Imaging of Reperfused Intramyocardial Hemorrhage with Cardiovascular Magnetic Resonance Susceptibility Weighted Imaging (SWI)

Purpose To report initial experience with TE-averaged susceptibility weighted imaging (SWI) in normal subjects and acute myocardial infarction (AMI) patients for the detection of intramyocardial hemorrhage (IMH). Materials and Methods 15 healthy control and 11 AMI subjects were studied at 1.5T before contrast agent administration with a dark blood double inversion recovery multiple spoiled gradient-echo sequence. Magnitude, susceptibility weighted and TE-averaged images were reconstructed from raw data. Contrast and signal-difference-to-noise were measured and compared between methods for IMH detection. Results There were six patients with microvascular obstruction (MVO) and four patients with IMH detected by TE-averaged SWI imaging. All patients with IMH on SWI scans had MVO on late gadolinium-enhanced (LGE) imaging. There was a three-fold increase in IMH contrast with SWI compared to magnitude images. IMH contrast decreased and signal-to-noise increased with increased TE averages. Conclusions TE-averaged SWI imaging is a promising method for myocardial tissue characterization in the setting of AMI for the detection of IMH. Along with gray-scale colormap inversion, it combines not only magnitude and phase information, but also images across TEs to provide a single image sensitive to IMH with characteristics similar to LGE imaging.

In a recent publication [43], we showed that IMH significantly reduces image phase below normal variations. In the present work, we report our initial experience with myocardial SWI imaging (combined gradient-echo magnitude and phase imaging) in normal subjects and AMI patients for the detection of IMH. We propose TE image averaging and gray-scale inversion as a means of providing a single image with good image SNR and excellent contrast for the detection of IMH. Additionally, we compared single-echo SWI and TE-averaged SWI quantitatively for image SNR and contrast characteristics.

Materials and Methods
Study data from a previous published report [43] which studied gradient-echo image phase was used to generate SWI images. LV functional measures and infarct sizes are listed in this publication. Twenty-six individuals underwent MR imaging at 1.5T using a standard body matrix phased array coil (Magnetom Siemens Avanto, Erlangen, Germany). The coil consisted of four clusters of three anterior and posterior elements. Elements were combined with the Auto (CP) coil mode and sampled as four clusters. Eleven subjects were enrolled at 3 days after their first myocardial infarction (age = 55.0 ± 11.5 years, range = 38.1-79.0) and fifteen were healthy control subjects (age = 61.7 ± 8.9 years, range = 41.9-76.4). AMI patients volunteered for this research study prior to hospital discharge after successful emergent reperfusion therapy with stent placement. Each subject underwent imaging according to a protocol approved by the institutional review board. All subjects signed an institutional review board approved, Health Insurance Portability and Accountability Act compliant consent form prior to study initiation.

MR Imaging
Acquisitions were performed during suspended respiration at end-expiration. Oxygen (2 l/min via a nasal cannula) was provided to improve breath-holding. For each part of the study, identical slice locations were used consisting of 7-10 parallel short-axis slices spanning the LV myocardium from the base to apex and three LV long axis views (two-, three, and four chamber). Before contrast agent administration, a dark blood double inversion recovery multiple spoiled gradient-echo sequence (1 slice per breathhold, repetition time = 20 ms; 12 echo times, 2.4-15.5 ms (1.2 ms spacing), flip angle = 20 degrees, bandwidth = 1860 Hz/pixel, in-plane spatial resolution = 2.3 x 1.7 mm 2 , slice thickness = 8 mm, velocity compensation in read and slice) was used for SWI imaging. Imaging was performed in mid-diastole using a prospectively gated segmented acquisition of k-space over 13 heartbeats with ECG gating using a signal from leads placed on the subject's chest. Raw k-space data were saved to the scanners hard disk and transferred to a personal computer for offline image reconstruction. A phase sensitive inversion recovery gradient-echo pulse sequence (1 slice per breathhold, repetition time = 8.7 ms; echo time = 3.4 ms; flip angle = 25 degrees, bandwidth = 130 Hz/pixel, in-plane spatial resolution, 1.7 x 1.3 mm 2 , slice thickness = 8 mm) was used for late gadolinium enhanced (LGE) infarct imaging [50]. LGE imaging was commenced 15 minutes after bolus contrast agent administration (0.15 mmol/kg gadopentetate dimeglumine, Magnevist, Bayer Healthcare, Wayne, NJ). The inversion time parameter was manually adjusted to minimize the signal of viable myocardium (250-350 ms) at the start of LGE acquisitions.

Image Reconstruction
Offline SWI image reconstruction was performed using Matlab R2009b (Mathworks, Natick, MA) from saved raw data files for both long and short axis images. Three classes of image reconstruction were performed: 1) standard magnitude and high-pass filtered phase 2) SWI and 3) echo time (TE) averaged.
1. Standard Magnitude and High-pass Filtered Phase Image Reconstruction. Images were independently reconstructed for each TE and coil element. Complex images were reconstructed using Matlab's two-dimensional discrete Fourier transform function (fft2) and full resolution magnitude and phase images were extracted. Next, a complex low pass filtered (LPF) image using an N = 64 point radial squared hanning filter LPF kx; ky ð Þ ¼ 0:5 1 À cos 2p ffiffiffiffiffiffiffiffiffiffiffiffi ffi was reconstructed by filtering k-space data and Fourier transform. High-pass filtered phase images were then calculated by dividing the low pass filtered image into the full resolution complex image and then extracting the phase component [44]. Individual coil element magnitude images were combined using a sumof-squares reconstruction [51]. 2. Susceptibility Weighted Image (SWI) Reconstruction. SWI images for each TE were constructed with 3 and 6 phase mask multiplications (N).
T2 Ã measurements were performed from magnitude images using ImageJ for region-of-interest (ROI) measurements and Matlab for curve fitting. ROIs were drawn on the short-axis images (basal, mid-ventricular and apical) according to the 16-segment AHA model [52]. For each ROI, the mean image intensity for each TE was computed and a T2 Ã decay curve was constructed. Each measured decay curve was fitted with an exponential function: S = So exp(-TE/ T2 Ã ) + C; where S is the image signal intensity, So and C are constants, TE is the echo time and T2 Ã is the transverse relaxation time.

Image Analysis
Average signal intensity was measured from identical region-of-interests for each reconstruction in the hemorrhagic infarct (identified as a hypointense lesion) and adjacent myocardium. The noise standard deviation was estimated from an ROI without signal (either outside of the body or in the lung cavity). Contrast and signal-difference-to-noise (SDNR) [53] were calculated from these measurements for source magnitude images, SWI images and TE-averaged images. Contrast = (S1-S2)/S1 and SNDR = (S1-S2)/STD, where S1 and S2 are mean image intensities and STD is and estimate of the noise standard deviation. Analysis of variance (ANOVA) and Student's t-test was used to determine significant differences between reconstruction methods and number of TE averages as appropriate. A p-value less than 0.05 was used to determine significant differences.
A single reader blindly reviewed TE-averaged SWI images for artifacts and recorded the number of control subjects and AMI patient with artifacts and their anatomical location. Using a semi-automatic detection algorithm, a signal intensity threshold of fifty percent of the maximum myocardial signal was applied to quantify the total LGE infarct volume from short-axis images [54]. Discrete areas of hypointensity within LGE were considered to be MVO. Although under the fifty percent signal intensity threshold, these pixels were included in the calculation of LGE infarct volume. Volumes of IMH and MVO were measured using visual intensity thresholding. LGE, IMH and MVO volumes are reported as a percentage of LV myocardial volume. For the analyses of T2 Ã and segments with IMH detected with TE-averaged SWI, repeated measures ANOVA and mixed model analysis was used. Marginal means and standard deviations reported. Results from a representative AMI subject are given in Figs 1-5. Source magnitude images (Fig 1) show an IMH lesion in the mid to basal inferior wall. SWI images (Fig 2) show better lesion definition via improved image contrast across all echo times. The IMH lesion is best defined at TE = 15.5ms, but image quality is severely degraded by increased noise giving it a cloudy appearance. TE-averaging (Figs 3 and 4) improves SNR for both magnitude and SWI images, but IMH contrast is better in SWI images, (see Fig 4, nTE = 9). Grayscale colormap inversion of echo combined SWI images (Fig 5) yields images characteristics similar to LGE imaging; normal myocardium is dark, with bright fat and blood. In this patient with MVO, IMH is seen as a hyperintense lesion in the inferior wall in the same segment as MVO.

Results
Quantitative results from ROI measurements displayed in Fig 6 show the reduction in IMH contrast and increase in IMH SDNR with increased averages (p<0.001). There was a three-fold increase in IMH contrast with SWI compared to source images. Contrast increased when comparing six vs. three phase mask multiplications (p<0.01). Contrast decreased and signal-tonoise increased with increased TE averages. SDNR increased with TE averages up to nTE = 9 (p = 0.07). When comparing unaveraged single echo magnitude source and SWI images, SDNR increased 75.1% and 77.5% respectively as echo time increased from TE = 2.3 to 15.5ms in this study. Both were always lower than TE-averaged image SDNR irrespective of the number of averages.
There were six patients with microvascular obstruction and four patients with intramyocardial hemorrhage detected by TE-averaged SWI imaging. All patients with IMH on SWI scans had MVO on LGE scans. Results of IMH and MVO volume measurements are displayed in Fig  7. The trend was that larger infarcts had MVO and IMH, but this was not true in all cases as the largest infarct had no evidence of IMH. Three additional patients are shown in Figs 8, 9  (Figs 9 and 10). Excellent IMH definition is realized with similar positive image contrast when compared to LGE images. Fig 11 shows a common artifact found in TE-averaged SWI images resulting from the coronary vein. 6/15 (40%) control subjects and 3/11 (27%) of AMI patients had at least one image with this artifact. Although common among patients, the artifact rarely affected more than one slice orientation. Six two chamber, three short axis and one three chamber orientations were affected, allowing interpretation of the myocardial segments in other orientations without image artifacts.

Discussion
This study has demonstrated the utility of SWI imaging for myocardial tissue characterization in the acute phase of myocardial infarction. SWI depicted myocardium in areas of MVO indicative of IMH. When compared to magnitude imaging, SWI provided incremental improvements in IMH contrast and TE-averaging provided incremental improvements in image SNR. Combined with grayscale colormap inversion, TE-averaged SWI imaging provided a high image contrast and SNR method for the depiction of IMH with characteristics similar to LGE infarct imaging.
There are a number of MR methods for the detection of IMH with encouraging reports in the literature. T1 weighted imaging is sensitive not only to IMH [24] but also to myocardial edema [19] and intramyocardial fat [55]. T2-weighted imaging [18] depends on the hyperintense signal from myocardial edema and depiction of IMH as a hypointense core [28,29,31,35,36,38]. Both HPF phase imaging [43] and T2 Ã imaging [25,26,30,38,41,42] are closely related to SWI imaging. T2 Ã weighted imaging with a single long TE is often compromised by dropout artifacts in the inferior and lateral walls, SNR and motion degradation. T2 Ã mapping combines images from multiple TEs into a quantitative map where a normal T2 Ã threshold could be applied, but also can suffer from image quality artifacts. HPF phase imaging produces a phase image which is also quantitative, but is typically not clinically viewed. TE-averaged SWI should be viewed as an incremental improvement to T2 Ã -weighted imaging of IMH as it utilizes the same pulse sequence, but provides better contrast via magnitude and phase combination and averaging over TEs to improve SNR. SWI similar to T2 Ã and T2 techniques detect IMH via a low contrast signal mechanism. Detecting hypointense IMH could be complicated by other low intense structures such as the LV cavity. IMH is typically seen within TE-averaged magnitude images from a patient with an inferior myocardial infarction at 3 days after emergent revascularization and stenting of a right coronary artery occlusion. Images are averaged over the nTE longest source echo time images (Fig 1). Note the improved SNR, but decreased IMH lesion (arrows) definition compared to source images.  TE-averaged susceptibility weighted (SWI) images from a patient with an inferior myocardial infarction at 3 days after emergent revascularization and stenting of a right coronary artery occlusion. Images are averaged over the nTE longest SWI images (Fig 2). Note the improved SNR and loss of contrast increased number of echo time averages (NTE). nTE = 9 was quantitatively shown to have the best SNR across patients with good hemorrhage contrast.   Imaging of Intramyocardial Hemorrhage with Magnetic Resonance SWI the LV myocardium and in this study was always surrounded by high contrast myocardium. SWI image artifacts can results from sources with susceptibility differences. Deoxygenated blood in the coronary sinus was a frequent source of an artifact identified in this study. The coronary vein can be distinguished from IMH due to its predictable location outside of the LV myocardium.
Higher main magnetic field strengths (3T or 7T) have the advantage of better susceptibility weighting, but may have more artifacts. Optimal field strength for IMH detection would have to be determined, but preliminary results obtained in this paper support the use of 1.5T SWI imaging for the detection of IMH. Gray scale inversion of SWI images yields positive contrast with bright injured myocardium and dark or nulled viable myocardium. Image contrast could improve adoption with bright IMH similar to infarcted myocardium in LGE imaging, but  could also cause confusion between the two imaging techniques. Another advantage of this technique and motivation of this work is the combination of multiple (in this case 24) magnitude and phase images into a single image. Similar to T2 Ã mapping. TE-averaged SWI imaging yields fewer images through image information combination to improve physician evaluation speed.
The results of this study should be considered in light of some limitations. Limitations of this study include the small study size, predominantly male study population. Patients in this study were physiologically stable and without contraindication to MR imaging. The patient population must be considered when interpreting the data and may not be representative of a larger population. IMH was not seen in normal volunteers or in segments without LGE. This study has demonstrated the feasibility of IMH detection with TE-averaged SWI imaging, but future larger studies are needed to determine its widespread performance and accuracy across a wide range of infarct ages and types. Optimal imaging parameters for routine clinical imaging have not been determined in this study. The optimal longest TE, number of echo times as well as the necessity of velocity compensation in imaging myocardium should be further investigated. We did not investigate lower bandwidth acquisitions which would increase signal-to-noise at the expense of a longer sampling intervals during the cardiac cycle and hence did not determine the optimal bandwidth for cardiac SWI imaging. Semi-automated image analysis was used and manual drawing of endo-and epi-cardial borders as well as intensity threshold may be operator dependent.
In conclusion, TE-averaged SWI imaging is a promising method for myocardial tissue characterization in the setting of AMI for the detection of IMH. Along with gray-scale inversion, it combines not only magnitude and phase information, but also images across several TEs to provide native images sensitive to IMH with characteristics similar to LGE imaging.