Accurate Intramyocardial Hemorrhage Assessment with Fast, Free-running, Cardiac Quantitative Susceptibility Mapping

Abstract

Purpose

To evaluate the performance of a high-dynamic-range quantitative susceptibility mapping (HDR-QSM) cardiac MRI technique to detect intramyocardial hemorrhage (IMH) and quantify iron content using phantom and canine models.

Materials and Methods

A free-running whole-heart HDR-QSM technique for IMH assessment was developed and evaluated in calibrated iron phantoms and 14 IMH female canine models. IMH detection and iron content quantification performance of this technique was compared with the conventional iron imaging approaches, R2*(1/T2*) maps, using measurements from ex vivo imaging as the reference standard.

Results

Phantom studies confirmed HDR-QSM’s accurate iron content quantification and artifact mitigation ability by revealing a strong linear relationship between iron concentration and QSM values (R², 0.98). In in vivo studies, HDR-QSM showed significantly improved image quality and susceptibility homogeneity in nonaffected myocardium by alleviating motion and off-resonance artifacts (HDR-QSM vs R2*: coefficient of variation, 0.31 ± 0.16 [SD] vs 0.73 ± 0.36 [P < .001]; image quality score [five-point Likert scale:], 3.58 ± 0.75 vs 2.87 ± 0.51 [P < .001]). Comparison between in vivo susceptibility maps and ex vivo measurements showed higher performance of HDR-QSM compared with R2* mapping for IMH detection (area under the receiver operating characteristic curve, 0.96 vs 0.75; P < .001) and iron content quantification (R², 0.71 vs 0.14).

Conclusion

In a canine model of IMH, the fast and free-running cardiac QSM technique accurately detected IMH and quantified intramyocardial iron content of the entire heart within 5 minutes without requiring breath holding.

Keywords: High-Dynamic-Range Quantitative Susceptibility Mapping, Myocardial Infarction, Intramyocardial Hemorrhage, MRI

Supplemental material is available for this article.

©RSNA, 2024

Summary

Free-running cardiac quantitative susceptibility mapping allows for accurate detection and quantification of intramyocardial hemorrhage in the setting of acute myocardial infarction.

Key Points

• ■ The developed free-breathing, non–electrocardiographically gated cardiac quantitative susceptibility mapping (QSM) MRI technique accurately detected intramyocardial hemorrhages without being influenced by breathing motion and cardiac arrhythmias (high-dynamic-range QSM vs R2*: area under the receiver operating characteristic curve, 0.96 vs 0.75; P < .001).

• ■ Mitigating the off-resonance artifacts, free-running cardiac QSM enabled precise regional myocardial iron content quantification of the whole heart (linear regression R², 0.71 vs 0.14).

• ■ Accurate quantitative intramyocardial hemorrhage assessment can improve patient care by facilitating accurate myocardial infarction risk stratification and has potential to guide the development of iron-targeted treatments.

Introduction

Recanalization of obstructed epicardial coronary arteries can result in intramyocardial hemorrhage (IMH), which can substantially augment reperfusion injury and contribute to infarct expansion (1). Myocardial infarction (MI) with IMH is the highest risk type of MI, classified as stage 4 acute MI (AMI) by the Canadian Cardiovascular Society (CCS) (2). Patients with this MI type have a two- to sixfold higher rate for major adverse cardiac events compared with patients with AMI without IMH (3–6). Importantly, recent studies have shown that therapies that deplete the iron within MI can significantly curb MI compositional changes and promote favorable remodeling (3). Because iron cardiotoxicity is concentration dependent (7), accurate myocardial iron quantification has high potential in longitudinal monitoring of iron content in patients with hemorrhagic MI and may facilitate the development of therapies targeting post-MI iron deposition (3,7–9).

T2*-based cardiac MRI is the current standard for IMH imaging due to its high sensitivity to iron (10–14). However, T2* cardiac MRI is known to be vulnerable to motion and off-resonance–induced artifacts. It has been shown that approximately 40% of clinical T2* cardiac MR images are affected by these artifacts (15–17), which can lead to nulled signal intensities at the heart-lung interfaces and corrupted T2* maps of the whole heart. This makes iron content assessment in patients with AMI particularly challenging (15,18). A more direct approach for quantifying iron using MRI is based on quantitative susceptibility mapping (QSM). QSM has been extensively validated and serves as the clinical standard for iron quantification in the brain (19). Because QSM utilizes the phase information at MRI and deconvolves the local field perturbation within the tissues, it allows for off-resonance correction and spatial mapping of the underlying tissues’ magnetic susceptibility (20). This provides direct measures of tissue iron concentration (6,21,22), allowing for a potential common standard for diagnosing abnormal myocardial iron content in patients with IMH. While the quantitative relationship between tissue iron content and QSM measurements has been extensively studied and validated in ex vivo hearts (4–6,10,23), its in vivo translation for hemorrhagic MI faces major technical obstacles. These include cardiac and respiratory motion–induced phase errors, particularly in patients with arrhythmic AMI who cannot perform stable breath holds, and QSM deconvolution errors caused by rapidly changing magnetic fields at the heart-lung interfaces and around the hemorrhagic focal iron deposition, making the conventional cardiac QSM unreliable for clinical AMI assessment.

Here, we developed a free-breathing, non–electrocardiographically (ECG) gated, whole-heart QSM technique with a high-dynamic-range (HDR-QSM) reconstruction algorithm to overcome the confounders for IMH detection and quantification. We first tested the method in an iron phantom with and without an off-resonance source. We then examined the IMH assessment ability of the technique in canines with surgically induced hemorrhagic MIs and compared the results against ex vivo measurements.

Materials and Methods

All animal studies were approved by the institutional animal care and use committee. Studies were performed prospectively between March 2021 and November 2022.

Free-running Whole-Heart Cardiac QSM Sequence Development and Image Reconstruction

To enable a free-running cardiac QSM for IMH assessment, a continuous ungated whole-heart multiecho gradient-recalled echo (mGRE) sequence was prescribed over the short-axis view covering the whole heart (24). The total acquisition time is less than 5 minutes. Data were acquired with randomized Gaussian-distributed Cartesian k-space sampling, interleaved with center k-space lines every other excitation to sample the respiratory and cardiac motion. A previously described low-rank tensor framework (25) was used for reconstruction. To facilitate a motion-robust QSM reconstruction, adaptive motion binning was applied to minimize motion artifacts while identifying a state with an optimized off-resonance profile.

Cardiac HDR-QSM Derivation

The HDR-QSM image processing pipeline was developed (see Fig 1) to eliminate common confounders. First, an HDR phase unwrapping approach (AUTO-HDR [26]; Fig S1) combining the temporal continuity and spatial smoothness was adopted using an iterative graph-cut–based algorithm (22). To optimize signal-to-noise ratio and extend the dynamic range of the field map with high unwrapping accuracy, the unwrapped HDR phase maps at all echoes were estimated using the following equation:

Figure 1:

High-dynamic-range quantitative susceptibility mapping (HDR-QSM) processing pipeline. The HDR-QSM imaging processing pipeline was developed for this study. First, motion-robust multiecho gradient-recalled echo (mGRE) images at end expiration and middiastole were reconstructed using a low-rank tensor framework. Phase maps of each echo were then unwrapped using the autonomous multiecho HDR phase (AUTO-HDR) unwrapping algorithm. In parallel, signal-to-noise ratio (SNR) maps were derived from the corresponding magnitude images with thresholding. Subsequently, the HDR phase maps were obtained using an SNR-weighted nonlinear least squares fitting algorithm. Finally, the QSM maps were derived using a variable-kernel sophisticated harmonic artifact reduction for phase data (V-SHARP) background removal algorithm with a multilevel QSM approach. ECG = electrocardiography, TE = echo time, 3D = three-dimensional.

where TE_i represents the i^th echo time; f_total is the fitted total frequency map describing the field inhomogeneity; S(TE_i) is the signal acquired at echo time TE_i; A(TE_i) is the amplitude of S(TE_i); and M_SNR,TEi is the signal-to-noise ratio guiding binary mask at echo time TE_i. M_SNR,TEi was obtained based on hard thresholding in all echoes (signal-to-noise ratio, 8) (27). The phase maps were then processed with variable kernel sophisticated harmonic artifact reduction for phase data (V-SHARP) (28) for background field removal, and a multistep QSM algorithm (29) was used to minimize streaking artifacts from IMH lesions with high iron concentration. A comprehensive description of the proposed sequence, image reconstruction pipeline, and HDR-QSM reconstruction is presented in Appendix S1.

Data Acquisition and Image Analysis

All imaging studies were performed with a clinical 3-T scanner (mMR Biograph; Siemens Healthcare). In controlled (phantom and ex vivo hearts) scans, images were acquired with a vendor-provided high-resolution two-dimensional mGRE sequence. For in vivo scans, the proposed motion-robust mGRE images were acquired using the developed sequence in 5 minutes without suspending respiration or ECG gating. Three-dimensional (3D) R2* (1/T2*) maps, conventional QSM maps, and HDR-QSM maps were reconstructed using the same data. In addition, standard two-dimensional (2D) breath-hold T2* and late gadolinium enhancement images were acquired using clinical sequences for comparison. The late gadolinium enhancement images in this study served as a reference to identify the affected territories in the reperfused MI and differentiate the false-positive voxels from susceptibility-induced T2* artifacts in the remote myocardium. Customized MATLAB (R2021b) scripts were developed for image reconstruction and analysis. All in vivo data were collected by specialized cardiac MRI technologists with over 10 years of experience. Image reconstruction, postprocessing, and quantitative analysis were performed by Y.H.

Phantom Study

Image acquisition.— Agar phantoms with 12 different iron concentrations, comprising the range observed in IMH lesions (0–0.167 mg/g), were made with ultrasmall superparamagnetic iron oxide particles (ferumoxytol; Schering) to validate the HDR-QSM approach. A cylindrical cup (radius, 10.0 cm) filled with air served as an off-resonance source and was placed at the center of the phantom. The developed mGRE sequence was prescribed without motion, covering the whole phantom for R2* and QSM map derivation. mGRE images were acquired with and without air at the center to validate the off-resonance correction ability of HDR-QSM.

Image analysis.—After image reconstruction, regions of interest were defined by circular contours in each vial, as shown in Figure 2 (red circle in A1). The mean R2* and susceptibility values from conventional and HDR-QSM were calculated for each vial. The mean R2* and susceptibility values were correlated to iron concentration using linear regression analysis.

Figure 2:

High-fidelity iron imaging in a calibration iron phantom. (A) Representative images of the iron phantom, including T2*-weighted images (A1), high-dynamic-range quantitative susceptibility mapping (HDR-QSM) without air as the off-resonance source (A2), conventional QSM image with air (A3), and HDR-QSM image with air (A4). In panel A3, evident off-resonance artifacts (indicated by red arrows) and susceptibility underestimation in high iron concentration vials can be observed. Furthermore, panel A5 illustrates the susceptibility difference between conventional QSM and HDR-QSM in the presence of air as the off-resonance source (A4 − A3). (B) Graphs show statistical analyses among the results from conventional QSM, HDR-QSM, and the reference standard QSM values. Panel B1 delineates the relationships among the proposed HDR-QSM without air (considered as the reference standard; depicted in black), HDR-QSM with air (blue), and conventional QSM with air (red). HDR-QSM demonstrates a strong correlation with the iron concentration in the phantom both with and without the off-resonance source (slope, 2.06 ppm/(mg/g); R², 0.98 and 0.98, respectively), while conventional QSM exhibits a weaker linear correlation compared with the reference standard (R², 0.70). Agreement assessment using Bland-Altman analysis of conventional QSM with air (red) and HDR-QSM with air (blue) versus the reference standard is presented in panel B2. The proposed method demonstrates a smaller bias and corrects the systematic underestimation of high susceptibility samples compared with conventional QSM (HDR-QSM vs conventional QSM: mean difference, −0.008 vs 0.046; 95% CI: −0.053, 0.037 vs −0.106, 0.014). ROI = region of interest, TE = echo time.

Animal Study

Animal preparation protocol.— Fourteen female canines (n = 14; 20–25 kg; 5–9 months old) underwent an experimental study involving surgically induced MI. Baseline cardiac MRI scans were conducted on all animals. Six canines were designated as healthy controls, while the remaining eight underwent left thoracotomy to induce MIs with 90-minute left anterior descending artery occlusion followed by reperfusion (10). Following successful reperfusion, the canines were given a minimum of 7 days to recover before undergoing imaging studies.

Image acquisition.— Animals underwent cardiac MRI scans at 1 week (acute) and 8 weeks (chronic) following MI. Clinical 2D breath-hold ECG-gated R2* maps (standard 2D R2*) and late gadolinium enhancement images were acquired over the whole left ventricle. The proposed free-breathing 3D mGRE sequence was prescribed to the same volume (acquisition time, 295 seconds). The 3D R2* maps, conventional QSM maps, and HDR-QSM maps were derived from the proposed reconstruction framework (24). A subset of animals (n = 5) was euthanized after in vivo scans for ex vivo imaging. Hearts were fixed and immersed in cylindrical containers with 10% formalin solution and scanned using a 2D multiple-section mGRE sequence with the same parameter described in the phantom study.

Image analysis.— Manual myocardial contouring, following the 16-segment American Heart Association model–based segmentation, was employed for both in vivo and ex vivo images. The image signal-to-noise ratio of in vivo 3D R2* maps, conventional QSM, and HDR-QSM maps were quantitatively evaluated using the intersegmental coefficient of variation of healthy and noninfarct segments. Image quality was assessed independently by a radiologist and an MRI physicist with over 10 years of cardiac MRI experience (L.T.H. and H.J.Y., respectively) who were blinded to imaging methods and scored on a five-point Likert scale (1 = severe artifacts, unreadable; 5 = no artifacts). The capabilities of the cardiac MR images for IMH detection and hemorrhagic myocardial iron content quantification were assessed. For IMH detection, receiver operating characteristic curve analysis was performed on the in vivo 3D R2*, conventional QSM, and HDR-QSM maps. For iron quantification, the measured R2* and QSM values were converted into tissue iron content. The iron content was calibrated using the tissue iron content measured with mass spectrometry. The in vivo hemorrhagic iron concentration was calculated by subtracting the iron concentration in the affected area from that of the septal-inferior remote segment obtained from the corresponding ex vivo images. The myocardial iron concentration was calculated based on a prior inductively coupled plasma mass spectrometry study from our group (30). Animal imaging study preparation, surgery, prior inductively coupled plasma mass spectrometry analysis, and image analysis details can be found in Appendix S1 and our previous work (31).

Statistical Analysis

All analyses were performed using IBM SPSS (version 28.0.1.1; IBM). For phantom data, linear regression was used to examine the relationship between the mean R2*, QSM, and iron concentration, serving as the calibration for ex vivo data. In addition, Bland-Altman analysis was performed to examine the limits of agreement and bias. In animal studies, data normality was assessed using the Kolmogorov-Smirnov and Shapiro-Wilk tests. Given the nonnormal distribution of the data, the Friedman test, repeated-measures analysis of variance, and mixed-model analysis of variance were used to evaluate intersegmental coefficient of variation, image quality scores, and regional iron content between modalities. The receiver operating characteristic curve analysis’ statistical significance and 95% CIs were determined using the DeLong test and Hanley and McNeil method. Linear regression and Bland-Altman analysis were used to compare the IMH extent and iron concentration across all modalities. Regression coefficients and 95% CIs were calculated, and the model’s goodness (R²) of fit was assessed. The Bland-Altman analysis was used to evaluate the agreement between the modalities. The statistical significance level was set at P < .05.

Results

Phantom Study

Results from the phantom study are depicted in Figure 2. Figure 2A shows images of the phantom with air (T2*-weighted image; echo time [TE], 9.94 msec), conventional QSM maps, and HDR-QSM maps. A QSM image without air was also acquired as the reference standard. HDR-QSM maps achieved uniform signal intensity in the vials despite the presence of a large off-resonance source. The high contrast-to-noise ratio and image quality illustrate HDR-QSM’s capability to correct for the conventional QSM bias under the influence of off-resonance artifacts. Panel A5 in Figure 2 illustrates the susceptibility difference between conventional QSM and HDR-QSM in the presence of off-resonance artifacts, where evident artifacts and underestimation of iron concentration are observed using the conventional QSM method. Quantitative analysis is presented in Figure 2B. Susceptibility values derived from HDR-QSM and the reference standard without an off-resonance source exhibited strong correlations with iron concentration (mg/g) (R², 0.98 and 0.98, respectively). In contrast, the conventional QSM method showed a weaker linear correlation in the presence of air (R², 0.70) and underestimation in vials with high iron concentration due to phase unwrapping and off-resonance artifacts. Bland-Altman plots comparing susceptibility differences from the reference (without air) with conventional QSM (bias, −0.046 ppm) and HDR-QSM (−0.008 ppm) are also shown. Notably, the conventional QSM measurements showed a significant bias and underestimation in high-susceptibility samples. The effect was successfully corrected by HDR-QSM.

In Vivo Study in Canine Models

Image quality evaluation.— The image quality of the conventional and proposed methods (standard 2D R2* maps, 3D R2* maps, conventional QSM, and HDR-QSM) from two animals with hemorrhagic MI at acute and chronic phases are compared in Figures 3 and 4. Figure 3 shows the images acquired at the acute phase of the injury with obvious microvascular obstruction and ventricular arrhythmia. Large hemorrhagic lesions with elevated iron content are shown on the T2*-based images accompanied by off-resonance–induced signal intensity dropout at the heart-lung interfaces (red boxes in Fig 3). Notably, motion artifacts from the unstable heartbeats are presented on the standard 2D R2* maps and lead to undiagnosable image quality, particularly on the basal section. Although 3D R2* and conventional QSM maps derived from the free-running sequence can mitigate the motion artifacts, obvious susceptibility artifacts still persist in the lateral-inferior segments. In addition, the conventional QSM maps presented evident streaking artifacts with hypointense streaks around the hemorrhagic zone (blue boxes in Fig 3). The proposed HDR-QSM alleviated the off-resonance and streaking artifacts and showed a homogeneous susceptibility map throughout the nonaffected myocardium. Similar trends were observed in the chronic lesions (Fig 4). In addition to the in vivo images, high-resolution artifact-free ex vivo R2* maps of the same heart are presented. High spatial correspondence is presented between the HDR-QSM and the ex vivo images, validating the high fidelity of iron content from HDR-QSM. To systematically assess the image quality of the cardiac MRI methods, the intersegmental coefficient of variation of the healthy myocardium and image quality scores were compared (Fig 5), demonstrating improved signal intensity homogeneity in noninfarct myocardium (lower intersegmental coefficient of variation) and image quality (higher IQ score) of HDR-QSM compared with the other methods. Notably, no R2* and conventional QSM images attain the highest quality score of 5, and no HDR-QSM images receive a quality score below 3, indicating the efficacy of the proposed method in mitigating the majority of artifacts and producing images with high diagnostic value. In some animals, the proposed method emerges as the sole approach to eliminating all artifacts, resulting in pristine artifact-free images.

Figure 3:

In vivo iron imaging from conventional and proposed methods of an acute myocardial infarction animal with ventricular arrhythmia. Representative short-axis cardiac MR images with (mid and apical) and without (basal) intramyocardial hemorrhage are presented in the same animal. Section-matched standard R2*(1/T2*) (obtained using a standard clinical breath-hold electrocardiographically gated T2* sequence), three-dimensional (3D) R2* maps, and quantitative susceptibility mapping (QSM) maps (reconstructed from the proposed 3D motion-robust framework) are presented with the late gadolinium enhancement (LGE) images. Persistent microvascular obstruction is shown in the IMH region (orange arrows). The standard R2* images show strong off-resonance artifacts in the lateral wall on all sections (red arrows). In addition, motion-induced signal intensity variation is seen in the septal region (green arrows), which significantly degrades the readability of the R2* maps. With the free-running 3D R2* maps and QSM maps, motion artifacts were alleviated on all images, yet off-resonance artifacts at the lateral walls (red arrows) still existed for both 3D R2* maps and conventional QSM maps. In addition, streaking artifacts at the periphery of the hemorrhagic lesion (blue arrows) persist in the conventional QSM reconstruction. High-dynamic-range–QSM (HDR-QSM) cleans up the artifacts and provides pristine image quality for quantitative IMH assessment. TE = echo time, T2*w = T2*-weighted image, 2D = two-dimensional.

Figure 4:

In vivo iron imaging from conventional and proposed methods of a chronic myocardial infarction animal with ex vivo comparison. Representative short-axis images from the conventional and proposed methods of an intramyocardial hemorrhage (IMH) dog in the chronic phase are presented. Late gadolinium enhancement (LGE) images and ex vivo R2* maps of the matching sections are shown to compare the in vivo results. Similar to the acute images, imaging artifacts strongly degrade the image quality of T2* (T2*-weighted [T2*w] and R2* maps) images and conventional quantitative susceptibility mapping (QSM) maps (red arrows). Good spatial IMH lesion correspondence (orange arrows) between the high-dynamic-range–QSM (HDR-QSM) and the ex vivo reference standard shows that the high-quality HDR-QSM maps can provide excellent accuracy for IMH assessment. In addition, blue arrows indicate streaking artifacts existed at the periphery of the hemorrhagic lesion on conventional QSM images that are mitigated with HDR-QSM. TE = echo time, 3D = three-dimensional, 2D = two-dimensional.

Figure 5:

Comparison of image quality (IQ) between iron imaging approaches. (A) Bar plot shows the intersegmental coefficient of variation in the healthy myocardium. A significantly lower variation is presented at high-dynamic-range quantitative susceptibility mapping (HDR-QSM) (0.31 ± 0.16 [SD]) compared with three-dimensional (3D) R2* map (0.73 ± 0.36) and conventional QSM (0.52 ± 0.50). (B) Bar plot shows IQ scores, where HDR-QSM showed the highest IQ score in all methods (3D R2*, 2.87 ± 0.51; conventional QSM, 2.43 ± 0.75; HDR-QSM, 3.58 ± 0.75). The data show that HDR-QSM provides a homogeneous signal intensity in noninfarcted myocardium across the whole heart and a boosted image quality for IMH assessment (* = P < .05, ** = P < .01). (C) Representative images classified based on the IQ of 3D R2* maps with R2* quality score from low (R2* IQ = 2) to high (R2* IQ = 4). Substantial off-resonance artifacts (red arrows) on both R2* maps and conventional QSM images are presented at the heart-lung interfaces. In addition, streaking artifacts (blue arrows) persist in the periphery region of hemorrhagic lesions and ventricular interfaces on the conventional QSM images. In several cases, these artifacts have a pronounced adverse effect on IQ, rendering those images either unreadable or yielding only minimal diagnostic information (score of 1 and 2). It is worth noting that HDR-QSM effectively reduces most of these artifacts and ensures that no images score below 3. Furthermore, the proposed HDR-QSM is the only approach capable of completely eliminating both the artifacts and resulting in high-fidelity iron images against the ex vivo reference standard (score 5). LGE = late gadolinium enhancement.

IMH detection.— The capability of HDR-QSM toward IMH detection and quantification was evaluated against the ex vivo reference standard and compared with the conventional methods (3D R2* and conventional QSM). Results for IMH detection are shown in Figure 6. HDR-QSM showed significantly improved IMH detection capability with the highest area under the receiver operating characteristic curve (AUC for 3D R2* map, 0.75 [95% CI: 0.56, 0.94]; AUC for conventional QSM, 0.69 [95% CI: 0.49, 0.89]; P (vs 3D R2*map) = .08; AUC for HDR-QSM, 0.96 [95% CI: 0.88, 1.00]; P < .001). HDR-QSM also showed improved sensitivity, specificity, and accuracy higher than 90% (sensitivity, specificity, and accuracy for 3D R2* map: 88.89% [eight of nine], 71.83% [51 of 71], and 73.75% [59 of 80], respectively; for conventional QSM: 88.89% [eight of nine], 61.97% [44 of 71], and 65% [52 of 80], respectively; for HDR-QSM: 100% [nine of nine], 92.96% [66 of 71], and 93.75% [75 of 80], respectively). A similar trend was observed in the accuracy analysis in three territories (lateral, septal, and inferior), where the HDR-QSM showed significantly improved accuracy (accuracy in lateral, septal, and inferior regions for 3D R2* map: 60% [15 of 25], 76.67% [23 of 30], and 84% [21 of 25], respectively; for conventional QSM: 44% [11 of 25], 83.33% [25 of 30], and 64% [16 of 25], respectively; for HDR-QSM: 92% [23 of 25], 96.67% [29 of 30], and 92% [23 of 25], respectively; Fig 5D).

Figure 6:

Intramyocardial hemorrhage detection comparison between the conventional methods and high-dynamic-range quantitative susceptibility mapping (HDR-QSM). (A) Receiver operating characteristic curve analysis shows that HDR-QSM provides significantly improved accuracy for intramyocardial hemorrhage detection compared with conventional iron-sensitive cardiac MRI (area under the receiver operating characteristic curve [AUC] for three-dimensional [3D] R2* map, 0.75; AUC for conventional QSM, 0.69; AUC for HDR-QSM, 0.96). The higher AUC for the HDR-QSM reflects the successful elimination of the off-resonance–induced imaging artifacts on in vivo images and provides a robust means to detect IMH. This is also reflected in the accuracy assessment in B, where HDR-QSM demonstrates superior sensitivity, specificity, and overall accuracy compared with the other methods. A consistent pattern emerges in the accuracy analysis across three myocardial territories (lateral, septal, and inferior). (C) Bullseye plot illustrates the representative segmentation model employed for the analysis of these territories. Accuracy comparison between the territories is presented in D. HDR-QSM displays significantly enhanced accuracy for all three territories (accuracy in the lateral, septal, and inferior region for 3D R2* map, 60%, 76.67%, 84%; conventional QSM, 44%, 83.33%, 64%; and HDR-QSM: 92%, 96.67%, 92%).

IMH iron quantification.— In vivo hemorrhagic iron content derived from R2* and QSM maps were compared between myocardial regions with and without IMH (MI_IMH, Remote_lateral, and Remote_{septal-inferior}). The results were validated against the ex vivo measurements of the same heart in Figure 7. In remote regions, HDR-QSM showed mean hemorrhagic iron content values closer to the ex vivo reference standard than the conventional methods (Remote_lateral[Fe]_Hemo for 3D R2*: 0.033 mg/g ± 0.216 [SD] [vs ex vivo, P = .001]; for conventional QSM: 0.003 mg/g ± 0.237 [P = .68]; for HDR-QSM: 0.003 mg/g ± 0.002 [P = .10]; for ex vivo: 0.001 mg/g ± 0.002; Remote_{septal-inferior}[Fe]_Hemo for 3D R2*: 0.000 mg/g ± 0.005 [vs ex vivo; P = .98]; for conventional QSM: −0.006 mg/g ± 0.007 [P = .009]; for HDR-QSM: 0.000 ± 0.003 [P = .83]; for ex vivo: 0.000 mg/g ± 0.003). In addition, conventional QSM showed a biased iron measurement in MI_IMH[Fe]_Hemo, which was successfully corrected by the HDR-QSM (MI_IMH[Fe]_Hemo for 3D R2*: 0.032 mg/g ± 0.170 [vs ex vivo; P = .58]; for conventional QSM: 0.018 mg/g ± 0.009 [P = .04]; for HDR-QSM: 0.026 mg/g ± 0.156 [P = .40]; for ex vivo: 0.030 mg/g ± 0.125). In Figure 7B–7D, iron content derived from in vivo 3D R2* maps, conventional QSM, and HDR-QSM were regressed against the ex vivo artifact-free maps. HDR-QSM demonstrated a substantially higher R² value and a slope closer to 1 (R², 0.71; slope, 0.73), indicating the removal of artifacts from the remote lateral region on the 3D R2* map and conventional QSM images (R², 0.14; slopes, 0.61 and 0.45, respectively). In addition, the relationship between iron content and ex vivo measurement of iron-sensitive MRI parameters is calibrated with mass spectroscopy in fixed hearts and shown in Figure S5. The Table displays the quantitative comparison results between iron imaging methods.

Figure 7:

Ex vivo comparisons of hemorrhagic iron content derived from the conventional methods and high-dynamic-range quantitative susceptibility mapping (HDR-QSM). (A) Bar plots show hemorrhagic myocardial iron content derived from the conventional methods, and HDR-QSM was compared against the ex vivo reference standard in different myocardial regions using a box and whisker plot. Only HDR-QSM showed comparable iron content to the ex vivo hearts in all myocardial territories, indicating its ability to mitigate imaging artifacts induced bias in the whole heart. (B–D) Similar trends are presented in the linear regression analysis between in vivo and ex vivo measurements. HDR-QSM shows a substantially tighter fit, and a slope closer to 1, reflecting the higher accuracy for iron quantification based on HDR-QSM maps (* = P < .05). MI_IMH = infarcted myocardium with intramyocardial hemorrhage, Remote_Sep-inf = remote myocardium in the inferior septal wall, Remote_Lat = remote myocardium in the lateral wall, 3D = three-dimensional.

Quantitative Comparison between Conventional and Proposed Iron Imaging Methods

Discussion

In this study, we developed a cardiac QSM technique to detect IMH and quantify regional iron content in hemorrhagic MI. Combining a free-breathing non-ECG–gated whole-heart cardiac MRI acquisition and a high-dynamic-range QSM algorithm, the developed technique significantly reduced imaging artifacts on conventional R2* and QSM maps and boosted the accuracy of quantitative IMH assessment. Comparisons between in vivo susceptibility maps and ex vivo measurements demonstrated that HDR-QSM outperformed R2* mapping in detecting IMH (AUC, 0.96 vs 0.75; P < .001) and quantifying iron content. The linear correlation between HDR-QSM and ex vivo measurements was 0.73x + 0.00 (R², 0.71), while for R2* it was 0.61x + 0.02 (R², 0.14).

Major advances in the pathologic consequences of IMH now point to the causal involvement of iron and its relationship to the rising prevalence of postreperfusion heart failure. Multiple studies (1,3,5,6) have demonstrated that focal iron deposition after a hemorrhagic MI is associated with increased arrhythmogenic risk, prolonged inflammation, arrhythmic fatty remodeling, and adverse left ventricular remodeling (2–6). Unfortunately, the current tools for myocardial iron imaging, such as T2 and T2* cardiac MRI, are either nonquantitative or unreliable for IMH assessment in patients with AMI. The precise and accurate iron quantification of different left ventricular territories based on HDR-QSM can boost IMH diagnosis accuracy, facilitate individualized patient management, and have the potential to guide the development of therapies against hemorrhage-mediated myocardial pathologic conditions (3,7–9).

At present, T2* cardiac MRI is widely recognized as the most viable method for evaluating cardiac iron levels (32,33). While studies have demonstrated a clear correlation between myocardial R2* values and myocardial iron content, indicating a potential noninvasive method for measuring iron in the heart (23), current cardiac T2* mapping sequences have limitations stemming from well-known imaging artifacts. Particularly, strong B0 inhomogeneity at the heart-lung interfaces typically causes severe off-resonance artifacts in the anterior, inferior, and lateral regions of the myocardium. This compromises the reliability of regional iron quantification based on T2*. In addition, T2* values are parameter dependent, which means the iron level is further complicated by differences in field strengths and imaging parameters during the T2* acquisitions. All of these hinder the clinical adoption of T2* mapping, making quantitative IMH assessment difficult for patients with acute MI (15,18).

QSM, however, offers consistent tissue susceptibility measurements across imaging parameters and field strengths (6). Although the feasibility of in vivo cardiac QSM has been demonstrated for blood oxygenation level quantification (34), the translation of QSM for IMH is additionally challenging (35). Imaging confounders, such as section mismatch from multiple breath holds, gating errors from cardiac arrhythmia, and prolonged imaging time from inefficient gating render current QSM techniques unsuitable for assessing IMH (5,34). Particularly, the conventional prospectively triggered sequences are prone to motion errors in patients with MI with compromised breath-holding ability and arrhythmia. The irregular motion can cause artifactual phase accumulation for late echoes and corrupt the QSM reconstruction. The developed HDR-QSM adopted a free-running acquisition that allows retrospective selection of motion states. This can mitigate artifacts caused by irregular motions in MI settings and identify a stable respiratory position with an optimized B0 profile at the heart-lung interfaces. The retrospective nature can also substantially reduce the complexity of the imaging acquisition and boost the efficiency and robustness of the studies (24). Although QSM can, in theory, remove the background off-resonance field from phase maps, the conventional QSM approach has a restricted dynamic range and is susceptible to unwrapping and streaking artifacts in extreme conditions. This is particularly prominent with IMH imaging, where the phase changes are swift at heart-lung interfaces and near the focal iron lesions (5,29,35–37). HDR-QSM extends the dynamic range of the phase maps by introducing a new regularization term in the TE dimension to derive accurate phase maps under the influence of a strong off-resonance field. Furthermore, it mitigates streaking artifacts with a multilevel approach that otherwise contributes to the underestimating iron content in the hemorrhagic core (evidenced in the conventional QSM in Fig 7A) (29,38).

To evaluate our in vivo results, we utilized high-resolution ex vivo images to assess the spatial distribution of the lesions and calibrated the findings with mass spectroscopy. In prior studies, R2* and QSM have been extensively studied and validated in the heart under different field strengths (4–6,10,23). Notably, susceptibility measurements from the artifact-free HDR-QSM showed a slope of 2.06 ppm/(mg/g) for iron concentration in the iron phantom studies, consistent with tissue iron content and earlier published values (5,6). Furthermore, our iron content measurements in the IMH hearts aligned well with the previously reported values (4).

The lower resolution of the in vivo images compared with ex vivo scans may lead to a blooming effect in R2* maps, potentially resulting in an overestimation of in vivo R2*-based IMH lesions. This effect could result in a nonstatistically significant overestimated IMH size when using R2* images. When comparing the measurements between the ex vivo high-resolution R2* maps and QSM maps, we found that the values were highly aligned when the blooming effect was alleviated by the smaller voxel size (P = .78).

In this study, most IMH lesions showed microvascular obstructions at acute scans after a 180-minute left anterior descending coronary artery occlusion. Notably, not all microvascular obstructions are hemorrhagic (39). Prior studies have shown that IMH is an independent prognostic marker of microvascular obstruction for adverse cardiovascular events, making an accurate and reliable IMH assessment an important prognostic measure as well as a therapeutic target when managing patients with acute MI (39).

Although not explored in the current study, the noninvasive and quantitative nature of HDR-QSM has the potential to facilitate longitudinal monitoring of IMH lesion changes over time and their relationship with scar size and location during recovery. This can assist the study of IMH-related myocardial remodeling and enable the understanding of its underlying mechanism.

This study faced certain limitations. We reconstructed diastolic in vivo QSM maps, which may represent a different contractile state of ex vivo hearts. The iron concentration in an IMH-induced lesion is expected to be consistent between cardiac phases. While systolic QSM maps can be potentially reconstructed from the proposed technique to provide thicker myocardium for iron assessment, they are outside the scope of this article. While HDR-QSM maps showed significantly improved IMH assessment ability compared with the conventional methods, further improvements are possible. First, regularization parameters for multilevel QSM were selected empirically for animal hearts with hemorrhagic MIs, but self-calibrated parameter selection can be achieved with algorithms to improve the robustness of the measurements (40). Additionally, it is possible to further expand the dynamic range for more demanding applications (eg, the presence of high-susceptibility contrast agents and metallic devices) (29). Although phase aliasing is mostly corrected with HDR-QSM, residual artifacts can still be observed outside the myocardium (eg, in the epicardial fat) in animals with extended B0 off-resonance, which may potentially compromise measurements. Advanced B0 shimming techniques can be potentially adopted to improve the image quality and precision of iron quantification (41). Another study limitation is that we developed the proposed technique in a 3-T scanner, which can boost the sensitivity of iron content. However, it is not the most commonly adopted cardiac MRI field strength. The high field strength can lead to amplified off-resonance artifacts and make QSM more challenging. Even so, our data present a high robustness against the off-resonance artifacts, and we foresee a smooth translation of HDR-QSM to 1.5-T imaging for wider clinical adoption. Furthermore, the current study included animals with left anterior descending coronary artery–based infarction to investigate well-defined lesions in territories with and without imaging artifacts. In four animals, MIs extended to the inferior-lateral walls in mid and apical sections, where HDR-QSM delineated the lesions from the imaging artifacts and boosted the accuracy of IMH detection.

In conclusion, the developed fast and free-running cardiac QSM accurately detected IMH and quantified the intramyocardial iron content of the entire heart within 5 minutes without ECG gating and breath holds. With clinical studies currently underway to further test and validate the proposed method in real-world clinical settings, HDR-QSM holds promise for enhancing IMH assessment in patients with AMI, thus enabling individualized MI management strategies, guiding the development of iron-targeted therapies, and ultimately reshaping the landscape of the management for patients with IMH.