Real-time ICA-based artifact correction of EEG data recorded during functional MRI
Ahmad Mayeli1,2, Vadim Zotev1, Hazem Refai2, and Jerzy Bodurka1,3

1Laureate Institute for Brain Research, Tulsa, OK, United States, 2Electrical and Computer Engineering, University of Oklahoma, Tulsa, OK, United States, 3College of Engineering, University of Oklahoma, Tulsa, OK, United States

Synopsis

Recording EEG signals during fMRI acquisition result EEG signals contamination from imaging and BCG artifacts in addition to other types of artifacts. In this abstract we introduced a novel method for detecting and reducing artifacts using second order blind separation in real time. The algorithm was tested on EEG signals from 12 subjects and it has proven successful for reducing various types of artifacts. The proposed algorithm can be applied in real time after the average artifact subtraction of imaging and BCG artifacts for various applications that require a real-time EEG & fMRI system.

Target audience:

Researchers and clinicians interested in employing simultaneous EEG-fMRI techniques.

Purpose:

Simultaneously EEG and fMRI measurement requires correction of various artifacts in EEG data. Besides Ballistocardiogram (BCG) and imaging (MR) artifacts, other types of artifacts, such as muscle and ocular artifacts can be present in EEG data regardless whether the EEG is recorded inside or outside the MRI scanner. To date, Average Artifact Subtraction (AAS) [1-2] is the only real-time method used to partially remove BCG and imaging artifacts; and no alternative real-time methods for removing other artifacts are available. However, independent component analysis (ICA) is widely used to suppress residual BCG and imaging, and other artifacts from EEG data offline. We developed a novel real-time ICA-based method for automatic detection and removal of various artifacts from EEG data simultaneously recorded with fMRI. The algorithm was validated in real time EEG & fMRI experiments and offline with prerecorded EEG data.

Methods:

EEG and fMRI data were acquired with a 32-channel MR-compatible EEG (Brain Products GmbH) and a GE Discovery MR750 3T MRI scanner. RecView software was used to suppress imaging and BCG artifacts from EEG data in real time, and then to down-sample data to 250 S/s. To obtain reliable and stable results from ICA decomposition, data submitted to the algorithm should be at least a multiple k of n2 [3] where n is the number of channels. In this study, to increase algorithm speed, 22 channels were selected as opposed to performing ICA on all channels. Given 22-channel data and k=20, the number of data points submitted to ICA must be more than 9,680, rounded up to 10,000 data points. We used SOBI [4], due to its faster convergence (compared to other ICA algorithms) for ICA decomposition. The algorithm was implemented in the following steps [5]: After the first 10,000 data samples are received, the first pass for artifact detection and reduction is executed on the last 1,000 data points. ICA is applied to the 10,000 data points each time, and the artifact-free EEG signal is updated every 4 sec. As subsequent 1,000 data samples are received, automated ICA proceeds and the artifact-free EEG signal is updated. Various features (i.e., energy, kurtosis, topographical map and power spectral density (PSD)) are extracted from Independent Components (ICs) for classification as brain activity and artifacts. Fig. 1 shows typical ICs distinguished as different types of artifacts along with their features. After ICA is performed on 22-channel data, maximum of overall 11 ICs for various types of artifacts are accounted for. Corrected EEG data can be obtained by removing components related to these artifacts, and then using inverse ICA to reconstruct the EEG signal.

Results:

The real-time ICA (rtICA) algorithm was applied on prerecorded EEG data from three healthy female subjects (mean age: 26±6 years) and three male subjects (mean age: 29±9 years) with combat related post-traumatic stress disorder (PTSD) diagnosis. EEG data were acquired under similar conditions during fMRI neurofeedback task runs (3640 sec) [6]. Furthermore, the method was tested on six healthy subjects (mean age: 36±14 years, 3 female) in real-time experiments. Four resting state runs (8 min 40 sec each) were collected on the subjects. The participants were instructed to relax and rest with eyes closed for two runs and eyes open and fixed on a fixation cross for the other two runs. Fig. 2 shows two examples of PSD before and after applying rtICA artifact correction. After artifact corrections, large power reductions were observed in EEG Delta and Theta bands. Smaller power reductions were observed in the Alpha and Beta bands. To quantify BCG artifact changes and correction improvements, we computed the normalized power spectrum ratio (INPS) [7] as the ratio of the sums of PSD values for cardiac harmonics before and after the BCG reduction and the results are shown in Fig. 3. To evaluate algorithm performance for detecting eye blinks, detection algorithm output was compared with a manual inspection of EEG data. The average percentage of correctly detected eye blinks for 12 subjects was found to be 94.87%±5%. Fig. 4 illustrates an example of RecView-corrected EEG signal (red lines) and further improved EEG data after removing artifacts detected by our ICA-based automatic method (black lines).

Discussion and Conclusion:

We developed a novel real-time ICA-based method for suppression EEG artifacts during simultaneous EEG & fMRI. This method can detect and remove BCG, imaging, muscle, eye blinks and motion artifacts in EEG data recorded simultaneously with fMRI. The method can improve real-time estimation of EEG signals and in particular multimodal EEG and fMRI neurofeedback applications.

Acknowledgements

This research was supported by W81XWH-12-1-0697 grant from the US Department of Defense.

References

1. Allen PJ, Josephs O, Turner R. A method for removing imaging artifact from continuous EEG recorded during functional MRI. Neuroimage 2000;12(2):230-239.

2. Allen PJ, Polizzi G, Krakow K, Fish DR, Lemieux L. Identification of EEG events in the MR scanner: the problem of pulse artifact and a method for its subtraction. Neuroimage 1998;8(3):229-239.

3. Onton J, Westerfield M, Townsend J, Makeig S. Imaging human EEG dynamics using independent component analysis. Neuroscience & Biobehavioral Reviews 2006;30(6):808-822.

4. Belouchrani A, Abed-Meraim K, Cardoso JF, Moulines E. A blind source separation technique using second-order statistics. IEEE Trans. Signal Processing1997;45(2):434-444.

5. Mayeli A, Zotev V, Refai H, Bodurka J. An automatic ICA-based method for removing artifacts from EEG data acquired during fMRI in real time. Proc. of 41st Annual Bioengineering Conference (NEBEC) 2015; pp.1-2.

6. Zotev V, Phillips R, Yuan H, Misaki M, Bodurka J. Self-regulation of human brain activity using simultaneous real-time fMRI and EEG neurofeedback. NeuroImage 2014;85:985-995.

7. Srivastava G, Crottaz-Herbette S, Lau KM, Glover GH, Menon V. ICA-based procedures for removing ballistocardiogram artifacts from EEG data acquired in the MRI scanner. Neuroimage 2005;24(1):50-60.

Figures

Figure 1: Automatic (A) eye artifact, (B) residual BCG artifact, (C) residual MR and Muscle artifacts and (D) motion artifact detection and results of removing them from EEG data.

Figure 2: PSD reduction using rtICA (A) applied on prerecorded EEG data from a subject with PTSD diagnosis during neurofeedback run (B) tested on real-time data from healthy subject during resting state.

Figure 3: INPS Reduction for channel F3 using rtICA compared to rtAAS (A) tested on prerecorded data from 6 subjects (B) applied on EEG data from 6 healthy subjects in real time.

Figure 4: Representative 8-second example traces of RecView-corrected EEG data recorded during continuous fMRI (red lines) and results of using our proposed ICA-based automatic artifact removal method (black lines).



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3773