Ashitha Pathrose1, Hassan Haji-Valizadeh1,2, Roberto Sarnari1, James Carr1,2,3, and Daniel Kim1,2
1Radiology, Northwestern University, Chicago, IL, United States, 2Biomedical Engineering, Northwestern University, Chicago, IL, United States, 3Medicine, Northwestern University, Chicago, IL, United States
Synopsis
Cardiac magnetic resonance feature-tracking (CMR-FT) has emerged as a
reference standard for the evaluation of cardiac morphology and function. But, the
performance of CMR-FT may be affected by data undersampling as done in several
real-time cardiac MRI techniques. In our study, we evaluated the performance of
CMR-FT for myocardial strain quantification from real-time cardiac cine MRI when compared to standard cardiac cine in cardiac disease patients. We found that
even though there is good agreement between the values derived from the real-time
and standard cine MRI, care should be taken as measures from the real-time cine can
be underestimated.
Introduction
Cardiac magnetic resonance feature-tracking (CMR-FT) has recently emerged as a useful tool for the quantitative evaluation of cardiovascular function owing to its ability to assess regional myocardial function impairment and lack of need
for a separate imaging sequence such as myocardial tagging, DENSE, or SENC. CMR-FT strain measures are usually derived from ECG-gated, breath-hold balanced steady-state free precession (bSSFP) cine acquisitions. This acquisition has several limitations including long scan-times and compromised image quality in patients
with arrhythmias or insufficient breath-holding capacity. Real-time cardiac MRI is an alternative technique that
allows dynamic imaging of the heart without the need for ECG synchronization or
breath-holding. But undersampling of the images in these real-time
techniques may affect the CMR-FT algorithms that rely on recognizing features or patterns in the images. Thus, the aim of this study was to evaluate the performance of CMR-FT based myocardial strain quantification from a 16-fold accelerated, real-time
cardiac cine MRI with radial k-space sampling and compressed sensing (CS) in patients with cardiac disease at 1.5T and 3T.Methods
Study cohort: Thirty-four
patients underwent standard breath-hold cardiac cine MRI and real-time cardiac
cine MRI acquisitions. The patients imaged belonged to two disease cohorts: prior
myocardial infarction group (MI; n = 15; mean age,
59.1 ± 12.6 y; male, n = 11; female, n = 4; mean heart
rate, 66.9 ± 7.0 bpm; left ventricular ejection fraction, 21%–70%) and chronic kidney disease group (CKD; n = 19; mean age, 63.9 ± 16.2 y; male, n = 9; female, n = 10;
mean heart rate, 65.8 ± 7.3 bpm; left ventricular ejection fraction, 46%–79%).
MRI data acquisition and reconstruction: The first cohort of MI patients were scanned on a 1.5T whole‐body MRI scanner
(Aera; Siemens). The second cohort of CKD patients were scanned on a 3T whole‐body MRI scanner
(Skyra; Siemens). Image acquisition parameters are detailed in table 1. Undersampled
data were reconstructed using the GRASP (radial CS) framework [1, 2]. GRASP reconstructions
were performed offline in MATLAB (MathWorks).
CMR-FT analysis: CMR-FT
was used to measure the left ventricular myocardial peak radial strain (Err) and
circumferential strain (Ecc) from the short-axes images from both these
acquisitions (cvi42, Circle). On all
images, the epicardial and endocardial borders were outlined in the
end-diastole (figure 1). The software then automatically propagated the contours throughout the remainder of the cardiac cycle. A
second investigator also evaluated 15 studies for the assessment of
inter-observer reproducibility. Pearson correlation coefficients and
Bland-Altman plots were used to assess agreement between the two methods. Interobserver reproducibility was assessed using
Intraclass Correlation Coefficient (ICC) (SPSS, IBM).Results
The standard breath-hold cine MRI obtained from four out of the 34
patients included in the study were not analyzable because of artifacts due
to arrhythmia. The real-time cine images of these patients were artifact-free
and showed reasonable strain values, but were excluded form statistical
analysis. Strain plots at peak systole for a representative subject is shown in
figure 2. There was good correlation and agreement between the strain
measurements derived from the standard breath-hold cine and the real-time cine;
Err [r2=0.69; p<.001; bias: -5.2%, 95% CI limits of agreement
(LOA): -28.2 to 17.7%] and Ecc [r2=0.72; p<.001; bias: 0.9%, 95%
CI limits of agreement (LOA): -5.2 to 6.9%] (figure 3). The mean Err (standard
cine: 40.1±20.5%; real-time cine: 35.3±14.6%; p<.001) and Ecc (standard
cine: -20.1±5.8%; real-time cine: -19.4±4.8%; p=.01) was significantly
underestimated by the real-time cine compared to the standard cine. The median
percentage underestimation for Err was -8.5% (IQR: -23.1% to 6.9%) and for Ecc
was -3.8% (IQR: -13.0% to 3.6%). There was excellent interobserver agreement for
the Err (ICC = 0.94, 95% CI LOA: 0.88 to 0.96) and Ecc (ICC = 0.93, 95% CI LOA:
0.88 to 0.96).Discussion
CMR-FT techniques falls in a general category of image
post-processing methods known as optical flow [3, 4]. The tracking method identifies a relatively small tracking window on one image and then searches for the most comparable image pattern in a window of
the same size in the subsequent frame. In CS reconstruction, regularization may
suppress features that are necessary for optical flow to track, reducing accuracy
of measurements. Low temporal resolution and
spatial resolution also negatively affects the tracking efficiency. Radial
strain is computed on the small distance between endo and epicardium, and was affected more than circumferential strain that are computed over larger regions
[5].Conclusion
This study demonstrates that CMR-FT derived strain measurements from a 16-fold
accelerated, real-time cine MRI with compressed-sensing is comparable to the
measures derived from standard breath-hold cine MRI. Although good agreement
was found between the two sequences, it should be kept in mind that strain measurements from undersampled cardiac MRI can be underestimated and is
dependent on the spatiotemporal resolution of the sequence and regularization induced by CS.Acknowledgements
No acknowledgement found.References
1.
Feng,
L., et al., Golden-angle radial sparse parallel MRI: combination of compressed
sensing, parallel imaging, and golden-angle radial sampling for fast and
flexible dynamic volumetric MRI. Magn Reson Med, 2014. 72(3): p. 707-17.
2.
Haji-Valizadeh,
H., et al., Validation of highly accelerated real-time cardiac cine MRI with
radial k-space sampling and compressed sensing in patients at 1.5T and 3T.
Magnetic Resonance in Medicine, 2018. 79(5): p. 2745-2751.
3.
Singh,
A., Optic flow computation : a unified perspective. 1991, Los Alamitos, Calif.:
New York : IEEE Computer Society Press, Institute of Electrical and Electronics
Engineers.
4.
Barron,
J.L., D.J. Fleet, and S.S. Beauchemin, Performance of optical flow techniques.
International Journal of Computer Vision, 1994. 12(1): p. 43-77.
5.
Langeland,
S., et al., Experimental assessment of a new research tool for the estimation
of two-dimensional myocardial strain. Ultrasound Med Biol, 2006. 32(10): p.
1509-13.