In simultaneous EEG-fMRI, the period of cardioballistic artifact (BCG) in EEG is required for the artifact removal. Recording the electrocardiogram (ECG) waveform during fMRI is difficult, often causing inaccurate period detection. Since the BCG artifact waveform in EEG-fMRI is relatively invariable compared to the ECG waveform, we propose a multiple-scale peak-detection algorithm to determine directly the BCG period from EEG-fMRI data. The algorithm achieves a high detection accuracy of the BCG artifact occurrence on a large EEG-fMRI dataset without using the ECG waveforms, virtually eliminating the need for ECG for BCG artifact removal.
The
simultaneous EEG-fMRI was conducted on a GE MR750 3T MRI with an 8-channel head
coil and 32ch MR-compatible EEG system (Brain Products GmbH). MRI artifacts were
removed, then ICA was applied to separate the BCG artifact. The cardioballistic
artifact components (CB-ICs) were automatically identified based on the
algorithm described in Ref.[8], with the algorithm extended to select all
CB-ICs instead of only the motion-related CB-ICs. The CB-IC with the highest
peak contrast (C: mean amplitude of the
peaks that lie between 1-4
standard deviations) was selected for the period detection (Fig.2).
Multiple-scale peak detection was originally proposed for peak-to-peak period determination in ECG.9 Here the algorithm is refined for the BCG period detection. The algorithm included two steps of band-pass filtering. In Step 1, the band-pass filtering of CB-IC (bp1) aims to estimate the period (Fig.2). In this step, a smoothed spectrum of the CB-IC was calculated (Fig.3), from which the filtering frequency of bp1 was chosen around the identified fundamental frequency of the heart beats. In Step 2, the CB-IC was smoothed (bp2) by band-pass filtering between 0.1Hz and the MR slice acquisition frequency (17Hz). Then a peak was selected from bp2 in each bp1 period based on the peak amplitude, peak rise, slope, and the past periods variations. Finally, the detected peak locations were adjusted if they exceed certain threshold determined by the reference frequencies found in Step 1 (Fig.3).
The proposed algorithm was applied to 281 resting scans from 48 subjects. Each scan lasted 526s. The first 6s of the scan and the time segments with unrecognizable ECG or CB-IC waveform caused by subject motions were removed from consideration, giving a total recording time of 39.98 hours. The automatically detected peaks of the CB-IC were compared to the correct peak timing, which was acquired in reference to both the CB-IC and ECG waveforms with a resolution of 10ms.
1. Allen PJ, Polizzi G, Krakow K, et al. Identification of EEG events in the MR scanner: the problem of pulse artifact and a method for its subtraction. NeuroImage. 1998;8:229-239.
2. Niazy RK, Beckmann CF, Iannetti GD, et al. Removal of fMRI environment artifacts from EEG data using optimal basis sets. NeuroImage. 2005;28:720-737.
3. Bell AJ and Sejnowski TJ. An information-maximization approach to blind separation and blind deconvolution. Neural Comput. 1995;7:1129-1159.
4. Debener S, Strobel A, Sorger B, et al. Improved quality of auditory event-related potentials recorded simultaneously with 3-T fMRI: removal of the ballistocardiogram artefact. NeuroImage. 2007;34:587-597.
5. Christov II. Real time electrocardiogram QRS detection using combined adaptive threshold. BioMed. Eng. Online. 2004;3:28.
6. Kim KH, Yoon HW and Park HW. Improved ballistocardiac artifact removal from the electroencephalogram recorded in fMRI. J. NeuroScience Methods. 2004;135:193-203.
7. Delorme A. and Makeig S. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics (pdf, 0.7MB). J. Neuroscience Methods. 2004;134:9-21.
8. Wong CK, Zotev V, Misaki M, et al. An automatic EEG-assisted retrospective motion correction for fMRI (aE-REMCOR). NeuroImage. 2016;129:133-147.
9. Wong CK, Zotev V, Yuan H, et al. Cardiac cycle detection in EEG with residual MRI gradient artifact by multiple-scale peak detetion. Proc. Org. Hum. Brain Map. 2014;20:2262.