Native T1 mapping detects both acute clinical rejection and graft dysfunction in pediatric heart transplant patients

In this IRB-approved study, with consent/assent as appropriate, heart transplant patients referred for clinically-indicated EMB were prospectively enrolled to undergo noncontrast-CMR at 1.5 T (MAGNETOM Aera, Siemens Healthineers, Erlangen, Germany) followed by cardiac catheterization, with EMB in an adjoining biplane fluoroscopy suite. CMR included right heart catheterization, standard volumetry cines, phase contrast imaging, and breath-held native T1 mapping using Modified Look-Locker Inversion recovery (MOLLI) in eight short-axis slices. The following parameters were used for MOLLI T1 map acquisition: field of view 360 mm × 307 mm, percent phase field of view 50–80% based on patient’s body habitus, and matrix size 1.4 × 1.4 mm. For patients with a RR interval > 700, echo time (TE) was 1.12 ms, repetition time (TR) was 2.80 ms, and flip angle of 35 degrees; for patients with RR interval < 700, TE was 1.06 ms, TR was 1.93 ms, and flip angle of 70 degrees. Slice thickness and slice skip also varied based on patient’s size, ranging from 4–8 mm to 0–2 mm respectively. Measurements of volume, function, and cardiac output were performed using standard offline software (Medis Medical Imaging, AJ Leiden, Netherlands). Catheterization included collection of standard oximetry and hemodynamic data and EMB. If clinically indicated coronary angiography was performed according to surveillance standard.

Clinical data were recorded, including patient demographics, transplant history, rejection history, and serum brain natriuretic peptide (BNP) levels. Hemodynamic measurements from echocardiography (left ventricular (LV) ejection fraction, mitral E/e’), catheterization (right atrial mean pressure, right ventricle (RV) systolic pressure, RV end diastolic pressure (RVEDP), main pulmonary artery mean pressure, and pulmonary capillary wedge pressure), and CMR (LV and RV end diastolic volume) were also included. Serum BNP and hemodynamic measurements listed above were considered markers of graft dysfunction for the purpose of analysis.

Evidence of cellular rejection and antibody mediated rejection on EMB were graded based on the International Society of Heart and Lung Transplantation (ISHLT) guidelines [36] by a surgical pathologist following standard clinical practices. Clinically, rejection was defined based on treatment plan created by transplant team following the catheterization/EMB procedure. Of note, CMR T1 measurements and analyses did not affect treatment decision, as these are not reported with standard clinical data at our institution. Per our institutional protocol, those cases with Grade 2R or above acute cellular rejection on biopsy received new rejection treatment including intravenous steroids, thymoglobulin, immunoglobulins, etc.. Those cases with Grade 0R or 1R acute cellular rejection on biopsy but no hemodynamic compromise did not receive any treatment or modification to immunosuppressive therapy. Cases with Grade 0R or 1R rejection, with hemodynamic compromise, on echocardiogram, catheterization, or CMR, received treatment per transplant team using other data including biopsy, history of rejection, clinical symptoms, and donor specific antibody (DSA) results, without impact of CMR T1 mapping data. Cases were divided into one of three outcome groups (A, B, C), based on their degree of rejection therapy recommended by their physicians, who were blinded to T1 mapping data. Group A included cases with no rejection and thus no changes made to their treatment regimen. Group B included cases with some evidence of rejection which required augmentation of maintenance treatment regimen or initiation of oral steroids under pulse dosing. Group C included cases with significant evidence of acute rejection, who received new rejection treatment such as intravenous (IV) or oral steroids at pulse dosing, thymoglobulin, or intravenous immunoglobulin (IVIG).

Native T1 parametric maps were deidentified and analyzed using additional offline software (OsiriX, Bernex, Switzerland). The middle 6 slices were used (2 basal, 2 mid, and 2 apical slices), to minimize through-plane motion artifacts associated with the most basal and apical slices. Based on the American Heart Association 17-segment model for myocardial segmentation [37], 16 total regions of interest (ROIs) were generated, as demonstrated in Fig. 1. Per lab standard [26], ROIs were traced in the septal and lateral walls only to avoid partial volume effect with fat and lungs in the anterior wall and partial volume effect with the diaphragm and stomach in the inferior wall commonly found in the pediatric population. ROIs were traced by a blinded reviewer using the “middle-third” technique to avoid artifacts and blood pool at the endocardial border, per lab standard [26, 38], yielding 16 segmental average regional voxel T1 values. Global mean and peak native T1 values were quantified using these 16 segmental T1 values.

A second reviewer traced ROIs on 20% of cases (10 cases) to measure interobserver variability.

Given the myriad of factors contribution to local variation in T1 values, our institution created and maintains a local normal T1 database [16], which has both general norms (900–1050 ms) and normal values per body surface area (BSA) quartile. Per internal local studies on healthy control patients, normal T1 values were considered 900–1050 ms, with values above 1050 ms considered to be abnormal (Additional file 1: Table S2).

LV myocardium was semiautomatically segmented using Otsu thresholding from T1 maps with offline software (Seg3D, Scientific Computing and Imaging Institute, Salt Lake City, Utah, USA). Blood pool and epicardial myocardium were carefully excluded. Image texture analysis (Fig. 2) was performed on a single mid-ventricular slice, using a previously-described texture toolbox [39] in MATLAB (The MathWorks Inc., Natick, Massachusetts, USA). Segmented maps were normalized to 256 intensity levels. A 256 × 256 Gy-level co-occurrence matrix (GLCM) was created, quantifying the frequency with which a pixel of a given intensity neighbors a pixel of another given pixel intensity. The GLCM is constructed such that each entry (i, j) in the GLCM quantifies the number of times that a pixel of intensity i neighbors a pixel of intensity j. An image with more homogeneous pixel intensity is thus more diagonally dominant than one with pixel heterogeneity. To quantify image heterogeneity, nine GLCM-based texture features were computed for each slice as previously described: [30, 40‐42] energy, contrast, entropy, homogeneity, correlation, sum average, variance, dissimilarity, and autocorrelation (Additional file 1: Table S1).

EMB specimens underwent staining with Masson’s trichrome following a standard laboratory staining protocol. Whole slide scanning was performed using Tissue Scope (Huron Technologies, Ontario, Canada). Using ImageScope (Aperio, Vista, California, USA), endocardial collagen and artifacts were manually excluded. A pixel counting algorithm was developed by setting a colorimetric threshold specific to blue stained collagen. Using this algorithm, percentage tissue fibrosis was calculated per biopsy. Surgical pathologists served as expert readers, with involvement in the creation and tuning of the percentage fibrosis tool.

Interclass correlation coefficient was used to verify interobserver variability. Student’s t-test was performed on native T1 values and parametric map texture features of cases with and without rejection, as defined by clinical rejection (Group A vs Groups B + C above) and by biopsy grade (Grade 0R vs Grade 1R or greater); ANOVA analysis was also performed across the three treatment groups. A receiver operating characteristic curve was created for mean T1 values, peak T1 values, and significant texture analysis features. Pearson correlation was performed on T1 mean and peak values against markers of graft dysfunction, which included serum BNP and hemodynamic data from echocardiography, catheterization, and CMR. All statistical analyses were performed using GraphPad Prism version 9.0.0 for Windows (GraphPad Software, San Diego, California, USA).

Twenty-four pediatric patients in 50 cases (12.2 ± 4.6 years, graft age 5.3 ± 4.1 years, 36% female) underwent study procedures for surveillance (70%), suspected rejection (4%), and follow-up of prior rejection (26%). Thirty-seven cases (74%) were in Group A (no rejection, no therapy changes), 6 cases (12%) in Group B (mild rejection, minor therapy changes), and 7 cases (14%) in Group C (rejection present, major therapy changes). For analysis purposes, groups B + C represented those cases with clinical rejection. There was no difference in mean age, graft age, or graft ischemia time between Group A and Group B + C (Table 1). LV ejection fraction on echocardiogram was lower in Group B + C than Group A (60.8 ± 3.4%vs 64.3 ± 5.2%, p = 0.009), though both were normal (Table 2). A higher percentage of cases in Group B + C, compared to Group A, were positive for DSAs (53.8% vs 20%, p = 0.051).

Similarly, a higher percentage of cases in Group B + C had biopsies with cellular rejection grade > 0R compared to Group A (84.6% vs 16.2%, p = 0.001).

Two patients in Group B + C who had Grade 0R biopsies and abnormal hemodynamics, prompted modification of immunosuppressive therapy (Table 2). Of the six cases in Group A with Grade 1R biopsies, all had normal hemodynamics on echocardiogram, catheterization, and CMR; further all six cases were receiving more frequent surveillance due to history of rejection and continued to show improvement during these cases. There were no cases of antibody mediated rejection among the 50 cases.

Intraclass correlation coefficient of 20% of cases, with regions of interest traced by 2 reviewers, was 0.829 for global mean T1 and 0.830 for peak T1.

A monotonic, increasing trend was noted in both mean and peak T1 values, with increasing degree of rejection (Figs. 3, 4), with peak T1 values in the abnormal range (T1 values greater than 1050) for Group B and C. Area under the curve (AUC) of receiver operating characteristic curve of mean T1 values was 0.746 (p = 0.007) and of peak T1 values was 0.730 (p = 0.012) (Fig. 5). Notably, ROC analysis demonstrated 100% sensitivity at peak T1 values > 1050 ms, which is the cutoff of normal vs abnormal T1 values at our institution based on internal studies.

Interestingly, T1 values did not differ between cases with Grade 0R and those with Grade 1R or higher cellular rejection [mean T1: 1018 ± 28 ms vs 1029 ± 45 ms respectively (p = 0.388); peak T1: 1067 ± 44 ms vs 1075 ± 44 ms respectively (p = 0.547)]. Of note, 2 of 13 cases (15%) in group B + C had biopsies with grade 0R rejection, were treated for clinical rejection, and both were found to have elevated peak T1 values.

Mean T1 values demonstrated moderate correlation with BNP (r = 0.59) and mean pulmonary artery pressure (r = 0.33). Peak T1 values demonstrated moderate correlation with BNP (r = 0.52), right atrial pressure (r = 0.40), mean pulmonary artery pressure (r = 0.46), RVEDP (r = 0.36), and average pulmonary capillary wedge pressure (r = 0.33). T1 values did not correlate with other hemodynamic markers (Table 3). When cases with clinical rejection were removed, leaving only those cases with possible graft dysfunction, peak T1 correlated with BNP moderately (r = 0.54) but with no other markers.

Fibrosis percentage, calculated from random EMB of the RV, did not demonstrate a difference between cases of clinical rejection (Group B + C), versus cases without rejection (Group A) (8.2% ± 3.1 vs 7.4% ± 5.2, p = 0.703). Fibrosis percentage did not correlate with hemodynamic markers listed above or with mean or peak T1 values.

Image texture analysis demonstrated that mid-ventricular short axis slices from Group A differed from groups B and C in three of nine texture features computed. Energy, a measure of image homogeneity, was higher in Group A than Groups B + C (p = 0.033). Entropy and variance, which reflect randomness and heterogeneity, were higher in Groups B + C than in Group A (p = 0.008 and p = 0.001 respectively). ROC analysis of these texture features demonstrated AUC of 0.750 (p = 0.016) for energy, 0.779 (p = 0.007) for entropy, and 0.831 (p = 0.002) for variance (Fig. 6). Other texture features did not differ between the groups.

There are limitations to this work; this is a single center study including a relatively small cohort which includes longitudinal, repeat encounters. We recognize that though this is a prospective study, the definition of clinical rejection is retrospectively based on decision to treat. However, these treatment groups are in line with the clinical treatment algorithm at our institution and therefore we find are a reliable outcome measure to compare. Also of note is that only septal and lateral segments were used for T1 value measurements. Given the nature of T1 mapping, we preferred to analyze only highly reliable data at the cost of excluding certain segments. In the pediatric population, particularly with patients as small as 15 kg, the anterior and inferior segments are subject to partial volume effect from epicardial fat and lungs for anterior segments and diaphragm and stomach for inferior segments, leading to unreliable data. Therefore, these segments were not included in the analysis, as is consistent with lab standards [26]. It is also important to note that very few patients underwent this procedure due to clinical concern for rejection, but rather for screening. Further, screening protocols and treatment protocols differ between institutions as standardized guidelines do not exist; therefore, further investigation at a multi-institutional level is required.

Despite these limitations, we find it promising that CMR serves as a noninvasive screening tool during surveillance encounters and may be used to identify those patients that may be at higher risk of rejection and therefore require further evaluation. Transplant rejection surveillance remains a multi-faceted approach, including assessment of clinical presentation, echocardiography, catheterization hemodynamics, and serum markers. Our team sought to demonstrate the possibility of a synergistic value of a combined CMR and EMB protocol for evaluating patients. In our population, two patients had Grade 0R biopsies but required treatment due to abnormal hemodynamics; these same patients demonstrated peak T1 values > 1050 ms, considered abnormal in per institutional normal values, further emphasizing the utility of CMR. ROC analysis demonstrates 100% sensitivity at peak T1 values > 1050 ms, demonstrating the possibility of a model in which patients requiring transplant rejection surveillance undergo CMR including parametric mapping, right heart catheterization, and hemodynamic measurements as initial screening. For those patients with abnormal findings, EMB would be performed and also inform treatment decision. In our analysis, if this model were used, all patients who required treatment would have underwent EMB and the appropriate treatment. Of the 37 patients who did not require treatment, 15 patients would have been saved from invasive EMB. Further, at our institution where CMR guided right heart catheterization is available and routinely performed for hemodynamic data, these patients are exposed to radiation only during coronary angiography, typically only performed at annual surveillance encounters. CMR, with its ability to perform radiation free evaluations that allow for hemodynamic assessment, particularly with the use of CMR guided right heart catheterization, and for entire myocardium assessment for fibrosis and edema, may be a promising tool in the pediatric heart transplant population. Further investigation is needed in the other ways in which CMR may serve as a useful tool in transplant screening, including the potential for guiding EMB. Preliminary work completed by our group shows parametric mapping patterns of T1/T2 elevations, or hotspots [53], in myocardial diseases such as rejection, reinforcing that these pathological changes are not uniform in nature. This has been suspected in the past given the false negative rate of random EMB [4‐6]. Evaluation of these hotspots using guided EMB may provide further insight.