Current T1 and T2 mapping techniques applied with simple thresholds cannot discriminate acute from chronic myocadial infarction on an individual patient basis: a pilot study

Background Studying T1- and T2-mapping for discrimination of acute from chronic myocardial infarction (AMI, CMI). Methods Eight patients with AMI underwent CMR at 3 T acutely and after >3 months. Imaging techniques included: T2-weighted imaging, late enhancement (LGE), T2-mapping, native and post-contrast T1-mapping. Myocardial T2- and T1-relaxation times were determined for every voxel. Abnormal voxels as defined by having T2- and T1-values beyond a predefined threshold (T2 > 50 ms, native T1 > 1250 ms and post-contrast T1 < 350 ms) were highlighted and compared with LGE as the reference. Results Abnormal T2-relaxation times were present in the voxels with AMI (=> delete acute infarction; unfortunately this is not possible in your web interface) acute infarction only in half of the subjects. Abnormal T2-values were also present in subjects with CMI, thereby matching the chronically infarcted territory in some. Abnormal native T1 times were present in voxels with AMI in 5/8 subjects, but also remote from the infarcted territory in four. In CMI, abnormal native T1 values corresponded with infarcted voxels, but were also abnormal remote from the infarcted territory. Voxels with abnormal post-contrast T1-relaxation times agreed well with LGE in AMI and CMI. Conclusions In this pilot-study, T2- and T1-mapping with simple thresholds did not facilitate the discrimination of AMI and CMI.


Background
Cardiovascular magnetic resonance enables myocardial tissue characterization by combining native and contrastenhanced techniques with differences in T 2 -and T 1weighting. Native T 2 -weighted imaging has been reported to detect myocardial edema [1,2], and T 1 -weighted late Gadolinium enhancement imaging (LGE) has been established to show necrosis and fibrosis. Recent studies demonstrated that the use of these techniques allows the differentiation of acute from chronic myocardial infarction (AMI, CMI) [3]. However, there is an ongoing controversial debate about the technical limitations and the pathophysiologic background of conventional T 2 -weighted edema imaging [4,5]. Recently, myocardial T 1 -and T 2 -mapping were introduced to quantify the T 1 -and T 2 -relaxation times, which may be superior to the semiquantitative or qualitative image assessment used with conventional T 2 -weighted imaging [6][7][8]. For patients with AMI, prolonged native T 2 -and T 1 -relaxation times as well as decreased post-contrast T 1 -relaxation times were reported for the infarcted areas [9][10][11][12]. In patients with CMI, increased native and decreased post-contrast T 1 -relaxation times were reported [12]. However, whether the utilization of T 2 -and T 1 -mapping helps to differentiate AMI and CMI in the individual patient has not been examined in detail. We hypothesized that applying T 1 -and T 2 -mapping with simple thresholds based on reference values from healthy controls will discriminate AMI and CMI on an individual patient basis.

Study population
Eight male patients (mean age 56 ± 13 years) underwent CMR within 9 ± 3 days (range 5-14 <space>days) after acute ST-segment elevation myocardial infarction and in a chronic state 139 ± 50 days (range 92-210 days) after the acute event. Repeated troponin measurements were not performed. But all patients remained clinically without onset of new cardiovascular symptoms or any cardiovascular event between both CMR scans. Patients' characteristics are described in Table 1. Note that patient two and four have the lowest release of myocardial enzymes but the lowest ejection fraction. In patient 2, this is attributable to a preexisting three-vessel disease with prior inferior infarction. In patient four, left-to-left collaterals might have compensated the cellular damage during LAD-occlusion. The results were compared to previously published T 1 -and T 2 -relaxation times in healthy controls [8].

CMR examination
All CMR examinations were performed with a 3 T MR system (Magnetom Verio, Siemens Healthcare, Erlangen, Germany). The protocol was identical for patients and healthy controls. An integrated body RF coil was employed for RF transmission and a 32-channel cardiac RF coil for signal reception if not otherwise stated. ECG was used for cardiac gating/triggering.

Cine imaging
Steady-state free-precession (SSFP) 2D cine images were obtained during repeated breath-holds in three long axes and in a stack of short axes (SAX) covering the left ventricle (LV) to assess wall motion and for cardiac chamber quantification. Imaging parameters were as reported recently [8].

T 1 -mapping
Data were acquired in basal, mid-ventricular, and apical SAX planes before and after administration of 0.2 mmol/ kg i.v. gadobutrol (Gadovist®, Bayer Healthcare Germany). The acquisition of the post-contrast T 1 -maps was started in every examination exactly 10 minutes after gadobutrol LGE imaging (LGE) LGE imaging was performed 15 min after the administration of gadobutrol in the same planes as SSFP CINE imaging using a segmented inversion-recovery gradient-echo sequence. Imaging settings were as reported recently, with an acquisition voxel size 1.4 × 1.6 × 6 mm 3 [8].

Image analysis
Defining the myocardium within the maps was done using CMR 42 (Circle Cardiovascular Imaging, Calgary, Canada) as previously described [8]. Much attention was invested to manually draw the endocardial and epicardial contours as accurate as possible to omit the inclusion of blood or epicardial fat. Based on the 95 % tolerance interval of the T 2 -and T 1relaxation times from a previous study in healthy controls [8], thresholds that discriminate normal from abnormal T 1 and T 2 were defined. The cut-off for abnormal T 2times was >50 ms, native T 1 > 1250 ms and post-contrast T 1 < 350 ms. All myocardial pixels that were abnormal based on these thresholds were automatically highlighted in color in the corresponding map. The distribution of abnormal pixels was correlated with the LGE, which was regarded as the reference for the localization and extent of the infarct.

Phantom experiments
Phantom experiments were done to evaluate the accuracy of the T 2 -and T 1 -mapping method, using the same MR scanner and coil as for the in-vivo exams. An agarbased phantom representing a range of T 1 and T 2 times was used. The T 1 -values were verified using an inversion recovery sequence with acquisition matrix 256x256, TR 15 s, 1 line/inversion, 90°FLASH readout, T 1 range 200 ms-1090 ms. T 1 values were calculated using a non-linear least square three-parameter fit. The T 2values were verified using a multi-echo spin echo (MESE) approach with matrix 256×256, TR 15 s, 1 segment, T 2 -range 16-235 ms. T 2 values were calculated using a mono-exponential least square fit. For comparison, the T 2 -and T 1 -mappings technique as described for the in-vivo-measurements were applied. All phantom studies were performed with a simulated heart rate of 60 beats per minute. The signal-to-noise ratio (SNR) was estimated as the signal intensity from a manually drawn region of interest within the corresponding compartment of the phantom and the standard deviation of the signal intensity from a region of interest in the background. For T 1 , the last image of the series was used, for T 2 the first.

Phantom experiments
The results of the phantom experiments are shown in Table 2. They show that the applied mapping techniques provide estimates of the T 1 -and T 2 -relaxation times close to the reference technique, with MOLLI underestimating the T 1 -times. This finding is in concordance with previous studies that tested the same techniques in phantom experiments [6,7].

In-vivo-measurements
All patients had evidence of AMI using LGE and T 2weighted imaging during the initial scan. Figure 1 provides images of the various imaging techniques for all subjects. Note that patient #2 had an old inferior infarction (red asterisk) but actually presented with LAD occlusion. The reference T 2 was acquired with multi-echo spin echo, the reference T 1 with inversion recovery. The measured T 2 and T 1 is based on the T 2 -and T 1 -mapping as described for the in-vivo-measurements. The estimate of the SNR stems from the first image of the T 2 -series and the last of the T 1 -series Figure 2a and b provides the T 2 -and T 1 -maps for each patient with all myocardial pixels that were beyond the predefined threshold highlighted in color. The absolute T 2 -and T 1 -relaxation times for every subject and every myocardial segment are given in Table 3.
Only in half of the subjects with AMI, abnormal T 2 -relaxation times corresponded well with the infarcted pixels as defined by LGE [subjects [1][2][3]6]. In the others [subjects 4,5,7,8], pixels with T 2 -values higher than the normal range were present, but did not match with the infarcted territory as defined clinically and by LGE. In CMI (where no increase of the T 2 -relaxation time was expected), pixels with T 2 -values higher than the normal range were present in all subjects. In some of them, these pixels matched with the chronically infarcted territory [subjects [1][2][3]6].
Similarly, abnormal native T 1 times corresponded with acutely infarcted pixels in five out of eight subjects [subjects 1, 3, 6-8] in AMI, but were also present remote from the infarcted territory [subjects [4][5][6]8]. In CMI, again abnormal native T 1 values corresponded with infarcted pixels in most of the subjects [subjects 1-3, 5-8], but native T 1 vales were also abnormal remote from the  Table 1. The red arrows in the LGE images highlight the infarct region. Note that patient #2 had an old inferior infarction (red asterisk) but actually presented with LAD occlusion. Patient five had microvascular obstruction with hemorrhage (white arrow) infarcted territory [subjects 3, 4, 6-8]. Pixels with abnormal post-contrast T 1 -relaxation times matched very well with LGE in all subjects both in AMI and CMI.

Discussion
Several reports described differences of the T 1 -and T 2relaxation times between infarcted myocardial segments and remote myocardium [10,11]. It seems easy to reproduce areas of myocardial infarction using T 2 -or T 1 -maps if the localization of myocardial infarction is known. However, if the investigator is blinded to any clinical information, the discrimination of normal from abnormal myocardium as well as acutely from chronically injured myocardium solely based on T 2 -and T 1 -maps is challenging, as demonstrated in the present case series.
The observed distribution of abnormal and normal relaxation times did not match closely with the extent of the myocardial lesion as defined by LGE, particularly in the T 2 -maps and the native T 1 -maps. In CMI, pixels with elevated T 2 -value that would indicate edema appeared in the myocardium. And both in AMI and CMI, pixels with abnormal T 1 -and T 2 -values appeared also in the remote myocardium, which is supposed to have values widely within the normal range.
This mismatch between the results of the thresholding and the expected distribution of T 2 -and T 1 -relaxation Fig. 2 a and b. Thresholds that discriminate normal from abnormal T 1 and T 2 relaxation times for every myocardial pixel were defined based on reference T 2 -and T 1 -values from healthy controls. These thresholds were applied on the T 2 -and T 1 -maps so that all abnormal pixels in the myocardium (between the red and green contour) that were outside the normal range became highlighted in a color (blue in T 2 -map, brown in T 1 -maps). The red arrows in the LGE images highlight the infarct region. Note that patient #2 had an old inferior infarction (red asterisk) but actually presented with LAD occlusion. Patient 5 had microvascular obstruction with hemorrhage (white arrow). a = subjects 1-4; b = subjects 5-8  The myocardial segments that were affected by myocardial infarction based on the LGE images are in Bold times within the myocardium may be attributed to several factors: i.) A challenge of this mapping approach is the interindividual scatter of myocardial T 2 -and T 1 -relaxation times that can be large in relation to the small difference between normal and abnormal myocardium [8]: The normal T 1 -value in healthy controls in the midventricular portion of the LV was reported to range from 1005 ms to 1296 ms at our institution and other sites report mean values of 1169 ± 73 ms mixed standard deviation [8,13]. On the other hand, some authors reported a mean native T 1value of 1257 ± 97 ms for acutely infarcted segments compared to 1196 ± 56 ms for normal unaffected segments in the same subject [10]. Similarly, the normal T 2 -values in healthy controls ranged from 38 to 59 ms at 3 T and 46 to 69 ms in a study at 1.5 T [8,14]. On the other hand, a T 2 value of 60 ms has been reported as an adequate cutoff to determine active myocarditis [15]. Therefore, using one simple threshold based on normal values with a large scatter is probably imperfect for the individual subject. For example, in a person with T 1 -values in the lower range of the reference, the T 1 -values may still be within the normal range even after an infarct-related increase of 150 ms. This concept certainly needs further analysis in studies with larger samples. The large normal range is presumably attributable to physiological variations as well as to many other influencing factors like a potential heart-rate dependency of some acquisition methods and partial volume effects as outlined below. A detailed description of factors influencing the precision and accuracy of T 1measurements is available elsewhere [16]. In future, new ways of image post-processing may correct for some of these influencing factors, as recently demonstrated by Xanthis et al. [17]. An alternative approach to analyze maps -instead of defining thresholds based on the relaxation time -maybe the analysis based on the signal intensity. Kali et al. recently studied the use of native T 1 -maps in CMI. CMI was defined as using the mean ±5 standard deviation criterion relative to the respective reference regions of interest. Using this approach, native T 1 -maps and LGE images showed a close agreement to determine regions with CMI [18]. ii.) Another aspect is the influence of partial volume. Pixels that include blood or epicardial fat quickly reach pathologic T 2 -and T 1 -values and are misleadingly classified as abnormal. Even though all attempts were made to minimize this error by drawing the contours exactly within the compact myocardium, a significant influence of partial volume effects still has to be assumed. Higher spatial resolution may solve this problem in the future, and single pixels with abnormal values located at the edge of the myocardium have to be interpreted with caution. iii.)In this case series microvascular obstruction was detected by LGE images in only one subject, therefore it does not explain the frequent mismatch between the thresholding and the expected distribution of abnormal T 2 -and T 1 -values. But generally, both T 2 -and T 1 -maps have been reported to be affected by microvascular obstruction leading to "hypoenhancement" within the "hyperenhanced" acute infarction [19]. Therefore, if large enough, microvascular obstruction may contribute to an inaccurate determination of the infarct area on T 2and T 1 -maps.

Limitations
The heart rate can influence T 1 -values by affecting the relaxation between the MOLLI segments. In this study, the heart rate ranged from 50 to 83 bpm. Together with the small sample size (n = 8), these may be the major factors for data overlapping. Using a different mapping sequences with less heart-rate sensitivity, such as the 5(3 s)3 MOLLI variant, could have resulted in an improved performance of T 1 -mapping [16]. A limitation of the phantom experiments is that SNR has only been estimated, because no correction for multi-element coils has been performed [20,21].

Conclusions
This pilot study demonstrated that T 2 -and T 1 -maps with a simple threshold-based analysis did not facilitate the detection of myocardial infarction and the discrimination of AMI and CMI in the individual patient. Despite all enthusiasm for myocardial mapping, improvements in the technology as well as additional concepts for discriminating normal from abnormal myocardium may be necessary.

Ethics approval and consent to participate
Informed consent was obtained from each patient and the study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the institution's human research committee (Charité Medical Faculty, EA2/077/10).

Consent for publication
Not applicable.

Availability of data and materials
The datasets supporting the conclusions of this article, including images, contours and databases, are stored on the institutional file server (smb://fs-cmrt.ecrc-berlin.com).
The data will not be shared publicly at the current stage, as this study is part of a multi-element project that is still ongoing. Of course, data will be shared on request.