-FDG PET images of that slice. For each patient, segmentation of the chosen slice therefore resulted in 32 transmural sector values (16 for apical slices) for each of the three image signals under investigation as shown in Fig. 1. An average of the two highest sector values was taken as representing maximum insult severity for each patient and each image signal in the sense of an “imaging biopsy”. These maximum values were determined independently for each imaging method, and therefore do not necessarily refer to the exact same intra-slice location (see Fig. 1). For the quantitative analysis of remote regions, three sectors as distal as possible to the sector with the highest value were chosen manually. Therefore, for each patient, the remote value of ECV, native T1 and ¹⁸F-FDG uptake respectively refers to the average from the so-defined remote sectors, and the maximum value refers to the average of the two highest sector values. Additional measures of infarct size in % of LV volume were obtained by manual delineation of enhancement regions in the LGE images as described previously [15].

Daily blood sampling was performed for up to 6 days after revascularization. Peak levels of blood parameters creatine kinase (CK), creatine kinase-MB (CK-MB), troponin T as well as peak leukocyte and monocyte counts were taken from this sampling period as previously reported [15].

The physiological characteristics of the final patient cohort (n = 25) are shown in Table 2. A total of 42 individual segments from four patients were excluded before quantitative analysis due to small susceptibility artifacts in the T₁ maps (n = 3) or breathing artifacts in the PET attenuation map (n = 1).

The total number of sectors after exclusions was 662 for each imaging method, consisting of 32 (mid/basal) or 16 (apical) sectors per patient for each of the 25 subjects. The upper panel of Fig. 2 shows the comparisons of these sector values in a globally pooled fashion. All three comparisons (FDG/ECV, FDG/native T1, ECV/native T1) exhibited similarly significant correlations between subject means (R = 0.60, 0.43, 0.51). The lower panel of Fig. 2 shows the corresponding within-patient correlations, i.e. a linearization of within-slice signal increase for each patient (i.e. each slice) individually. Within-patient analysis yielded much higher correlation factors (R = 0.91, 0.87, 0.88) than the globally pooled comparison.

The underlying inter-patient variability of absolute values at the high end was most clearly discernible for the comparison of FDG/ECV. Consequentially, if only the average of the two maximum sector values from each patient were compared, no significant correlations were observed for FDG/ECV, FDG/native T1 nor ECV/native T1 (Fig. 3). Table 3 shows ranges and averages for maximum and remote values across all patients for each imaging method. While maximum values were determined individually for each imaging method, average distances between sectors containing these were small across modalities, i.e. on average 1.2 sectors between ECV/FDG, 2 sectors between nT₁/FDG and 1.9 sectors between nT₁/ECV.

Results for comparing infarct sizes with corresponding maximum values of ECV, ¹⁸F-FDG uptake and native T₁ are shown in Fig. 4. Infarct sizes calculated on the basis of LGE images yielded an average infarct size of 17.3 ± 7.1% of LV volume. A highly significant correlation of cellular damage as indicated by maximum ECV with infarct size was observed (p = 0.002). Conversely, no significant correlation between infarct size and either maximum ¹⁸F-FDG uptake or maximum native T₁ was observed.

Maximum ECV (Fig. 5), ¹⁸F-FDG uptake (absolute (Fig. 6) and TBR) and native T₁ (Fig. 7) were compared to peak values of peripheral blood parameters CK, CK-MB, troponin T as well as peak leukocyte and monocyte counts. Monocyte counts were unavailable for two patients, and were excluded for one additional patient due to splenectomy, resulting in 22 instead of 25 data points for Figs. 5d, 6d, 7d. For maximum ECV, a strong trend towards association with peak CK (p = 0.052) and a significant correlation with peak CK-MB (p = 0.006) were observed, despite no correlation to peak troponin or monocytes. The comparison to peak CK-MB revealed a tight relationship at the low end and a larger variability at the high end of values (Fig. 5b). For absolute ¹⁸F-FDG uptake, only a narrowly significant correlation was found with troponin (p = 0.042), and none was found for ¹⁸F-FDG TBR normalized to LV blood (see Additional File 1). Maximum native T₁ values did not show significant correlations to CK or CK-MB, but a highly significant correlation (p = 0.005) to peak monocyte counts and a significant correlation with troponin (p = 0.024). The corresponding comparisons to peak leukocyte counts are not shown, yielding a significant trend (p = 0.033) with native T₁.

The first step was the comparison of inter-patient and intra-patient (i.e. intra-slice) signals as shown in Fig. 2. The excellent correlations for the latter suggest a very good co-localization of pathophysiological processes indicated by the three different image signals. While in the vast majority of studies, extents of image signals are compared by global thresholding, e.g. using multiples of standard deviations as binary cutoff values, the intra-patient correlations shown herein are independent of differences in absolute signal increase between patients, i.e. differences seen in the correlation slopes in Fig. 2c-d. Despite the good co-localization of signal increases, the heterogeneity of these slopes and equivalently the comparison of maximum values in Fig. 3 suggest sensitivities to different underlying tissue properties. Therefore, ECV, ¹⁸F-FDG uptake and native T₁ appeared as representing mutually distinct combinations of features with respect to infarct-related pathophysiology, while the respective underlying processes seemed to be largely co-localized.

For myocardial ECV, estimates derived from pre- and post-contrast T₁ mapping have been shown to be sensitive to a number of different disease processes [5] and histologically verified to correlate with fibrosis [24]. With respect to AMI, quantitative ECV mapping has only recently been shown to be associated with functional outcome in patients [11] but further investigation of pathophysiological mechanisms are lacking. While significant association of ECV with edema has been documented at day 1 after AMI in pigs [25], the same study reported a disappearance of this association after 7 days, which suggests edema if at all as a minor contributor to ECV estimates from the study at hand obtained 5 days after AMI. A possible pathophysiological correlate for absolute ECV in this study has been provided in the form of CK/CK-MB blood markers for (myocytic) cellular damage. The significant correlation between peak CK-MB and ECV is in fact remarkable as it reflects the association of a global, peripheral blood parameter with a focal, biopsy-like imaging result. It may however be seen as an epiphenomenon to the additional finding of a highly significant correlation between ECV at the infarct center and infarct size, where peak CK-MB activity is an established marker for the latter [26]. This observed relationship between the extent of the area being subject to ischemic insult and the amount of myocytic damage at its center may be seen as somewhat mechanistically plausible, considering a decrease of the probability for remaining collateralization with distance to the nearest non-infarcted tissue regions. The much smaller significance for the corresponding correlation observed between peak CK and ECV is consistent with a lower specificity of CK to myocardial damage compared to CK-MB.

With respect to FDG, data from this study suggest no correlation of maximum ¹⁸F-FDG uptake at the infarct center with peak monocyte or leukocyte counts or global infarct size. This was irrespective of whether maximum ¹⁸F-FDG was evaluated in absolute terms as SUV LBM or normalized to LV blood activity as a TBR. A similar finding [15] has been interpreted as a conceivable disproportionality between systemic/peripheral inflammatory cell counts and the presence of migrated inflammatory cells within the myocardium begetting the imaging signal. While it is known that ¹⁸F-FDG is taken up by inflammatory cell populations [27], the interpretation of a corresponding image signal from the post-ischemic myocardium is challenging due to the concurrent presence of background contributions from myocyte uptake. Despite a somewhat reliable suppression of physiologic FDG metabolism in healthy cardiomyocytes, the potential presence of post-ischemic FDG uptake due to a switch of metabolism from fatty acids towards glucose consumption early after AMI is a major confounder to the interpretation of ¹⁸F-FDG uptake as a purely inflammatory signal [28]. Therefore, ¹⁸F-FDG image signals are generally regarded as a mixture of background/blood pool, post-ischemic and inflammatory constituents in this context.

Native myocardial T₁ may reflect a variety of pathologic tissue alterations, but is generally accepted to indicate the edematous increase of free water content early after AMI [8]. The expected increase of infarct-centric native T₁ observed in this study is therefore attributed to an edematous reaction, which, however, did not show a correlation with infarct size as ECV did. With respect to native T₁, the most interesting finding from this study is a highly significant correlation with peak monocyte counts and a weaker but still significant association to peak leukocyte counts, of which monocytes are a subset specific to inflammatory activity. As for ECV, this association of biopsy-like imaging results with peripheral blood markers is remarkable, even more so considering that native T₁ was not found to be related to infarct size. Therefore, the data at hand provides evidence for the fact that myocardial edema and the systemic inflammatory reaction post AMI are quantitatively associated.

Summarizing the comparison of imaging results with peripheral blood parameters, the data at hand suggest ECV as a marker of cellular damage early after reperfused AMI, with maximum values related to infarct size and therefore reflecting most likely a mechanistic property of the respective infarct. Conversely, the missing correlation to infarct size for maximum ¹⁸F-FDG uptake and native T₁ suggest their association with a more patient- than infarct-specific reaction to the ischemic insult. This notion is strongly supported by the observed association of native T₁ with peak monocyte counts.

An important translational limitation to results from this study is that patients exhibiting MVO post AMI were not amenable to the presented analysis. While there exist propositions on the segmentation of MVO border zones for e.g. ECV mapping [11], the difference in spatial resolution between PET and CMR precluded comparable sub-segmentations of the myocardial wall. For similar reasons, only transmural short-axis sectors were defined for LV myocardium, where especially cardiac motion hampers a meaningful sub-segmentation across the myocardial wall in PET. With respect to segmentation, locations for maximum values were determined individually for each modality because the alternative of having one of the three methods be the reference standard for locating the corresponding sectors would have introduced a bias into the comparison and additionally precluded the translation of conclusions to situations where the reference modality is not available. The small spatial differences introduced by individual determination of reference sectors did not suggest this as a significant limitation to the finding that all three modalities indicate mutually distinct processes within the tissue.

The acquisition scheme (3(3)3(3)6) used in this study for MOLLI T₁ mapping has been shown to be more sensitive to variations in heart rate than other, more recently proposed schemes [29]. However, the fact that resting heart rates in the examined cohort did not vary strongly and that heart rate did not correlate with remote native T₁ (R = − 0.22, p = 0.3) do not suggest this as a major confounder to the presented findings.

Additionally, the practice of using peak values of peripheral blood markers as a surrogate for their summed activity may have introduced additional variability into the reported results. While the described practice is known to be relatively accurate for CK/CK-MB [26], the inflammatory reaction indicated by monocyte counts may behave in a more complex fashion. With respect to statistics, the large number of performed correlation analyses may have justified the use of a 1% significance level to make type 1 errors less likely. However, the main conclusions presented herein only rely on findings with p-values < 0.01.