Removing muscle and eye artifacts using blind source separation techniques in ictal EEG source imaging

Hans Hallez, Maarten De Vos, Bart Vanrumste, Peter Van Hese, Sara Assecondi, Koen Van Laere, Patrick Dupont, Wim Van Paesschen, Sabine Van Huffel, Ignace LemahieuClinical 2009 · DOI

TL;DR

This paper tackles the challenge of accurately pinpointing the origin of epileptic seizures (the epileptogenic focus) using EEG data, which is often corrupted by muscle and eye movement artifacts. By applying blind source separation techniques (BSS-CCA and SCICA) as a preprocessing step before source localization, the researchers demonstrate improved accuracy in identifying the seizure onset zone, making presurgical evaluations more reliable.

What Problem Does This Paper Address?

Epilepsy, a neurological disorder characterized by abnormal electrical activity in the brain, can be treated surgically by removing the epileptogenic focus. However, accurately locating this focus is crucial for successful surgery. Electroencephalography (EEG) is a key tool for this, offering high temporal resolution to capture the rapid electrical activity during a seizure (ictal period). The problem is that ictal EEG is frequently contaminated by artifacts from muscle activity and eye movements. These artifacts distort the electrical signals, leading to inaccurate source localization and potentially misidentifying the seizure origin. Existing methods for artifact removal often have limitations, such as requiring additional sensors (like EOG for eye artifacts) or filtering out valuable seizure information. This paper aims to address this by investigating the effectiveness of advanced blind source separation techniques for artifact removal before performing EEG source localization.

Methods

The core of this study lies in a two-step process:

  1. Artifact Removal using Blind Source Separation (BSS):
    • BSS-CCA (Canonical Correlation Analysis): This technique is used to remove muscle artifacts. It leverages the fact that muscle artifacts have lower autocorrelation compared to brain activity. The method is semi-automatic, allowing an expert to visually inspect and adjust the number of components removed.
  2. SCICA (Spatially Constrained Independent Component Analysis): This method is employed for removing eye blink and eye movement artifacts. It's a fully automatic process that utilizes prior knowledge about the spatial distribution of these artifacts.
  3. EEG Source Localization using RAP-MUSIC:
    • After artifact removal, the cleaned EEG data is processed using the RAP-MUSIC (Recursively Applied and Projected Multiple Signal Classification) algorithm. This is a multi-dipole source estimation technique that separates the EEG signal into signal and noise subspaces.
  4. The estimated dipole locations are then compared to the ictal Single-Photon Emission Computed Tomography (SPECT) activity, which is coregistered to MRI (SISCOM). The SPECT provides a 3D map of brain activity during the seizure, serving as an independent reference for the seizure onset zone. The accuracy of the dipole localization is assessed by measuring the Euclidean distance between the estimated dipoles and the SPECT activation area.

Datasets

The study utilized EEG data from 8 epilepsy patients who underwent presurgical evaluation. The data included ictal EEG segments, MRI scans, and ictal SPECT (iSPECT) images. The iSPECT images were coregistered with MRI to create a realistic head model for source localization. No new benchmarks were introduced; the study focused on evaluating existing artifact removal techniques in conjunction with a standard source localization method.

Key Insights and Ideas

  • Artifacts are a major hurdle for accurate EEG source localization: The study strongly emphasizes that muscle and eye artifacts significantly distort ictal EEG, leading to unreliable source estimations.
  • BSS techniques offer a promising solution: Blind source separation methods like BSS-CCA and SCICA can effectively isolate and remove these artifacts without discarding crucial seizure information, unlike traditional filtering methods.
  • Preprocessing is key for reliable source localization: Applying artifact removal before source localization significantly improves the accuracy and reliability of identifying the epileptogenic focus.
  • Qualitative improvement is evident: Visual inspection of dipole estimations before and after artifact removal clearly shows a tighter clustering of dipoles around the SPECT activation zone after filtering.
  • Quantitative metrics support qualitative findings: The median distance between estimated dipoles and the SPECT activation decreased in a majority of patients after artifact removal, indicating improved localization.

Key Results

  • Improvement in 5 out of 8 patients: The study found that in 5 out of 8 patients, the artifact removal techniques led to more accurate dipole estimations, characterized by tighter clusters or new clusters appearing near or overlapping with the ictal SPECT activity.
  • Reduced dipole-SPECT distance: For these 5 patients, the median distance between the estimated dipoles and the ictal SPECT activation was reduced after filtering.
  • Qualitative visual evidence: Figures in the paper visually demonstrate how dipole estimations shift from widespread, artifact-driven locations (e.g., near the eyes or muscles) to more focal areas consistent with SPECT findings after artifact removal.
  • Semi-automatic vs. automatic artifact removal: BSS-CCA for muscle artifacts required some manual intervention, while SCICA for eye artifacts was fully automatic, highlighting potential for clinical integration.

Limitations and Future Work

  • Head Model Limitations: The study used a standard head model derived from a generic MRI, not patient-specific MRIs. This lack of individualized head models introduces potential errors in source localization accuracy. Future work should focus on creating patient-specific head models using individual MRIs and precise electrode localization.
  • Semi-Automatic Nature of Muscle Artifact Removal: While BSS-CCA is semi-automatic, the manual selection of components in 10-second windows can lead to abrupt changes in signal-to-noise ratio and dipole estimates. Future research should aim for more robust and automatic component selection.
  • Potential for Over-filtering: In cases of severe muscle contamination, the artifact removal process might inadvertently remove components related to the ictal activity itself, leading to a loss of relevant information. Careful application and validation are advised.
  • Pilot Study: The authors acknowledge this is a pilot study and suggest that further research is needed to fully integrate these techniques into routine clinical practice. This includes developing user-friendly interfaces and automating more aspects of the process.
  • Focal vs. Non-Focal Activity: The paper notes that for non-focal epileptic activity, dipole models might not be suitable, which could explain why some patients did not show improvement.

Summary updated

BibTeX
@inproceedings{hallez2009removing,
  author = {Hans Hallez and Maarten De Vos and Bart Vanrumste and Peter Van Hese and Sara Assecondi and Koen Van Laere and Patrick Dupont and Wim Van Paesschen and Sabine Van Huffel and Ignace Lemahieu},
  title = {{Removing muscle and eye artifacts using blind source separation techniques in ictal EEG source imaging}},
  booktitle = {Clinical},
  year = {2009},
  doi = {10.1016/j.clinph.2009.05.010},
}
Figure