3937

Time variant cardiac noise correction in resting state fMRI without physiologic noise measures
Wanyong Shin1 and Mark J Lowe1

1Radiology, Cleveland Clinic, Cleveland, OH, United States

Synopsis

The cardiac and respiratory response functions derived from RETROICOR have substantial variation. We found the cardiac response function is phase, or time shifted across the brain while the respiratory function has a fixed phase, but changes polarity across the brain. Based on this finding, we propose a cardiac and respiratory noise correction method in rsfMRI data without external cardiac and respiratory fluctuation measures, and compare its performance to the RETROICOR model.

Introduction

We previously found that the cardiac response function is time-delayed or phase shifted while the respiratory response function has fixed polarity shifts(1), as shown in Fig1. Voxelwise RETROICOR(2) considers time or phase shift of the cardiac and respiratory fluctuations. However any representative cardiac and respiratory signal regressor method including those from PESTICA (3), CORSICA (4), or FIX (5) are limited to detect the time varying cardiac noise. Figure 2 demonstrates the cardiac hemodynamic response function (HRF) calculated from RETROICOR, then cardiac HRF regressor is regressed with and without time shift. As shown, the optimal cardiac HRF regressor varies with a time shift. We propose a cardiac phase shift corrected PESTICA, called as cPESTICA, and validate its performance compared to RETROICOR.

Method

Twenty eight healthy controls were scanned at 3T using single band EPI with pulse plethysmograph and respiratory belt recording (TR=2.8s, 128x128 matrix, 31 slices, 132 repetitions). After the first 4 volumes were removed, the second harmonic RETROICOR was applied using modified RetroTS.m (https://afni.nimh.nih.gov/sscc/staff/glend/matlab_compiler). RETROICOR includes the physiologic noise removal process to minimize the error of physiologic regressor model fit. In this study, “RETROICOR model” indicates Fourier series of regressor fitting based on the defined phase. We calculated the voxelwise cardiac and respiratory HRFs from the top 5% of F value voxels in RETROICOR model for the development and validation of the proposed cPESTICA.

PESTICA generates the physiologic noise estimator using slicewise ICA (3). The first, we define the highest peaks from the estimated cardiac noise estimators from PESTICA, as shown in Fig3, and the linear phase in the time domain from 0 to 2π between the peaks. Note that this phase, defined from PESTICA is not phase from the actual cardiac pulsation measures. The second order Fourier Series were modeled over the defined phase in the same ways that RETROICOR model doesTherefore, cPESTICA has 5 degree of freedom (DOF) while the original PESTICA and RETROICOR have 2 and 8 DOFs.

While HRF is NOT used for cPESTICA or RETROICOR correction, HRFs over phase between cPESTICA and RETROICOR are compared for the validation purpose. F value and t-score maps from cPESTICA and RETROICOR are presented.

Result

The correlation coefficient (CC) of the cardiac and respiratory HRFs between cPESTICA and RETROICOR are calculated with 0.93±0.08 and 0.90±0.11 across 28 subjects, respectively. CC between the timeseries of cardiac and respiratory HRF regressors are 0.54±0.19 and 0.57±0.17, respectively. Twelve representative HRFs and HRF regressors are presented in Fig4. Fig 5 shows the example of the cardiac and respiratory model fits when using RETROICOR and cPESTICA. The colored areas (p < 10-4) are commonly observed in both RETROICOR and cPESTICA.

Discussion

We found that cPESTICA estimates cardiac noise regressors that are equivalent to those produced from the RETROICOR model. Our observed high correlation of the cardiac HRF between cPESTICA and RETROICOR model indicates that cPESTICA can provide reliable physiologic estimators when external physiologic noise measures are not available during fMRI scanning or when the physiologic data is corrupted due to bad contacts or acquisition issues.

This study shows that physiological noise correction with single regressor is only valid for respiratory noise but limited for the time delayed cardiac noise. cPESTICA considers the shift of the cardiac HRF and removes the cardiac noise effectively.

Acknowledgements

Authors gratefully acknowledge technical support by Siemens Medical Solutions.

References

1. Shin W, Lowe MJ, editors. A Comprehensive Investigation of Physiologic Noise Modeling in Resting State FMRI; Phase Shifted Cardiac Response Function in EPI. Proceeding of the 26th Meeting of the Society for Magnetic Resonance in Medicine; 2018; Paris, France.

2.Glover GH, Li TQ, Ress D. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magn Reson Med. 2000;44(1):162-7.

3. Beall EB, Lowe MJ. Isolating physiologic noise sources with independently determined spatial measures. Neuroimage. 2007;37(4):1286-300.

4. Perlbarg V, Bellec P, Anton JL, Pelegrini-Issac M, Doyon J, Benali H. CORSICA: correction of structured noise in fMRI by automatic identification of ICA components. Magnetic resonance imaging. 2007;25(1):35-46. PubMed PMID: 17222713.

5. Griffanti L, Salimi-Khorshidi G, Beckmann CF, Auerbach EJ, Douaud G, Sexton CE, Zsoldos E, Ebmeier KP, Filippini N, Mackay CE, Moeller S, Xu J, Yacoub E, Baselli G, Ugurbil K, Miller KL, Smith SM. ICA-based artefact removal and accelerated fMRI acquisition for improved resting state network imaging. Neuroimage. 2014;95:232-47.

Figures

Figure1. The example of the first harmonic coefficients of cosine and sine terns in RETROICOR model from the top 5% F value voxels. The ratio of cosine to sine coefficient terms indicates the phase shift in the first harmonic oscillation. The cardiac response function has wide range of phases while the respiratory response function has the polarity phase. The error bar of the cardiac response function is decreased with the polarity correction and even further with the phase shift (time delayed) correction while the decrease in the error bar of the respiratory response function is not noticeable between the polarity and phase corrections.

Figure 2. t-score maps when the cardiac HRF is regressed (left) and when the cardiac HRF is dithered with +/-2 times of slice timing shift (~0.2s) (middle), scaled between from 1/-1 (red/blue). Note that the segments of middle cerebral arteries regions are additionally found when the cardiac HRF is shifted. The right map presents the time delay of cardiac HRF in the areas where cardiac HRF is significantly correlated to EPI (p < 10-3), scaled between +/- 0.2s (red, arrived fast/blue, delayed). The delay map shows that regions near the basilar arteries and their branches have a faster arrival time than the downstream tissue areas.

Figure 3. The example of the cardiac response function (red) from RETROICOR and cardiac estimator (blue) from PESTICA. Note that RETROICOR model (red) is bounded between -1 to 1 within a period while PESTICA fluctuates relatively freely since it is chosen in EPI time series using ICA. The blue dots in PESTICA estimator indicates the defined high peaks between which the second order Fourier Series signal is modeled. Then cPESTICA generates 4 regressors for the cardiac noise as in the same way as RETROICOR does

Figure 4. The examples of the cardiac (1st column) and the respiratory (2nd column) HRF over phases. The third and the forth columns plots the cardiac and respiratory HRF regressors during 10s and 30s after 1 mins of scans. Red and blue lines indicate HRF from RETROICOR and cPESTICA, respectively. Note that HRF is not used for either RETROICOR or cPESTICA. Twelve out of 28 subjects data are shown here.

Figure 5. The example of physiology noise model fit using RETROICOR and cardiac phase corrected PESTICA (cPESTICA). The colored voxels indicate each noise model with the significance, p < 10-4. F value and t-score maps were scaled from -50 (blue) to 50 (red), and from -10 (blue) to 10 (red), respectively.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
3937