Author + information
- Leon Axel, PhD, MD and
- Daniel K. Sodickson, PhD, MD∗ ()
- Bernard and Irene Schwartz Center for Biomedical Imaging, Department of Radiology, New York University School of Medicine, New York, New York
- ↵∗Reprint requests and correspondence:
Dr. Daniel K. Sodickson, Department of Radiology, New York University School of Medicine, 660 First Avenue, Fourth Floor, Room 407, New York, New York 10016.
Assessment of cardiac function is a cornerstone of the diagnosis and care of patients with heart disease. The use of various imaging methods for measuring cardiac function has become increasingly important in both clinical practice and research. In particular, cardiac magnetic resonance (CMR) has become the de facto standard for assessment of global cardiac function through measurement of the change in ventricular volumes over the cardiac cycle and calculation of the associated ejection fraction (EF). This reflects the strengths of CMR, including its ability to generate a set of high-quality tomographic images, synchronized with different phases of the cardiac cycle, with accurately known 3-dimensional (3D) spatial relationships, thus permitting reliable segmentation of the boundary contours of the ventricular walls and calculation of the corresponding ventricular volumes. These global function measurements can be combined in a single imaging session with the use of other CMR-derived measures of, for example, regional and functional cardiac anatomy, blood flow, myocardial perfusion, and other tissue characteristics such as signal relaxation times and late gadolinium enhancement. Thus, CMR has long been touted as a sort of “1-stop shop” for cardiac evaluation. However, there are still some important limitations in the use of CMR for measurement of global function and calculation of EF:
1. The relatively long time required to acquire conventional CMR data requires that the necessary volumetric data be pieced together over multiple breath-holds, resulting in prolonged examination times and potential inconsistencies among different breath-hold periods.
2. The inside of the ventricular wall is rough, due to the presence of trabeculations and papillary muscles. Although, by convention, these irregular structures are generally included with the cavity, it is more difficult to delineate them at end-systole, potentially introducing a bias into the calculated end-systolic volumes and the EF.
3. The frequent presence in cardiac patients of cardiac arrhythmias and difficulty with breath-holding can lead to associated decreases in image quality.
4. The need to interact frequently with CMR analysis programs to manually correct errors in automated image segmentation introduced by the imaging limitations mentioned in the previous text increases the time needed to carry out the analysis and decreases its reproducibility.
In cardiac imaging more than in many other settings, the goal is simultaneous characterization of complex 3D tissue structure and dynamics; therefore, imaging speed is of the essence. Various classes of methodological and technological advances have been brought to bear over the years to accelerate CMR. These include a menagerie of advanced pulse sequences and hardware improvements. The advent of parallel magnetic resonance imaging, in which arrays of radiofrequency coils are used to gather essential data simultaneously rather than in a traditional sequential order, has circumvented previous speed limits, and parallel imaging techniques have played a key role in modern CMR examinations. A recent and highly promising addition to the armamentarium of rapid CMR is compressed sensing (CS), which combines incoherently undersampled data acquisition with sparsity-enforcing image reconstruction. CS may be considered as a form of image pre-compression, in which time-consuming data acquisition steps are omitted beforehand with knowledge that key information will be recovered robustly in reconstruction. Soon after the seminal work by Donoho, Candès, and Tao introducing CS appeared in the applied mathematics literature, CS methods were applied to magnetic resonance imaging by Lustig et al. (1). Since then, interest in CS for accelerated imaging has burgeoned, various productive combinations with other acceleration techniques such as parallel imaging have been developed, and numerous practical implementations for CMR have been demonstrated. However, despite vigorous research interest in CS methods, to date there have been comparatively few evaluations in a truly clinical setting. The current study is, therefore, a welcome addition to the literature.
In this issue of iJACC, Vincenti et al. (2) report on a CS-based approach to acquire multislice CMR data in a single breath-hold, together with a 3D model–based approach to calculate global function measures from the resulting dataset. Images were qualitatively assessed for overall quality and the presence of artifacts, and the derived function measurements were validated against comparable results from both standard volume analysis of conventionally acquired CMR data of the ventricles and integrated CMR flow data acquired in the aortic root.
The reported results are encouraging, with reasonable-looking image quality and excellent agreement of the new functional measurements with those derived from the more conventional approaches. In addition to the time savings of the new approach, the acquisition of complete datasets in a single breath-hold allows for better spatial registration between individual slices, potentially enabling more accurate volume calculations. The model-based volumetric image analysis method used here provides both an attractive 3D display of the underlying ventricular structure and a means of regularizing volume calculations despite a relatively coarse spatial sampling.
There are also some limitations of the current study:
1. As the researchers acknowledge, the coarse spatial sampling used here would not be adequate for assessment of the regional anatomy and function of the heart and thus would not eliminate the need for acquiring a conventional set of more closely spaced images.
2. There were some systematic biases toward decreased estimates of end-diastolic volume and associated increased estimates of EF, as might be expected from the somewhat lower resolution of the CS-based imaging used.
3. Although the relatively short acquisition time needed for each slice (2 heartbeats) may decrease the image-degrading effects of respiratory motion or arrhythmias, it will not eliminate them and will not capture associated real variations in the underlying functional measures.
4. The computation-intensive nature of iterative CS-based image reconstruction makes it a relatively lengthy process. However, as suggested by the researchers, the use of more powerful specialized computers, such as graphics processing units, is likely to make this less of a practical problem.
5. The use of global measurements derived from aortic flow as a reference can be problematic, due to the frequent presence of uncontrolled baseline offsets in the flow measurements.
Although the results of Vincenti et al. (2) are already encouraging, the exploration of CS-based approaches to imaging and image analysis is an active and rapidly expanding area. Although the researchers used a conventional Cartesian-based sampling of the CMR data, non-Cartesian sampling offers potential advantages. For example, radial sampling patterns (3) reduce the impact of aliasing from structures in the outer parts of the field of view (a problem encountered in the current study). Radial acquisitions have been shown to be inherently well suited for robust CS, and they are also compatible with continuous acquisition approaches (e.g., employing golden-angle schemes) (4). CS-based methods may also enable significant speed-up of flow measurements, which could both reduce the time required and make the extension of such imaging to more fully 3D volumetric flow measurements more practical. Other CS-based CMR approaches also have the potential to substantially reduce the deleterious effects of respiratory motion and arrhythmias on the images, and they have even been shown to enable robust characterization of irregular cardiac and respiratory motion patterns together with improved structural and functional assessments (5).
In summary, the paper of Vincenti et al. (2) provides welcome evidence of some of the potential value of CS CMR for improved cardiac function assessment, but it is only the barest beginning. In fact, there are early indications that a more complete integration of advanced acceleration methods into everyday imaging practice may be in store, in the form of a new paradigm of rapid, continuous comprehensive imaging. Such an approach would be both simpler and more efficient than traditional multiplanar CMR protocols, eliminating “dead time” between separate specialized acquisitions and allowing extraction of multiple dynamic as well as tissue contrast parameters simultaneously. This approach, enabled by key advances such as CS, would revolutionize cardiac imaging workflow while enabling increasingly rich characterization of both structure and function in the human heart.
↵∗ Editorials published in JACC: Cardiovascular Imaging reflect the views of the authors and do not necessarily represent the views of JACC: Cardiovascular Imaging or the American College of Cardiology.
Drs. Axel and Sodickson have contributed to various patent applications in the area of rapid imaging using compressed sensing methods.
- American College of Cardiology Foundation
- Vincenti G.,
- Monney P.,
- Chaptinel J.,
- et al.
- Feng L.,
- Grimm R.,
- Block K.T.,
- et al.
- Feng L.,
- Axel L.,
- Latson L.A.,
- Xu J.,
- Sodickson D.K.,
- Otazo R.