Fast and fully automatic calibration of frequency offset for balanced steady-state free precession cardiovascular magnetic resonance at 3.0 Tesla
© Tang et al.; licensee BioMed Central Ltd. 2013
Received: 27 June 2012
Accepted: 20 March 2013
Published: 11 April 2013
This study proposed a fast and fully automatic calibration system to suppress the dark banding artifacts in balanced steady-state free precession (bSSFP) cardiovascular magnetic resonance (CMR) at 3.0 T.
Twenty-one healthy volunteers (18 men, 3 women; mean age 24.9 years) participated in this study after providing institutionally approved consent. The optimal frequency was obtained using sweep scans of transition-band low flip-angle bSSFP (bSSFP-L), performed with three conditions: breath-hold plus electrocardiography (ECG) triggering (BH + ECG), breath-hold only (BH), and free breathing (FB). A real-time feedback system was implemented to allow the performing of bSSFP-L calibration scanning and conventional cine bSSFP within one breath-hold. For each scan condition, the optimal phase was estimated using 20-point and 10-point spline fitting.
Linear regression analysis indicated high correlation between the optimal phases obtained using BH and FB and those obtained using BH + ECG (R2 = 0.91 to 0.98, n = 21). The optimal phases obtained using 10-point datasets showed high correlation with the 20-point BH + ECG datasets (R2 = 0.92 to 0.99, n = 21); although the within-subject coefficient of variation (wsCV) was larger using 10-point fitting. The variation of repeated measurements was largest with FB acquisition and smallest with BH + ECG acquisition. The optimal frequency obtained by offline calculation and by real-time feedback calibration significantly reduced dark-band artifacts in cine bSSFP images (both p < .01).
The proposed real-time feedback calibration method for bSSFP imaging is rapid and fully automatic. This method could greatly reduce dark-band artifacts in bSSFP images and facilitate clinical CMR at 3.0 T.
Cardiovascular magnetic resonance (CMR) has become a practical tool for evaluating cardiac anatomy and function. Of all CMR techniques, balanced steady-state free precession (bSSFP)  combines the advantages of sub second scan time, high fluid-tissue contrast, and three-dimensional imaging compatibility. Therefore, the bSSFP sequence has become a major imaging method in clinical CMR protocols . The applications include, but are not limited to, coronary MR angiography , accelerated dobutamine stress CMR , diagnostics of congenital heart diseases , and the detection of myocardial infarction using late gadolinium enhancement CMR .
The most prominent artifact uniquely found in bSSFP images is the dark-band artifact arising because of global heterogeneity of the main magnetic field. This occurs because the steady-state signal intensity is a function of the reference offset angle, which is the phase evolution within one repetition time (TR) relative to the radio-frequency (RF) pulse. The location-dependent off-resonance can result in local variations in the SSFP signal and, hence, non-uniformity in signal intensity. Such a phenomenon is referred to as dark-band artifacts. In cardiac imaging for in vivo human study, the magnetic field is generally inhomogeneous at the heart-lung interface . At 3.0 T, the effect is more prominent; dark-band artifacts thus commonly occur in bSSFP-based CMR images.
To solve this problem, localized high-order shimming  can reduce the field inhomogeneity. Frequency scout scans  can also be used to identify the optimum frequency offset, which further minimizes the presence of the dark-band artifacts in the cardiac structures of interest [9–11]. The frequency scout is a discrete search approach. It acquires bSSFP images with varying RF frequency offsets. The operator observes the image appearances to determine the frequency offset for the subsequent bSSFP scans. However, the precision (i.e., step size of RF frequency offset) of scout scans is limited by the breath-hold time. The user interaction required to identify the optimal frequency offset also delays the clinical routine. If the frequency scout scan is performed only at the beginning of the study, the potential temporal magnetic field instability caused by the heating of the passive shims , gradient system [13, 14], or chest motion [15, 16] might degrade the precision of frequency calibration.
This study presents a method that can rapidly and automatically achieve frequency calibration by proposing a real-time feedback system to obtain the optimal frequency offset immediately after a calibration scan. The study aims to consecutively perform frequency calibration and clinical bSSFP imaging within one breath-hold, and to facilitate clinical routine practice and improve the quality of cardiac images.
Low-flip angle transition-band bSSFP sequence
The transition-band bSSFP with low flip angle (bSSFP-L) detects brain functional activations, as reported by Miller et al. . The sequence removes the RF phase alternation of the conventional bSSFP sequence and uses a relatively small flip angle (typically 3 to 5°). The sequence detects the minute frequency shifts caused by changes in the concentration of deoxyhemoglobin [13, 17]. The present study proposes to use the frequency sensitivity of bSSFP-L to detect field inhomogeneity in the cardiac structures of interest. The major steps for estimating the optimal frequency offset using bSSFP-L are performing a sweep scan with varying RF phase angle (β, the phase difference between consecutive RF pulses), plotting an intensity profile of the cardiac region against β, approximating the optimal β corresponding to the global maximum of the intensity profile, and obtaining frequency offset Δf using Δf = β/360TR.
In vivo imaging experiment: bSSFP-L sweep scan and cine cardiac imaging
All imaging experiments were performed using a 3.0 Tesla whole-body MR system (Siemens, Tim Trio, Germany) equipped with a 12-channel cardiac coil. The study was approved by our institution's Internal Review Board Committee. Twenty-one volunteers (18 men, 3 women; aged 24.9 ± 3.17 years) participated in the experiments after providing informed consent. Scout localizers were first obtained and a midlevel short-axis slice was selected. The scan parameters of a bSSFP-L sweep scan were TR/TE = 4/2 ms, flip angle = 4°, matrix = 64 × 64, FOV 280 mm, BW = 555 Hz/pixel, and slice thickness = 6 mm. Each sweep scan yielded 20 images using β = -180 to 162°, with a step size of 18° (equivalently, Δf = - 125 to 112.5 Hz, with a step size of 12.5 Hz), to cover a cycle of potential bSSFP off-resonance phase angles. The steady state was not established before image acquisition. Shimming was accomplished before the patient was first scanned with the bSSFP-L sequence. The operator positioned the FOV for the heart region to be located at the center of the acquired image. Three types of bSSFP-L sweep scan were performed: breath-hold plus ECG triggering (BH + ECG, trigger delay = 130 ms), breath-hold (BH), and free breathing (FB). The BH + ECG method was considered the reference standard to justify if BH or FB methods are appropriate for frequency calibration using the bSSFP-L method. When performing the breath-hold experiment, the operator asked participants to hold their breaths at the end of expiration. The total acquisition times for BH + ECG, BH, and FB were 20 heartbeats, 5.4 s, and 5.4 s, respectively. Each type of sweep scans was performed 3 times to assess reproducibility. All sweep scans were performed consecutively to prevent frequency drift over time. After the sweep scans, the optimal Δf was calculated using an external computer. The calculation procedures are presented in a later section.
Image processing: automatic region of interest selection and calculating optimal frequency offset
Subsequently, the procedure calculated the averaged intensities over the heart ROI of a bSSFP-L image dataset to plot a signal intensity curve against Δf (hereafter termed sweep scan curve). The samples of the sweep scan curve were increased using the cubic spline interpolation technique. Finally, Δf, corresponding to the peak of the interpolated curve, was regarded as the optimal frequency offset. In real-time calibration, a shorter total acquisition time facilitates its clinical application. The procedures were, therefore, repeated to calculate Δf using only 10 images (Δf = - 125 to 100 Hz, with a step size of 25 Hz) of the sweep scan. The 10-image datasets are referred to as 10-point (10-pt) datasets hereafter.
where σi and were the standard deviation and mean, respectively, of the three repeats on subject i. N was the number of subjects included in the calculation. Out of 126 sets of repeated measurements (21 subjects × 6 = 126), four were excluded because their means were very close to zero where wsCV is mathematically undefined. The combination of BH + ECG acquisition and 20-pt spline fitting was considered as the reference standard to which other five methods were compared using linear regression analysis including the four cases previously removed from wsCV calculation.
Real-time feedback system: automatic frequency calibration and bSSFP imaging
Real-time feedback system: in vivo experiments
Ten participants (male, aged 24.8 ± 4.9 years) participated in the real-time calibration experiments. They underwent three multiphase cine bSSFP scans without frequency calibration, with frequency-scout calibration, and with bSSFP-L calibration. The cine scans acquired the same short-axis slice using the following scan parameters: TR/TE = 3.6/1.8 ms, flip angle = 49°, matrix = 256 × 256, FOV = 300 mm, bandwidth = 930 Hz/pixel, ECG-gated, number of cardiac phase = 25, and two-fold GRAPPA  acceleration with 44 reference lines. For the bSSFP-L real-time feedback calibration, a 10-pt bSSFP-L sweep scan (Δf = -125 to 100 Hz, with a step size of 25 Hz, scan time = 2.7 s) and a cine bSSFP scan were performed consecutively with a 1-s time delay. The two imaging sequences were set to acquire the slice at an identical position. A midlevel short-axis slice was selected for each volunteer. Additional slices (short axis, long axis, four-chamber view, right ventricular outflow tract) were acquired for two of the volunteers. The participants were asked to hold their breath during both scans. To calculate the total processing time of the feedback system, a trigger signal synchronized with the last excitation pulse of the bSSFP-L sequence was transmitted using fiber-optic cabling to the dedicated PC.
Approximate assessment of dark-band artifacts
To assess the dark-band artifacts of the three cine datasets (without frequency calibration, with frequency-scout calibration, and with bSSFP-L calibration), an ROI covering the myocardium and blood pool of the left ventricle (LV) was manually outlined for each cardiac phase of the cine datasets. The average ROI signal intensity was also calculated. A pixel inside the ROI was classified as part of a dark-band artifact when the signal intensity was below a certain threshold. Although the simple bisection method can be employed to detect the banding artifacts, it may not be able to distinguish dark-band pixels from tissues with inherently low signals such as the papillary muscle. Therefore, we used six thresholds to assess the dark-band pixels to compare the performance of the frequency calibration methods. Thresholds #1, #2, and #3 were 10%, 15%, and 20% of the average signal intensity of the ROI in each dataset, respectively. Threshold #4 was the average value of the threshold #1 values for the three datasets (without frequency calibration, with frequency-scout calibration, and with bSSFP-L calibration). Similarly, thresholds #5 and #6 were obtained from thresholds #2 and #3, respectively.
Linear regression analysis against 20-pt BH + ECG datasets
95% C.I. (Hz)
BH + ECG
Summary of within-subject coefficient of variation (wsCV) computed for six combinations of imaging acquisition methods (BH + ECG, BH, FB) and curve-fitting schemes (20-pt and 10-pt spline fitting)
BH + ECG
This study implemented a fully automatic and rapid frequency calibration method to reduce dark-band artifacts in bSSFP imaging, by taking advantage of the sensitivity the bSSFP-L sequence to minute frequency drifts. Results indicated the feasibility of performing the sweep scan obtaining Δf without ECG triggering. The entire calibration time was further reduced by using a sparser sampling scheme (10-pt sweep scan) and a real-time feedback system. The shortened calibration time allows this method to serve as a pre-adjustment module for bSSFP imaging sequences. It is feasible to perform the calibration scan and bSSFP imaging within one breath-hold.
Considering the breath-hold period, this study used 20-pt sweep scans with BH + ECG to increase the precision of Δf estimation, using the acquired 20-pt BH + ECG datasets as reference standards with which to judge the performance of the remaining data sets. Quantitative assessment of cine data demonstrated a significant reduction in dark-band artifacts following calibration using the 20-pt BH + ECG method. The linear regression analysis showed high correlation between the 20-pt BH datasets and the 20-pt BH + ECG datasets (R2 = 0.98, slope = 0.99). Variation analysis of repeated measurements supports that the reproducibility of the 20-pt BH method is comparable to that of the 20-pt BH + ECG method. This result indicates that cardiac motion minimally affects the signal intensities of bSSFP-L images. Using ROI-based analysis, the optimal Δf can be accurately obtained without ECG triggering. Without using ECG triggering, the image acquisition time of the 20-pt sweep scan reduces from 20 heartbeats to 5.4 s. The 20-pt FB datasets show a high correlation with the 20-pt BH + ECG datasets, albeit with lower slope (R2 = 0.91, slope = 0.89) and relatively higher variation of the repeated measurements (wsCV = 23.7%). This result indicates that respiratory motion reduces precision and reproducibility of the Δf measurement. These comparisons suggest that the BH method, which provides greater precision (R2 = 0.98, slope = 0.97, wsCV = 21.1%) and fast data acquisition, is the optimal choice for frequency calibration using bSSFP-L sweep scans. The FB method could be particularly useful in patients with limited breath-hold capacity.
To further reduce image acquisition time, this study investigated the feasibility of obtaining accurate frequency offset with a sparser sampling scheme, the 10-pt sweep scan method. Analyses of linear regression and reproducibility suggested that the 10-pt BH method is capable of producing accurate and reliable Δf, which is comparable to that obtained using the 20-pt BH + ECG method. The shortened acquisition time of the 10-pt BH method allows the performing of real-time frequency calibration and clinical bSSFP imaging within one breath-hold. Using the 10-pt BH method shortened the total data acquisition time of the Δf measurement to 2.7 s. The total processing time of the feedback system (data networking and software calculation of Δf) is less than 1 s. The results indicate that the real-time feedback system can calibrate frequency offset for the subsequent conventional bSSFP imaging and substantially reduce the dark-band artifact in the bSSFP images. Frequency calibration for each bSSFP scan in the clinical protocol is, therefore, feasible. It is possible to obtain optimal Δf values for different slice locations and orientations. In addition, calibration of each bSSFP scan in the clinical protocol could potentially enable the tracking of frequency drift during a long study.
Using the 10-pt method, the frequency resolution is 25 Hz (step size of β = 36° and TR = 4 ms). In clinical practice, the frequency scout method obtains optimal results while matching the parameters of the scout scan and bSSFP imaging, such as TR and bandwidth. The purpose of the proposed bSSFP-L method is to calculate the off-resonance frequency around the heart. Theoretically, this frequency would be unaffected by the modulation of imaging parameters. However, for other factors that would affect the bSSFP signal such as the eddy current, RF imperfection, and flow effect, matching the imaging parameters of the cine sequence should be an optimal approach.
Our data indicate that the performance of the proposed bSSFP-L method matches that of the conventional frequency scout method. The main advantages of the bSSFP-L method are the fully automatic procedure and rapid acquisition. The sweep scan curve obtained using the bSSFP-L sequence without an ECG trigger is affected minimally by cardiac motion. In contrast, the sweep scan curve obtained using the conventional bSSFP frequency scout without an ECG trigger shows prominent oscillations. Consequently, it is relatively difficult to achieve an automatic procedure for the bSSFP datasets, especially when a sparse sampling scheme (10-pt method) is used and the ECG trigger is turned off.
The limitation of this study is that the validation of the proposed method is performed only on healthy young adults. A potential pitfall of this method is the automatic ROI selection procedure. Whether the ROI selection algorithm is still robust on patients with cardiac dysfunction remains unanswered. The proposed algorithm requires the operator to position the heart region at the center of the FOV. This could lead to wraparound artifacts, especially when parallel imaging is used. Nonetheless, it is potentially feasible to implement a tool similar to the FOV of the “localized volume shim” on the user interface of the MR scanner console to position a rectangle covering the heart region. This procedure could further improve the robustness of the frequency calibration and avoid the restriction of the FOV position.
In summary, the proposed real-time calibration method, which rapidly and accurately optimizes the frequency offsets of bSSFP imaging, is fully automatic and can facilitate clinical bSSFP cardiac imaging and improve image quality, especially at 3.0 T.
Supported by the National Science Council under grant NSC-100-2628-R-011-001.
- Oppelt A, Graumann R, Barfuss H, Fischer H, Hartl W, Shajor W: FISP - A new fast MRI sequence. Electromedica. 1986, 54: 15-18.Google Scholar
- Finn JP, Nael K, Deshpande V, Ratib O, Laub G: Cardiac MR imaging: state of the technology. Radiology. 2006, 241 (2): 338-354. 10.1148/radiol.2412041866.View ArticlePubMedGoogle Scholar
- Katoh M, Stuber M, Buecker A, Gunther RW, Spuentrup E: Spin-labeling coronary MR angiography with steady-state free precession and radial k-space sampling: initial results in healthy volunteers. Radiology. 2005, 236 (3): 1047-1052. 10.1148/radiol.2363040865.View ArticlePubMedGoogle Scholar
- Jahnke C, Paetsch I, Gebker R, Bornstedt A, Fleck E, Nagel E: Accelerated 4D dobutamine stress MR imaging with k-t BLAST: feasibility and diagnostic performance. Radiology. 2006, 241 (3): 718-728. 10.1148/radiol.2413051522.View ArticlePubMedGoogle Scholar
- Makowski MR, Wiethoff AJ, Uribe S, Parish V, Botnar RM, Bell A, Kiesewetter C, Beerbaum P, Jansen CH, Razavi R, Schaeffter T, Greil GF: Congenital heart disease: cardiovascular MR imaging by using an intravascular blood pool contrast agent. Radiology. 2011, 260 (3): 680-688. 10.1148/radiol.11102327.View ArticlePubMedGoogle Scholar
- Ibrahim T, Hackl T, Nekolla SG, Breuer M, Feldmair M, Schomig A, Schwaiger M: Acute myocardial infarction: serial cardiac MR imaging shows a decrease in delayed enhancement of the myocardium during the 1st week after reperfusion. Radiology. 2010, 254 (1): 88-97. 10.1148/radiol.09090660.View ArticlePubMedGoogle Scholar
- Atalay MK, Poncelet BP, Kantor HL, Brady TJ, Weisskoff RM: Cardiac susceptibility artifacts arising from the heart-lung interface. Magn Reson Med. 2001, 45 (2): 341-345. 10.1002/1522-2594(200102)45:2<341::AID-MRM1043>3.0.CO;2-Q.View ArticlePubMedGoogle Scholar
- Schar M, Kozerke S, Fischer SE, Boesiger P: Cardiac SSFP imaging at 3 Tesla. Magn Reson Med. 2004, 51 (4): 799-806. 10.1002/mrm.20024.View ArticlePubMedGoogle Scholar
- Deshpande VS, Shea SM, Li D: Artifact reduction in true-FISP imaging of the coronary arteries by adjusting imaging frequency. Magn Reson Med. 2003, 49 (5): 803-809. 10.1002/mrm.10442.View ArticlePubMedGoogle Scholar
- Wieben O, Francois C, Reeder SB: Cardiac MRI of ischemic heart disease at 3 T: potential and challenges. Eur J Radiol. 2008, 65 (1): 15-28. 10.1016/j.ejrad.2007.10.022.View ArticlePubMedGoogle Scholar
- Wansapura J, Fleck R, Crotty E, Gottliebson W: Frequency scouting for cardiac imaging with SSFP at 3 Tesla. Pediatr Radiol. 2006, 36 (10): 1082-1085. 10.1007/s00247-006-0255-6.View ArticlePubMedGoogle Scholar
- Henry PG, van de Moortele PF, Giacomini E, Nauerth A, Bloch G: Field-frequency locked in vivo proton MRS on a whole-body spectrometer. Magn Reson Med. 1999, 42 (4): 636-642. 10.1002/(SICI)1522-2594(199910)42:4<636::AID-MRM4>3.0.CO;2-I.View ArticlePubMedGoogle Scholar
- Miller KL, Smith SM, Jezzard P, Pauly JM:High-resolutionat FMRI, at 1.5 T using balanced SSFP. Magn Reson Med. 2006, 55 (1): 161-170. 10.1002/mrm.20753.View ArticlePubMedGoogle Scholar
- Scheffler K, Seifritz E, Bilecen D, Venkatesan R, Hennig J, Deimling M, Haacke EM: Detection of BOLD changes by means of a frequency-sensitive trueFISP technique: preliminary results. NMR Biomed. 2001, 14 (7–8): 490-496.View ArticlePubMedGoogle Scholar
- Jezzard P, Clare S: Sources of distortion in functional MRI data. Hum Brain Mapp. 1999, 8 (2–3): 80-85.View ArticlePubMedGoogle Scholar
- Pfeuffer J, Van de Moortele PF, Ugurbil K, Hu X, Glover GH: Correction of physiologically induced global off-resonance effects in dynamic echo-planar and spiral functional imaging. Magn Reson Med. 2002, 47 (2): 344-353. 10.1002/mrm.10065.View ArticlePubMedGoogle Scholar
- Wu ML, Wu PH, Huang TY, Shih YY, Chou MC, Liu HS, Chung HW, Chen CY: Frequency stabilization using infinite impulse response filtering for SSFP fMRI at 3 T. Magn Reson Med. 2007, 57 (2): 369-379. 10.1002/mrm.21138.View ArticlePubMedGoogle Scholar
- Nock R, Nielsen F: On weighting clustering. IEEE Trans Pattern Anal Mach Intell. 2006, 28 (8): 1223-1235.View ArticlePubMedGoogle Scholar
- Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A: Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med. 2002, 47 (6): 1202-1210. 10.1002/mrm.10171.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.