Respiration-Induced B0 Field Fluctuation Compensationin Balanced SSFP: Real-Time Approach for Transition-Band SSFP fMRI

Jongho Lee,1 Juan M. Santos,1 Steven M. Conolly,2 Karla L. Miller,3

Brian A. Hargreaves,4 and John M. Pauly1

In functional MRI (fMRI) the resonance frequency shift inducedfrom respiration is a major source of physiological noise. Intransition-band SSFP fMRI the respiration-induced resonanceoffset not only increases noise interference, it also shifts theactivation band. This leads to a reduction in the contrast-to-noise ratio (CNR) and the potential for varying contrast levelsduring the experiment. A novel real-time method that compen-sates for the respiration-induced resonance offset frequency ispresented. This method utilizes free induction decay (FID)phase information to measure the resonance offset. For com-pensation, one can update the resonant frequency in real timeby changing the transmit RF pulse and receiver phases to trackthe measured offset. The results show decreased signal powerin the respiration frequency band and increased numbers ofactivated voxels with higher Z-scores compared to uncompen-sated experiments. Magn Reson Med 55:1197–1201, 2006.© 2006 Wiley-Liss, Inc.

Key words: fMRI; steady state; real time; respiration; SSFP

Respiration effects in functional MRI (fMRI) have beenwell investigated because they are one of the major sourcesof physiological noise (1–4). A bulk susceptibility changefrom the varying lung volume causes B0 field fluctuationsover time. This results in respiration-induced resonanceoffsets in the brain in addition to the actual brain positionchange. A typical range for the resonance offsets is be-tween 0.15 Hz and 1 Hz in the brain at 1.5 T (2). The effectmonotonically decreases with distance from the lungs andis relatively uniform across an axial slice (4).

Transition-band steady-state free precession (SSFP)fMRI, or blood oxygenation sensitive steady state (BOSS)fMRI, is a recently developed functional imaging tech-nique that utilizes the frequency sensitivity of SSFP todetect the frequency shift of deoxyhemoglobin (5–8). Itpossesses the SNR efficiency of balanced SSFP and hasless imaging distortion compared to conventional BOLD

fMRI (7). However, this technique is highly sensitive toboth spatial and temporal off-resonance shifts because thefunctional contrast occurs only in brain regions that arewell-matched to the center frequency of the scanner,which creates bands of activation across the brain (Fig. 1aand b). Since the activation band is limited to �2 Hz nearthe on-resonance frequency, the respiration-induced reso-nance offset can shift this band substantially (up to 1 Hz at1.5 T), resulting in signal fluctuation in accordance withthe respiration pattern (Fig. 1c). This signal fluctuation canbe large despite the small resonance offset changes be-cause the slope of the magnitude and phase profiles ofsmall-flip-angle balanced SSFP are very steep. Conse-quently, this effect results in a major noise source thatinterferes with the activation time course in transition-band SSFP fMRI. In addition to the varying signal level,respiration-induced resonance offsets cause the activationband to shift, potentially changing the amount of func-tional contrast in a given voxel over time. This is animportant factor that differentiates the respiration effect inSSFP-based fMRI from that in conventional BOLD fMRI,where the effect only induces additive noise. Hence, it isessential to compensate for this effect in order to performa successful and reliable transition-band SSFP fMRI exper-iment.

A number of techniques have been developed to removeartifacts induced by respiration in BOLD fMRI, includingdigital filter methods (9,10), navigator-based methods (11–13), and image-based methods (14,15). However, all ofthese techniques are based on postprocessing methods thatremove the respiration-induced artifacts in reconstructionstages. As a result, they are not suitable for SSFP-basedfMRI, in which the activation band shifts during the ex-periment, creating time-varying functional contrast withina voxel. Therefore, it is crucial to compensate for therespiration-induced artifacts in real time. Here we presenta new real-time respiration-induced B0 field fluctuationcompensation method.

MATERIALS AND METHODS

Respiration-Induced Resonance Offset Measurement inBalanced SSFP

In balanced SSFP the respiration-induced resonance offsetcan be measured from the phase accrual of the free induc-tion decay (FID) as the phase evolves linearly over time(2,16). To measure the resonance offset with this methodrequires a long readout period. Another way to find theresonance offset is to use the phase value measured from asingle point of the FID. As the respiration induces a reso-

1Magnetic Resonance Systems Research Laboratory, Department of Electri-cal Engineering, Stanford University, Stanford, California, USA.2Department of Bioengineering, University of California, Berkeley, California,USA.3FMRIB Centre, Oxford University, Headington, Oxford, United Kingdom.4Department of Radiology, Stanford University, Stanford, California, USA.Grant sponsor: NIH; Grant number: R21EB002969; Grant sponsor: GE Med-ical Systems.*Correspondence to: Jongho Lee, Room 208, Packard Electrical EngineeringBldg., Stanford University, Stanford, CA 94305-9510. E-mail:jonghoyi@mrsrl.stanford.eduReceived 21 September 2005; revised 18 January 2006; accepted 24 January2006.DOI 10.1002/mrm.20879Published online 5 April 2006 in Wiley InterScience (www.interscience.wiley.com).

Magnetic Resonance in Medicine 55:1197–1201 (2006)

© 2006 Wiley-Liss, Inc. 1197

nance offset, the phase profile (and magnitude profile) ofthe image shifts. This results in an FID phase value changein accordance with the resonance offset. In this method, afactor that scales the phase change to the frequency offsetis necessary. Since the FID phase is the phase of the vectorsum over all voxels, the scaling factor depends on the fielddistribution within the object. In the present study wemanually found the scaling factor during the experimentby observing the compensation results in real time.

To measure respiration-induced resonance offset, a con-ventional 2DFT balanced-SSFP sequence was modified tohave a FID period between the slice-selection refocusinggradient and the spatial-encoding gradients (Fig. 2). Thephase value was measured in every TR. Only the firstsample of the FID was used for the measurement. Toreduce the eddy-current-induced phase change, the phaseencoding was ordered in alternating polarity (one phaseencode followed by its opposite polarity phase encode).This ordering differs from conventional ordering schemes(17) because the pulse sequence does not perform thetransmit RF pulse phase cycling. Hence, this scheme ef-fectively cancels the eddy-current-induced phase offset byintroducing the opposite phase evolution in the next TR.

Compensation Method Using the Real-Time System

For real-time updates of the scanner parameters based onthe measurement, we utilized the custom-developedRThawk real-time system (18). This system can acquireand process readout data in one TR and change scan pa-rameters (TR, flip angle, transmit RF pulse phase, etc.) inthe next TR.

The transmit RF pulse and receiver phase changes thateffectively act as a resonance frequency shift (19) wereutilized to track the measured resonance offset. The reso-nance frequency was shifted by the amount of the mea-sured resonance change so that the slice resonance fre-quency tracked the phase of the magnetization. As a result,the respiration-induced resonance offset effect was can-celed and the activation bands remained spatially stable.

Once the compensation begins, the scanner measuresthe phase in the FID every TR and then feeds the value tothe real-time system within the next TR. The real-timesystem averages the phases over eight TRs (62.4 ms in this

FIG. 1. a: Small-flip-angle balanced-SSFP image(TR � 7.8 ms, flip angle � 4°). The signal is verysensitive to off-resonance and is thus not uniformlydistributed. b: The off-resonance frequency distri-bution (field map) of a. The on-resonance area inthe field map has higher signal magnitude in a. c:The average signal magnitude change in ROI over50 s. The ROI is shown by the red line in a. Thesignal magnitude fluctuates with the respirationcycle.

FIG. 2. Modified balanced-SSFP sequence for resonance offsetdetection. An FID period is introduced between the slice-selectiongradient and the imaging gradients to measure the phase of theFIDs. In this experimental setup, 10 samples (320 �s) of the FIDwere acquired in every TR.

1198 Lee et al.

experiment) to reduce the phase noise. It then accumulatesand scales the value to the corresponding frequency offsetvalue. Next, the real-time system feeds back the frequencyoffset value to the scanner on the eighth TR. The scannercalculates the matching transmit RF pulse phase value andadds the new transmit and receiver phase values startingfrom the next TR until a new update is received. Since therespiration cycle (�3 s) is much longer than each update(62.4 ms), the resonance offset is tracked rapidly enough toavoid large magnetic profile shifts.

Experimental Setup

All experiments were performed using a 1.5 T GE EXCITEsystem (40 mT/m and 150 mT/m/ms) with a 3-inch re-ceive-only surface coil and a body transmit coil. Foursubjects, who provided written consent (approved byStanford University), were instructed to avoid any volun-tary motion. Subject motion was further restricted by pads.An axial slice including the visual cortex area was selectedafter 3D localization. The modified 2DFT balanced-SSFPsequence (TR � 7.8 ms, flip angle � 4°, FOV � 24 cm,matrix � 64 � 64, slice thickness � 5 mm, BW � �31.25 kHz) with an FID period of 10 samples was used toacquire images (Fig. 2). Linear shimming was targeted atthe occipital lobe of the brain by a custom-built targetedshim program.

Three scans were performed: 1) a respiration compensa-tion experiment (1 min without respiration compensationfollowed by 1 min with respiration compensation), 2) afunctional experiment without respiration compensation(2 min), and 3) a functional experiment with respirationcompensation (2 min). The order of the functional exper-iments was randomized for each subject. During the ex-periments all of the subjects were instructed to breathenormally. For the functional studies a visual stimulus(10-Hz contrast-reversing annulus grating flashing) waspresented for 2 min (15 s on and 15 s off starting with a 15 sresting period). The subject was instructed to gaze at cross-hairs located at the center of the visual stimulus.

Data Analysis

The ROI for each experiment was the same as the targetedarea for shimming. The power spectrums were calculated

in each voxel and averaged over the ROI. The respirationpeak was chosen as a peak amplitude location within0.1–0.5 Hz of the power spectrum. The respiration bandwas chosen to be �0.1 Hz of the respiration peak. Thepercentage signal power in the respiration band was cal-culated and compared in each subject.

For the functional experiments, analysis was performedusing FEAT FSL (FMRIB, Oxford, UK). The baseline driftwas removed by high-pass filtering. No motion correctionwas performed, since the band shift due to the respiration-induced resonance offset can be mistaken as slice motion.Spatial smoothing was not performed. For activation de-tection, a simple uncorrected voxel P-value threshold of0.01 was used. The time series of the data were inspectedfor subject motions. Voxels with a Z-score higher than 2.33in both experiments were selected for comparison.

RESULTS

Respiration Compensation Results

Figure 3a and b show the magnitude signal change in theROI and the phase measurement from the first sample ofeach FID. The white band shifts due to the respiration-induced resonance offset were obvious in the magnitudetime-series images (Fig. 3a). The cross-correlation coeffi-cient of these two time course signals was, on average,0.74. The power spectrum distributions (Fig. 3c and d)indicate that respiration-induced signal change is a dom-inant noise source in small-flip-angle balanced SSFP. Theobserved respiration peaks were between 0.24 and0.32 Hz. The respiration band signal power calculatedfrom the power spectrum of the magnitude time series wasreduced by 53% on average, and the signal power in thephase measurement was decreased by 97%.

Transition-Band SSFP fMRI Results

In functional experiments the average Z-scores from theselected voxels were increased by 23–59% after compen-sation. On average, a 57% decease in the signal power inthe respiration band was observed (Table 1). The nulldistribution was preserved in all experiments. The mean

FIG. 3. Respiration compensa-tion experiment results. The com-pensation began after 60 s. a:Magnitude signal change aver-aged over the ROI. b: Phase sig-nal change from the first FID sam-ples. c: Power spectrums of themagnitude signal change in theROI. The solid line is the PSD cal-culated from 1 to 60 s, and thedashed line is from 61 to 120 s. d:Power spectrums of the phasesignal change in FID. The solidline is the PSD calculated from 1to 60 s, and the dashed is from 61to 120 s. The small box in d is a100-times-enlarged plot of thedashed line.

Respiration Compensation for BOSS fMRI 1199

time courses and the power spectra of the selected voxelsare given in Fig. 4. The comparison of the two time coursesclearly shows reduced interference from the respiration-induced resonance offset in the compensated experiment.The power spectrum also shows a significantly decreasedsignal peak in the respiration band. An increased numberof activated voxels was observed in each experiment forthe same level of Z-score (Z � 2.33; Table 1). On average,30% more activated voxels were detected with respirationcompensation. In the activation maps (Fig. 5), the threearrows point to statistically significant strips, which matchwell between the two experiments. However, the Z-scoresare much higher in the respiration-compensated experi-ment (Fig. 5a) compared to the uncompensated one (Fig.5b).

DISCUSSION AND CONCLUSIONS

A real-time approach to compensate for respiration-in-duced off-resonance has been proposed and demonstrated.We utilized the phase value of the FID to detect the reso-nance offset and changed the transmit RF pulse phase andreceiver phase in real time to track it. This method suc-cessfully suppressed respiration-induced noise and re-sulted in more reliable functional data. Since other meth-ods for reducing the effects of respiration are based onretrospective techniques that are not adequate to restorethe degraded functional contrast induced from respirationin SSFP-based fMRI, this real-time method is essential.

Another benefit of our method is that it compensates forthe slow drift in the main magnetic field that is oftenobserved in time-series acquisitions. This slow scannerdrift is a critical problem in transition-band SSFP fMRIbecause it changes the location of the activation band overtime (8). As mentioned before, once the activation band isshifted to another region, the original region will no longerhave the same functional contrast. Since our approachstabilizes any signal change that is slower than the maxi-mum tracking speed of the feedback loop, the slow scannerdrift is also compensated.

In this experiment we manually selected the factor thatscales the phase value to the frequency offset by observingthe compensation results in real time. This method issimple and quick because the feedback loop has a toler-ance to a range of scaling values. The FID phase signalpower reduction ratio also confirms the success of thismethod. The ratio was on average 97%, which implies thatthe respiration band signal power decreased dramaticallywith the use of the manually selected scaling factor. An-other way to find a more exact scaling value is to calculatethe ratio of the frequency estimation from the FID periodsand phase values. However, this method requires enoughFID periods to obtain a reliable frequency estimation. Thescaling factor only has to be calculated once for the samefield distribution.

Respiration-induced off-resonance might disturb themagnetic steady state because it changes the resonancefrequency over time. However, the respiration cycle is

Table 1Transition-Band SSFP fMRI Experimental Results

Subject 1 Subject 2 Subject 3 Subject 4

Average Z-score without compensation 4.60 3.70 3.99 6.08Average Z-score with compensation 5.66 5.53 6.35 7.57Respiration signal power reduction ratioa 0.34 0.65 0.62 0.66Number of activated voxels without compensationb 55 29 59 67Number of activated voxels with compensationb 56 47 89 81

a1-(respiration band signal power with compensation)/(respiration band signal power without compensation).bThe activated voxels were voxels in the ROI that had a Z-score � 2.33.

FIG. 4. Transition-band SSFPfMRI experiment results. Voxelswith a Z-score higher than 2.33 inboth experiments were selectedand averaged to generate theplots. a and c: Magnitude signalchange of the uncompensatedexperiment and its power spec-trum. b and d: Respiration-com-pensated experiment results. Thebars at the bottom of a and brepresent the stimulated period.

1200 Lee et al.

much slower than the limit to disturb the steady statesuggested by Foxall (20). Therefore, respiration has a verysmall effect on disturbing the magnetic steady state. More-over, our approach tracks the induced off-resonance andmakes the magnetic profile stationary. Hence, our methodeliminates the steady-state disturbance from the respira-tion-induced resonance offset.

Since the resonance offset shift from respiration is rela-tively uniform across axial slices, the proposed methodworks best for axial slices. In different slice orientations,the results will experience partial compensation. For in-stance, a coronal slice will have overcompensation in thesuperior part of the image and undercompensation in theinferior part of the image as the respiration-induced reso-nance offset decreases along the Z-axis toward the top ofthe head.

REFERENCES

1. Weisskoff RM, Baker JR, Belliveau JW, Davis TL, Kwong KK, CohenMS, Rosen BR. Power spectrum analysis of functionally-weighted MRdata: what’s in the noise? In: Proceedings of the 1st Annual Meeting ofSMRM, New York, NY, USA, 1993. p 7.

2. Noll DC, Schneider W. Respiration artifacts in functional brainimaging: sources of signal variation and compensation strategies. In:Proceedings of the 2nd Annual Meeting of SMRM, San Francisco, CA,USA, 1994. p 647.

3. Raj D, Paley DP, Anderson AW, Kennan RP, Gore JC. A model forsusceptibility artefacts from respiration in functional echo-planar mag-netic resonance imaging. Phys Med Biol 2000;45:3809–3820.

4. Van de Moortele PF, Pfeuffer J, Glover GH, Ugurbil K, Hu X. Respira-tion-induced B0 fluctuations and their spatial distribution in the humanbrain at 7 Tesla. Magn Reson Med 2002;47:888–895.

5. Cho ZH, Ro YM, Chung SC, Chung JY. A direct susceptibility measure-ment in fMRI using SSFP interferometry (SSFPI) technique. In: Pro-ceedings of the 3rd Annual Meeting of SMR, Nice, France, 1995. p 806.

6. Scheffler K, Seifritz E, Bilecen D, Venkatesan R, Hennig J, Deimling M,Haacke EM. Detection of BOLD changes by means of a frequencysen-sitive trueFISP technique: preliminary results. NMR Biomed 2001;14:490–496.

7. Miller KL, Hargreaves BA, Lee J, Ress D, deCharms RC, Pauly JM.Functional brain imaging using a blood oxygenation sensitive steadystate. Magn Reson Med 2003;50:675–683.

8. Miller KL, Smith SM, Jezzard P, Pauly JM. High-resolution FMRI at1.5T Using balanced SSFP. Magn Reson Med 2006;55:161–170.

9. Biswal B, DeYoe AE, Hyde JS. Reduction of physiological fluctuationsin fMRI using digital filters. Magn Reson Med 1996;35:107–113.

10. Buonocore MH, Maddock RJ. Noise suppression digital filter for func-tional magnetic resonance imaging. Magn Reson Med 1997;38:456–469.

11. Hu X, Kim S-G. Reduction of physiological noise in functional MRIusing navigator echo. Magn Reson Med 1994;31:495–503.

12. Hu X, Le TH, Parrish T, Erhard P. Retrospective estimation and com-pensation of physiological fluctuation in functional MRI. Magn ResonMed 1995;34:210–221.

13. Pfeuffer J, Van de Moortele P-F, Ugurbil K, Hu X, Glover G. Correctionof physiologically-induced global off-resonance in dynamic echo-pla-nar and spiral functional imaging. Magn Reson Med 2002;47:344–353.

14. Glover GH, Li TQ, Ress D. Image-based method for retrospective cor-rection of physiological motion effect in fMRI: RETROICOR. MagnReson Med 2000;44:162–167.

15. Chuang K-H, Chen J-H. IMPACT: image-based physiological artifactsestimation and correction technique for functional MRI. Magn ResonMed 2001;46:344–353.

16. Freeman R, Hill HDW. Phase and intensity anomalies in Fourier trans-form NMR. J Magn Reson 1971;4:366–383.

17. Bieri O, Markl M, Scheffler K. Analysis and compensation of eddycurrents in balanced SSFP. Magn Reson Med 2005;54:129–137.

18. Santos JM, Wright GA, Pauly JM. Flexible real-time magnetic resonanceimaging framework. In: Proceedings of the 26th Annual Meeting ofIEEE EMBS, San Francisco, CA, USA, 2004. p 1048.

19. Hinshaw WS. Image formation by nuclear magnetic resonance: thesensitive-point method. J Appl Phys 1976;47:3709–3721.

20. Foxall DL. Frequency-modulated steady-state free precession imaging.Magn Reson Med 2002;48:502–508.

FIG. 5. Transition-band SSFP fMRI activationmaps. a: Without respiration compensation (Z-score � 3.55). b: With respiration compensation(Z-score � 3.55). The Z-score scale is color codedfrom red (Z � 3.55) to yellow (Z � 8). The highZ-score areas (arrowed) in b match well with theactivated areas in a. However, b shows muchhigher Z-scores than a in the same areas.

Respiration Compensation for BOSS fMRI 1201