Imaging of Myocardial Infarction for Diagnosis and Intervention Using Real-Time Interactive MRI Without ECG-Gating or Breath-Holding

doi:10.1002/mrm.20174

Magn Reson Med. Author manuscript; available in PMC 2007 August 6.

Published in final edited form as:

Magn Reson Med. 2004 August; 52(2): 354–361.

doi: 10.1002/mrm.20174

PMCID: PMC1939888

NIHMSID: NIHMS27039

Imaging of Myocardial Infarction for Diagnosis and Intervention Using Real-Time Interactive MRI Without ECG-Gating or Breath-Holding

Michael A. Guttman,¹^* Alexander J. Dick,² Venkatesh K. Raman,² Andrew E. Arai,¹ Robert J. Lederman,² and Elliot R. McVeigh¹

¹Laboratory of Cardiac Energetics, National Heart, Lung and Blood Institute, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland.

²Cardiovascular Branch, National Heart, Lung and Blood Institute, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland.

^*Correspondence to: Michael Guttman, NIH/NHLBI, 10 Center Dr., Building 10, Room B1D416, Bethesda, MD 20892-1061. E-mail: mguttman/at/nih.gov

The publisher's final edited version of this article is available at Magn Reson Med

See other articles in PMC that cite the published article.

Abstract

Current methods for MRI of infarcted myocardium require ECG-gating and breath-holding during contrast-enhanced segmented k-space inversion-recovery (IR) imaging. However, ECG-gating can be problematic in MRI, and breath-holding can be difficult for some patients. This work demonstrates that infarcted tissue can be visualized without ECG-gating or breath-holding with the use of intermittent inversion pulses during real-time (RT) interactive imaging with steady-state free precession (SSFP). The sequence generates a RT image stream containing a myocardium-nulled image every few frames, which allows nearly simultaneous observation of both infarcted regions and wall motion. First-pass per-fusion and wall motion can be simultaneously observed with minor parameter modifications. This method may reduce diagnostic scan time, expand the target population, improve patient comfort, and facilitate targeted, interventional treatment of infarcted myocardium. Supplementary material for this article can be found on the MRM website at http://www.interscience.wiley.com/jpages/0740-3194/suppmat/index.html. Published 2004 Wiley-Liss, Inc.^†

Keywords: real-time, interventional, MRI, infarct, delayed hyperenhancement, SSFP, FISP, myocardial ischemia, perfusion

Inversion-recovery (IR), gradient-recalled-echo (GRE), delayed hyperenhancement (DHE) imaging is a validated method for detecting the presence and extent of myocardial infarction (MI) (1,2). Breath-holding and ECG-gating for segmentation of k-space are employed to increase spatial resolution, temporal resolution, and contrast. However, ECG-gating can be problematic for patients with irregular heart rhythms, and breath-holding can be uncomfortable and difficult for ill patients. These difficulties may reduce patient compliance and ultimately limit the population that can be examined with current techniques of infarct detection using DHE.

In addition, there is currently no established method to visualize infarcted myocardium with a real-time (RT) imaging sequence, which may be desirable during an interventional cardiovascular procedure. Switching between RT and ECG-gated pulse sequences during a sensitive interventional procedure is an unattractive alternative. RTMRI monitoring of cardiac function would be temporarily interrupted, which could increase procedural risks. A more desirable approach is to modify the pulse sequence interactively during RT imaging (3) in order to alter image contrast as needed.

Toward that goal, we developed an interactive RT imaging method that produces images frequently enough to monitor cardiac function, and can optionally produce infarct-enhanced images every few frames. The technique is based on a high signal-to-noise-ratio (SNR), steady-state free precession (SSFP) pulse sequence (4) modified with intermittent preparation pulses (5,6). For diagnostic use, this method may facilitate the search for diseased myocardium by detecting abnormal wall motion and DHE in the same scan without ECG-gating and breath-holding, and by allowing the use of interactive slice positioning tools to locate abnormal myocardium. We compared DHE obtained from RT imaging with that from conventional IR-GRE by estimating infarct size and contrast in images from infarcted pigs and from patients with known infarcts. With minor parameter changes, the sequence also allows simultaneous observation of first-pass perfusion (7) and wall motion.

For interventional use, infarcted tissue is visible in the myocardium-nulled images, and the view-shared images show device position relative to surrounding anatomy while they provide essential monitoring of cardiac function. The composite information from this image stream may facilitate navigation of an interventional device toward the target infarcted tissue. This multipurpose imaging technique can depict the position of a catheter-based receiver coil relative to infarcted myocardium in the left ventricle (LV) of an infarcted animal.

MATERIALS AND METHODS

Theory of Operation

Conventional DHE imaging uses the accumulation of a T₁-shortening contrast agent in infarcted regions of myocardium. After systemic injection of contrast agent and a delay of 10–30 min, the infarcted regions will contain more contrast agent than normal myocardium, and will exhibit a shorter T₁ relaxation time (1). Tissue magnetization is typically prepared before imaging with a 180° inversion pulse and a delay such that a subset of phase-encoding lines are collected around the time when the longitudinal magnetization, M_z, of normal myocardium has recovered to zero (i.e., the inversion time (TI)). At this time, the infarcted tissue will have recovered to some positive value, due to its shorter T₁. After the subset of data has been acquired, there is a delay (typically skipping the next heart cycle) to allow the longitudinal magnetization to recover fully to the initial value, M_z0, to maximize SNR. The process is repeated until all the data for a “myocardium-nulled” image have been collected. The patient is asked to stop breathing during the data acquisition, and ECG-gating is used to collect data during the relatively motionless period of diastasis.

In the proposed technique (see Fig. 1), RT-SSFP imaging is interrupted every few images by a nonselective 180° inversion pulse and a delay (TD1), as in the conventional method. With the use of a linear phase-encode trajectory, all of the phase-encoding lines for the myocardium-nulled image are collected. The delay TD1 is chosen interactively such that the central phase-encoding views (i.e., near k_y = 0) are collected near the time the magnetization of normal myocardium has recovered to zero. Partial k-space coverage in the phase-encoding direction with zero-filling is employed to decrease the image acquisition time. If a short TI is required (e.g., soon after Gd-DTPA administration), the center of k-space is collected early. For a longer TI, one can reverse the phase-encode order to collect the center of k-space later. This allows an increase in TI while it also minimizes the increase in delay time TD1.

FIG. 1

Schematic illustration of a pulse sequence for RT-SSFP imaging with intermittent inversion pulses for infarct enhancement. One inversion cycle is shown, starting with an inversion pulse, a myocardium-nulled image, and then several view-shared images during (more ...)

SSFP continues after the myocardium-nulled image without interruption, but alternates between collection of all the odd-numbered phase-encoding views and collection of all the even-numbered views. A set of view-shared images are produced during signal recovery, each of which is reconstructed from the most recently acquired collections of even- and odd-numbered views. Once steady state is reached, an optional delay time TD2 may be employed for recovery of magnetization, followed by an inversion pulse to produce the next myocardium-nulled image.

We set TD2 to zero for all of the experiments presented herein to decrease the time required to acquire the myocardium-nulled image; however, this was done at the expense of some SNR. SNR is reduced because the longitudinal magnetization before the inversion pulse is the SSFP value, not the fully recovered M_z0. However, SSFP is intrinsically a higher-SNR technique than spoiled GRE, and the additional images obtained during magnetization recovery can be of value for diagnosis (wall motion) and intervention (device positioning). We investigated these benefits in vivo. In addition to decreased SNR, we also expected lower temporal resolution to cause motion blurring, since the image acquisition window covers a larger portion of the cardiac cycle than the conventional technique.

The following scan parameters were interactively adjustable during imaging: scan plane, TD1, TD2, view-sharing on/off, and number of view-shared images. In addition, the inversion pulse could be turned off to perform only RT SSFP imaging. A button could be pressed to display only the myocardium-nulled images (i.e., view-shared images would not be reconstructed) to facilitate optimal adjustment of delay times, and to avoid the distraction of dynamic image contrast in the same window.

Experiments

Apical-septal MIs were induced in six swine by transcatheter closed-chest coil occlusion of the distal left anterior descending (LAD) coronary artery, approximately 1 week prior to imaging. In addition, three human patients with known coronary artery disease (CAD) were imaged as volunteers. A solution of 0.2 mmol/kg Gd-DTPA (Magnevist; Berlex, Inc., Wayne, NJ) was injected systemically as a contrast agent. One posterior and two anterior surface coils were used, and RT imaging was performed in a Siemens Sonata 1.5T MR scanner with SSFP. The first four experiments were conducted with the following parameters: matrix = 120 × 160, slice thickness = 6 mm, FOV = 255 × 340 mm, TR = 2.84 ms, 3/4 partial phase k-space acquisition with zero-filling, and bandwidth = 1008 Hz/pixel, yielding 2.1 × 2.1 × 6 mm pixels. A slightly higher spatial resolution was used for the remaining experiments (including the human volunteers): matrix = 144 × 192, FOV = 283 × 340 mm, and bandwidth = 1002 Hz/pixel, yielding 2.0 × 1.8 × 6 mm pixels (all other parameters unchanged). The view-shared images were reconstructed at rates of 7.8 and 6.5 frames per second, respectively, for the two sets of parameters.

After the myocardium-nulled image was acquired, up to four view-shared images were acquired during magnetization recovery. TD1 was set according to the time elapsed since the Gd-DTPA injection, and was adjusted interactively during scanning. The optimal delay time tended to be 25–50 ms, resulting in a TI of 150–200 ms, depending on the k-space matrix size.

For reference, we performed conventional ECG-gated, breath-held IR-GRE imaging (2) using the Siemens Turbo-FLASH sequence with a 168 × 256 matrix, 32 views per cardiac phase, 6-mm slice, and 223 × 340 mm FOV, yielding 1.3 × 1.3 × 6 mm pixels. The TI was 240–300 ms, and one additional heartbeat was used for signal recovery. One slice location from each experiment was used to compare infarct-to-normal contrast-to-noise ratios (CNRs) and infarct area between the conventional and RT imaging methods. For each experiment, we compared the relative performance of the sequences by calculating the ratio between CNR values from IR-GRE and the average of several RT-IRSSFP images. The use of this ratio reduces variability between experiments and facilitates comparison with simulation data. Previous reports have suggested that accurate estimations of the infarct area can be problematic (2,8), and we present our comparison with the following caveats: RT imaging is asynchronous with the cardiac cycle, and each myocardium-nulled image appears at a different cardiac phase. Since we could not choose an RT image from exactly the same cardiac phase as the ECG-gated image, we measured multiple RT images that were as close as possible to the cardiac phase of the gated image, and then averaged the measurements. Differences in cardiac phase, as well as differences in respiratory phase and manual segmentation uncertainties, are expected to increase error in the infarct area comparison, but not appreciably in the CNR comparisons.

In four experiments (two in swine, and two in humans), imaging was performed during first-pass perfusion of GdDTPA. RT imaging was started with view-sharing and no saturation preparation. Just before the injection was given, a button was pressed to play a 90° nonselective saturation preparation immediately prior to each view-shared image acquisition. This mode was used to visualize the enhancement pattern of first-pass perfusion simultaneously with myocardial wall motion.

To simulate a possible interventional scenario, a prototype MR-active catheter (Boston Scientific, Inc.) was placed in the LV of two of the infarcted pigs to test simultaneous RT visualization of device and infarct using intermittent inversion pulses. RT imaging without saturation preparation was used to guide the device into the LV, as described in Ref. 9, using a slice depicting the ascending and descending aortas. When the device was positioned in the LV, intermittent inversion pulses were enabled to enhance infarcted regions while allowing continued visualization of the catheter for further navigation. The higher-resolution imaging parameters described above were used for these experiments.

The animal experiments were approved by the NHLBI Animal Care and Use Committee. All of the human research protocols were approved by the NHLBI Institutional Review Board, and the subjects gave written informed consent.

Simulations

We performed numerical simulations of the RT-IR-SSFP and IR-GRE pulse sequence variants used in the DHE experiments to gain insight into their relative performance. These simulations were run in MATLAB (Mathworks, Inc.) and employed iterative applications of the Bloch equations (10). The simulations consisted of delays and repetitions timed as in the experiments, using typical relaxation parameters for human tissue obtained from published sources (11) and data compiled from patient studies at the NIH. Contrast estimates were made for tissue at 20 min postcontrast at the point in time when the center of k-space was collected. Signal values from RT-IR-SSFP and IR-GRE simulations were normalized for comparison by means of background noise measured in representative images. The ratio of CNR values for IR-GRE and RT-IR-SSFP was then calculated as was done for the image data.

RESULTS

First-Pass Perfusion

Figure 2 shows selected frames from RT first-pass perfusion imaging after Gd-DTPA was injected intravenously in one of the human volunteers. Wall motion and first-pass perfusion defects are simultaneously visible in the image stream without ECG-gating or breath-holding, and correspond spatially (See the movie provided in supplementary material for this article, which can be found on the MRM website at http://www.interscience.wiley.com/jpages/0740-3194/suppmat/index.html).

FIG. 2

RT short-axis imaging shortly after systemic 0.2 mM/kg injection of Gd-DTPA in a human with previous MI (consecutive frames). Each view-shared SSFP image was preceded by a 90° nonselective saturation pulse. The image parameters were as (more ...)

We encountered no workflow problems in switching from perfusion to DHE imaging. The RT interactive imaging features allowed us to search for infarcted tissue by stepping through short-axis slices, switching to a long-axis view, and rotating about the LV long-axis. This search was enhanced by our ability to observe wall motion and DHE simultaneously.

DHE

In all of the experiments, infarcts were visible in both the IR-GRE and RT-IR-SSFP images, and corresponded spatially (see Figs. Figs.33 and and4).4). Table 1 shows the CNR comparison between RT-IR-SSFP and conventional IR-GRE. In the human experimental data, IR-GRE images exhibited 1.5 times higher CNR between infarcted and normal myocardium than RT-IR-SSFP, whereas it was comparable in the pig experimental data. Simulation results suggest that IRGRE will produce 2.4 times greater CNR than RT-IR-SSFP for this protocol with human tissue parameters.

FIG. 3

Image stream from RT-IR-SSFP showing DHE in an infarcted pig. Consecutive short-axis images covering two complete inversion cycles 28 min after injection with a TI of 123 ms, and 2.1 × 2.1 × 6 mm pixels. Infarcted myocardial tissue is (more ...)

FIG. 4

Image stream from RT-IR-SSFP showing DHE in a human with previous MI (not the same subject shown in Fig. 2). Consecutive short-axis images covering two complete inversion cycles, 12 min after injection with 2.0 × 1.8 × 6 mm pixels, and (more ...)

Table 1

Comparison Between RT-IR-SSFP and Conventional IR-GRE in Pig and Human Experiments. CNR Values are an Average of Images in all Experiment. The Ratios are Calculated Per Experiment and Then Averaged. IR-GRE Produced Better Infarct-Normal CNR in Humans, (more ...)

LV blood brightness was variable in conventional IRGRE, and almost always nulled in RT-IR-SSFP, which resulted in consistently greater contrast between infarct and blood with the RT imaging technique. The infarct area was overestimated in the RT-IR-SSFP images by 5–10%. This may have been due to temporal blurring, but the above-described sources for measurement error should also be considered.

We adjusted the TI interactively using RT controls to achieve optimal myocardial nulling (Fig. 5). Segmental akinesia was readily discernible in the view-shared images. Regions of DHE always corresponded spatially to regions in which wall motion abnormalities were seen. In a nontransmural infarct example (Fig. 6), DHE is observable with no associated absence of wall thickening in that region.

FIG. 5

RT interactivity facilitates manual optimization of TI (selected, nonconsecutive short-axis frames). Images were acquired from a human (same subject as in Fig. 4) 34 min after Gd-DTPA injection. The TI (displayed on the images) was adjusted in 25-ms increments (more ...)

FIG. 6

Nontransmural infarct in a long-axis view: comparison between RT-IR-SSFP without gating or breath-holding and conventional IR-GRE imaging in an infarcted pig, approximately 1 hr after Gd-injection. RT-IR-SSFP imaging (top row, consecutive images) was (more ...)

Interventional

In the active-catheter experiments, RT imaging with no saturation preparation provided satisfactory frame rate and image quality for catheter navigation along the aorta from femoral access into the LV. The images in Fig. 7 show that once the catheter was positioned in the LV, healthy and infarcted myocardium, cavitary blood, and catheter signal were all distinctly visible. The endocardial border in infarcted regions was often more discernible in the RT-IR-SSFP images due to better nulling of cavitary blood. In the IR-GRE images or the RT-SSFP without magnetization preparation, the infarcted tissue and cavitary blood often exhibited similar brightness levels, obscuring the endocardial border. The RT-IR-SSFP images provided improved feedback for guiding the catheter toward diseased myocardial tissue.

FIG. 7

Interventional RT imaging example with prototype MR-active catheter (Boston Scientific, Inc.) in an infarcted pig, long-axis view. Images a–c show navigation into the LV cavity using RT-SSFP with view-sharing. Images d–f were acquired (more ...)

DISCUSSION

Wall motion defects and sometimes signal enhancement in infarcted regions were discernible in the RT images obtained before Gd-DTPA injection. This helped us to choose the initial postinjection slice, and simplified the search for first-pass perfusion defects and DHE. The integrated approach to imaging all of these effects is thus beneficial from a workflow standpoint, and may reduce the total scan time.

It is apparent in some of the experiments (e.g., Fig. 3) that infarcted tissue is visible in the RT-SSFP images without any magnetization preparation (12). While we were able to demonstrate this consistently in the pig experiments with large infarct areas, smaller infarcts in the human studies were less readily apparent. This may be due to the close proximity of the infarct to the bright blood signal, and could perhaps be improved by the use of image postprocessing or blood saturation methods. To null the myocardium in steady-state imaging, one could use window/level adjustments or other postprocessing methods.

Since the RT-IR-SSFP image stream contains images of differing contrast, one could improve the user interface by displaying the myocardium-nulled image in a separate window from the view-shared images. However, for interventional applications, some observers found it preferable to view a single image stream containing the infarct enhancement, wall motion, and device in the same window. The myocardium-nulled image could also be color-coded and overlaid translucently on the view-shared images for more persistent guidance to infarcted regions.

For diagnostic applications, this technique could be extended to a multiple-slice RT method whereby images from each slice could be displayed in a separate window. With multiple-slice imaging, increased CNR would be expected, since the magnetization would recover more completely in a given slice while data are being acquired on other slices.

The longer data acquisition window used in RT imaging increases the difficulty in nulling the myocardium. It also increases temporal blurring of wall motion, and may be a cause of the observed overestimation of infarct size. One of the concerns about this RT technique is whether it can visualize small, nontransmural infarcts, given the effects of the longer acquisition window and the lower spatial resolution. Our initial experiences suggest that nontransmural infarcts are well visualized, but further study is needed to determine the smallest infarct that can be detected with the proposed technique. The data acquisition window can be shortened using parallel imaging methods, such as sensitivity encoding (SENSE) (13,14). Temporal SENSE (TSENSE) has been shown to exhibit less temporal blurring of wall motion than view-sharing (15), which may reduce this overestimation of infarct size. Parallel imaging should also eliminate the ghosting seen in the first view-shared image obtained after the myocardium-nulled image is acquired (16).

In the proposed technique, spatial and temporal resolution are sacrificed for increased interactivity during scanning without gating or breath-holding. A protocol that does not require breath-holding should increase comfort, especially for very sick patients, and could result in better compliance. In addition, there is potential clinical value in the ability to observe wall motion and DHE in the same scan. In some studies with small infarcts (e.g., Fig. 6), we observed DHE with no appreciable wall-thickening abnormality. This highlights the value of combining different types of cardiac MRI studies to obtain a more complete picture of an individual patient's heart structure and function.

The TIs required in RT-IR-SSFP for TD2 = 0 were shorter than those needed for IR-GRE. This is a consequence of playing the inversion pulse when tissues are at their SSFP values, before the magnetization is allowed to recover fully. This reduces the achievable SNR, but has the positive effect of increasing the overall frame rate, since a shorter TD2 reduces the required TI. However, SNR increases greatly when SSFP is used, resulting in satisfactory image quality. Another consequence of playing the nonselective inversion pulse with TD2 = 0 is that it is almost always possible to null both the normal myocardium and the LV blood pool. This facilitates identification of the endocardial border of infarcted regions.

CONCLUSIONS

RT interactive SSFP imaging with magnetization preparation can be used for simultaneous observations of infarcted myocardium by DHE, heart wall motion, and the position of an MR-active interventional device. With minor changes to imaging parameters, simultaneous visualizations of first-pass perfusion and wall motion may be achieved. Although spatial resolution is lower, and temporal blurring is greater than in conventional IR-GRE methods, ECG-gating and breath-holding are not required. This simplifies the procedure and may reduce the number of patients who are excluded from the imaging protocol. This fast interactive method has potential utility for qualitative infarct screening, as well as for interventional procedures that require targeting in or around infarcted regions.

Supplementary Material

Click here to view.^{(15K, pdf)}

ACKNOWLEDGMENTS

The authors thank Joni Taylor, Gina Orcino, and Diana Lancaster for their help in the animal experiments; Annette Stine and Victor Wright for their assistance in the patient studies; Scott Smith for the prototype catheters; and Al Zhang, Ph.D., for his help with pulse sequence programming issues.

Footnotes

^†This article is a US Government work and, as such, is in the public domain in the United States of America.

REFERENCES

1. Kim RJ, Fieno DS, Parrish TB, Harris K, Chen EL, Simonetti O, Bundy J, Finn JP, Klocke FJ, Judd RM. Relationship of MRI delayed contrast enhancement to irreversible injury, infarct age, and contractile function. Circulation. 1999;100:1992–2002. [PubMed]

2. Simonetti OP, Kim RJ, Fieno DS, Hillenbrand HB, Wu E, Bundy JM, Finn JP, Judd RM. An improved MR imaging technique for the visualization of myocardial infarction. Radiology. 2001;218:215–223. [PubMed]

3. Holsinger AE, Wright RC, Riederer SJ, Farzaneh F, Grimm RC, Maier JK. Real-time interactive magnetic resonance imaging. Magn Reson Med. 1990;14:547–553. [PubMed]

4. Oppelt A, Graumann R, Barfuβ H, Fischer H, Hartl W, Scajor W. FISP—a new fast MRI sequence. Electromedica. 1986;54:15–18.

5. Deimlink M, Heid O. Magnetization prepared True FISP imaging; Proceedings of the 2nd Annual Meeting of ISMRM; San Francisco. 1994. p. 495.

6. Schreiber WG, Schmitt M, Kalden P, Mohrs OK, Kreitner KF, Thelen M. Dynamic contrast-enhanced myocardial perfusion imaging using saturation-prepared TrueFISP. J Magn Reson Imaging. 2002;16:641–652. [PubMed]

7. Atkinson DJ, Burstein D, Edelman RR. First-pass cardiac perfusion: evaluation with ultrafast MR imaging. Radiology. 1990;174:757–762. [PubMed]

8. Kellman P, Arai AE, McVeigh ER, Aletras AH. Phase-sensitive inversion recovery for detecting myocardial infarction using gadolinium-delayed hyperenhancement. Magn Reson Med. 2002;47:372–383. [PMC free article] [PubMed]

9. Lederman RJ, Guttman MA, Peters DC, Thompson RB, Sorger JM, Dick AJ, Raman VK, McVeigh ER. Catheter-based endomyocardial injection with real-time magnetic resonance imaging. Circulation. 2002;105:1282–1284. [PMC free article] [PubMed]

10. Liang ZP, Lauterbur PC. Principles of magnetic resonance imaging: a signal processing perspective. SPIE Optical Engineering Press; Bellingham, WA: 2000. New York: IEEE Press.

11. Haacke EM. Magnetic resonance imaging: physical principles and sequence design. Wiley-Liss; New York: 1999.

12. Raff GL, Bartholomew BA, Gentry R, Shetty AN, Bis KG, Goldstein JA, O'Neill WW. Simultaneous evaluation of myocardial function, viability, and microvascular dysfunction by a new magnetic resonance technique; Proceedings of the 52nd Annual Scientific Session of ACC; Chicago. 2003. p. 1167.

13. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med. 1999;42:952–962. [PubMed]

14. Kellman P, Epstein FH, McVeigh ER. Adaptive sensitivity encoding incorporating temporal filtering (TSENSE) Magn Reson Med. 2001;45:846–852. [PubMed]

15. Guttman MA, Kellman P, Dick AJ, Lederman RJ, McVeigh ER. Real-time accelerated interactive MRI with adaptive TSENSE and UNFOLD. Magn Reson Med. 2003;50:315–321. [PMC free article] [PubMed]

16. Kellman P, McVeigh ER. Ghost artifact cancellation using phased array processing. Magn Reson Med. 2001;46:335–343. [PMC free article] [PubMed]