T1 mapping and T2 mapping at 3T for quantifying the area-at-risk in reperfused STEMI patients

18 PPCI-treated STEMI patients were recruited over a 7-month period from one UK center. The main exclusion criteria were previous MI and standard contraindications to CMR (significant claustrophobia, severe allergy to gadolinium chelate, estimated glomerular filtration rate <30 mL/min/1.73 m², presence of ferromagnetic implants). All eligible patients provided informed written consent and the local ethics committee (London - Harrow) approved all study-related procedures. The patients recruited in this study formed part of a cohort of patients included in a recently conducted hybrid simultaneous Positron Emission Tomography/MR study [18].

Patients underwent CMR at a median of 5 (4–6) days post-PPCI using a 3T MR scanner (Biograph mMR; Siemens Healthcare, Erlangen, Germany). The MR imaging protocol included cine imaging for function, followed by T1 maps and T2 maps for AAR and Late Gadolinium Enhancement (LGE) for MI size. T1 maps and T2 maps (Works in Progress, software #699, Siemens Healthcare, Frimley, UK) were acquired as previously described [17]. For T1 maps by SSFP-based Modified Look-Locker Inversion Recovery (MOLLI) technique, imaging parameters were: repetition time = 2.6−2.7 ms; echo time = 1.0−1.1 ms; flip angle = 35°; matrix = 256x144; slice thickness 6 mm. Images were acquired at different inversion times (5-3-3 modified MOLLI protocol to reduce heart rate variability by acquiring 5 images after the first inversion pulse, followed by a 3 heartbeat pause and then acquiring the last 3 images after the second inversion pulse [19]) and registered prior to a non-linear least-square curve fitting to generate a pixel-wise coloured T1 map. For T2 maps, 3 single shot images were acquired at different T2-preparation times (0 ms, 24 ms, and 55 ms, respectively) and imaging parameters were: repetition time = 2.4 ms; echo time = 1 ms; flip angle = 70°; repetition time = 3× R-R interval; acquisition matrix 116×192; slice thickness 6 mm; field of view adjusted as per subject size. Motion correction and fitting were performed to estimate coefficients of the decay function, which were then used to estimate the T2 times. An in-built specific colour look-up table was then used to derive the coloured T2 maps.

The T1 and T2 maps were acquired to match the short-axis cines to cover the entire left ventricle. Gadolinium contrast (Gadoterate meglumine, gadolinium-DOTA, marketed as Dotarem, Guerbet S.A., Paris, France) was administered at a dose of 0.1 mmol/kg. LGE was performed using segmented two-dimensional inversion-recovery turbo fast low-angle shot LGE sequences (repetition time = 864 ms; echo time = 1.56 mm; acquisition matrix = 123×256; inversion time = 400 – 500 ms, flip angle = 20°; slice thickness 8 mm) 10–15 min after single bolus contrast agent injection and short-axis slices of the entire left ventricle were acquired to match the T1 maps and T2 maps (Fig. 1).

Quantification of LV volumes, LV mass, LV ejection fraction and MI size were performed using CVI42 software (Version 5.1.0[280], Calgary, Canada). MI size was quantified following manual delineation of the endocardium and epicardium of short axis slices and using 5 standard deviations (SD) threshold above the mean remote myocardium [20]. Transmural extent of LGE was also quantified using 100 chords for each short axis slice and averaging the mean transmural extent for each segment using the modified 16 segment American Heart Association (AHA) model [21]. Segments were assigned an LGE score of either “0” for no LGE and “1” for the presence of LGE. Mean segmental T1 and T2 values were also automatically generated for the modified 16 segment AHA model using CVI42 after manually drawing the endocardial and epicardial borders on all short axis slices. The time taken to acquire the 2 mapping sequences for each patient (LV coverage from base to apex) were recorded. AAR by T1 mapping and T2 mapping were quantified using an in-house macro written in ImageJ (Version 1.45 s, National Institute of Health, USA). The endocardium and epicardium borders of matching LV short axis T1 maps and T2 maps were manually segmented (excluding the papillary muscles) to obtain the myocardium volume. Two experienced observers performed all subsequent quantification on the pre-segmented images blindly and independently, and one of the observers performed the analysis twice, 3 months apart. The affected myocardium on the T1 maps and T2 maps were quantified using 3 analytical techniques namely manual delineation, 2SD above the mean remote normal myocardium, and the automated Otsu detection method (Otsu technique) [22]. In brief, the Otsu technique uses an algorithm to automatically divide the signal intensity histogram into normal and enhanced. An exhaustive search for values that minimize the intraclass variance between two populations of signal intensities is used to establish the threshold [23]. Analysis was performed for all the slices for each patient to obtain the “enhanced” myocardium as a percentage of the whole LV. Regions-of-interest were drawn within the AAR (avoiding areas of MVO), and within the remote myocardium to obtain representative T1 and T2 values at 3T CMR. Manual correction was performed for areas of pseudonormalization within the MI zone (corresponding to areas of MVO) and areas of hyperenhancement due to any obvious blood pool or pericardial partial voluming and off-resonance artifacts in the remote myocardium. All slices were visually assessed by the 2 experienced observers and those with significant partial voluming and susceptibility or motion artifacts overlapping with the affect myocardium were excluded by consensus. In cases of doubt or disagreement, the raw images were used to decide whether that slice was to be excluded or not as previously described by von Knobelsdorff-Brenkenhoff et al. [17].

Statistical analysis was performed using SPSS (Version 22, IBM Corporation, Illinois, USA) and MedCalc (Version 15.6.1, MedCalc Software bvba, Ostend, Belgium). Continuous data was expressed as mean ± SD or median (interquartile range). Categorical data was reported as frequencies and percentages. Both per-slice and per-patient comparison was performed. Paired student t-tests and Wilcoxon Rank sum test were performed to compare mean or median between paired groups. Pearson’s correlation coefficient expressed as its square (R²) was used to assess inter-method correlation. A linear regression analysis was performed to obtain the regression slope and its 95 % confidence interval to compare against the reference line with a slope of 1 which would represent AAR by T1 and T2 to be identical. Bland-Altman analysis was performed to assess agreement and bias detection between methods and presented as average difference ± 2SD. Inter-observer and intra-observer variability was assessed using intra-class correlation coefficient (ICC) and mean difference between interobserver and intraobserver measurements of T1 and T2 AAR ± SD. Receiver Operating Characteristic (ROC) analysis was performed to provide cut-off values for T1 and T2 for detecting acute myocardial necrosis defined by an LGE score of 1. ROC curves were compared for statistical difference using the method described by Delong et al. [24]. All statistical tests were two-tailed, and P-values of less than 0.05 were considered statistically significant.

The AAR by the 3 techniques (manual/Otsu/2SD) for T2 mapping were 31.8 ± 11.7 %, 31.6 ± 11.2 % and 38.7 ± 15 % and for T1 mapping were 32.0 ± 11.5 %, 32.3 ± 11.5 % and 38.4 ± 13.6 % respectively. The ICC for intra-observer and inter-observer variability of the 3 analytical techniques were excellent, both for T1 and T2 mapping and was highest for the Otsu technique. The 2SD technique had the largest differences both for intra-observer and inter-observer measurements for both mapping techniques. These findings are summarized in Table 2. The 2SD technique overestimated the AAR compared to manual delineation (as the reference standard) but there was no difference between Otsu and manual delineation for both T1 and T2 mapping (Fig. 3). The Otsu derived T1 and T2 AAR was therefore used for the analysis below.

On a per-slice analysis, there was an excellent correlation between the T1 mapping and T2 mapping with an R² of 0.95 and a regression slope 0.97 (95 % CI 0.93 – 1.01). There was no bias on Bland Altman analysis (mean ± 2SD: bias 0.0 ± 9.6 %) as illustrated in Fig. 4a. On a per-patient analysis, the correlation and agreement remained excellent with no bias (R² 0.95, regression slope 0.95 (95 % CI 0.84 – 1.07), bias 0.7 ± 5.1 %; Fig. 4b). The mean AAR (expressed as a % of the LV) quantified by T1 mapping was similar to that by T2 mapping (32.3 ± 11.5 % of the LV, range 6 to 52 % by T1 mapping, versus 31.6 ± 11.2 % of the LV, range 5 - 48 % by T2 mapping, P = 0.25).

The mean MI size by LGE was 18.8 ± 9.4 % of the LV, (range 2.0 - 34.0 %). Myocardial salvage (AAR subtract the MI size) was 12.8 ± 10.0 % of LV (range 0 – 42.0 %) by T2 mapping. The myocardial salvage index (myocardial salvage/AAR) was 0.40 ± 0.26 (range 0 – 0.89). There were no difference in either myocardial salvage (12.8 ± 10.0 % of LV by T2 mapping versus 13.5 ± 10.4 % of LV by T1 mapping, P = 0.25) (Fig. 3) or the myocardial salvage index (0.40 ± 0.26 by T2 mapping versus 0.39 ± 0.27 by T1 mapping, P = 0.20) between the 2 mapping techniques.

The T1 and T2 values in the AAR was significantly higher than those in the remote myocardium (T1 AAR: 1524 ± 116 ms, T1 remote myocardium: 1163 ± 78 ms, P < 0.001; T2 AAR: 72 ± 7 ms, T2 remote myocardium: 46 ± 3 ms, P < 0.001).

Both T1 and T2 mapping performed well to detect acute myocardial necrosis delineated by the presence of LGE as illustrated in the ROC curves in Fig. 5. The area under the curve (AUC) was 0.87 ± 0.02 for T1 and 0.86 ± 0.02 for T2, P = 0.96. A T1 value of > 1249 ms had a sensitivity of 83 % and specificity of 80 % and a T2 value of > 52 ms had a sensitivity of 85 % and specificity of 82 % to detect acute myocardial necrosis defined by an LGE score of 1.

We found that the Otsu thresholding method performed best with excellent inter-observer and intra-observer variability for both mapping techniques. The algorithm automatically divides a signal intensity histogram into two classes requires minimal user input compared to manual delineation, SD thresholding and full width half maximum techniques. It automatically calculates an optimal threshold [22] and has previously been shown to be more accurate and reproducible for quantifying acute MI size by CMR. [25].

Currently, it would appear that T1 and T2 mapping could be used interchangeably to assess the AAR in reperfused STEMI patients. In studies investigating post-contrast T1 and extracellular volume fraction in acute myocardial infarction patient, there is the possibility of shortening the scanning time by omitting T2 maps as the T1 maps would be available for AAR quantification. However, T1 mapping may not be suitable in acute myocardial infarction patients who also have a chronic infarct in the remote myocardium. T1 mapping CMR has recently been shown to identify chronic infarct with high diagnostic accuracy [26] in a canine model, and using this technique in these patients would require taking into account areas of chronic infarct in the remote myocardium. Our study excluded patients with previous infarct and therefore we cannot comment on the performance of T1 mapping over T2 mapping in patients with co-existing chronic infarcts. From an MR physics point of view, these two techniques are assessing different properties of the myocardium and may explain the limit of agreement of around 10 % on a per-slice comparison and of around 5 % on a per-patient comparison and more work remains to be done to establish the advantage of one mapping sequence over the other.

Certain cardioprotective therapies such as ischemic postconditioning [10] and remote ischaemic conditioning (using transient arm or leg ischaemia and reperfusion) [11, 12] has been shown not only to reduce MI size, but also the extent of edema delineated by T2 mapping and T2-weighted imaging leading to an underestimate of the AAR in reperfused STEMI patients. Whether the AAR delineated by T1 mapping is also affected by these therapeutic interventions needs to be investigated. It is currently believed that the AAR delineated by T2-weighted imaging is maximum and constant within the first week following an acute myocardial infarction [27, 28]. Whether the T1 signal remains stable for a longer period of time remains to be tested. A recent pre-clinical study using a porcine model of acute MI has suggested that myocardial edema delineated by T2 CMR may vary over the first week with 2 phases of edema [29]. Whether this is present in reperfused STEMI patients and whether it is apparent with T1 mapping is unknown.

Although, we only included a small number of patients, we did include a range of ischemic times and performed detailed slice-per-slice comparisons, including inter and intra-observer performance. Unfortunately we did not recruit any patients with a circumflex territory myocardial infarction as it would have been interesting to assess the effect of off-resonance artifacts in the lateral wall in those patients with lateral wall MI on both mapping techniques. Our study was performed at 3T and whether T1 mapping performs as well as T2 mapping in quantifying the AAR at 1.5 T in the clinical setting remains to be determined.