Removal of eye blink artifacts in wireless EEG sensor networks using reduced-bandwidth canonical correlation analysis

Abstract

Removal of eye blink artifacts in wireless EEG sensor networks using reduced-bandwidth canonical correlation analysisJournal of Neural Engineering, vol. 13, no. 6, pp. 066008, 2016. Archived versionAuthor manuscript: the content is identical to the content of the published paper, but without the final typesetting by the publisher Abstract-Objective: Chronic, 24/7 EEG monitoring requires the use of highly miniaturized EEG modules, which only measure a few EEG channels over a small area. For improved spatial coverage, a wireless EEG sensor network (WESN) can be deployed, consisting of multiple EEG modules, which interact through short-distance wireless communication. In this paper, we aim to remove eye blink artifacts in each EEG channel of a WESN by optimally exploiting the correlation between EEG signals from different modules, under stringent communication bandwidth constraints. Approach: We apply a distributed canonical correlation analysis (CCA-)based algorithm, in which each module only transmits an optimal linear combination of its local EEG channels to the other modules. The method is validated on both synthetic and real EEG data sets, with emulated wireless transmissions. Main Results: While strongly reducing the amount of data that is shared between nodes, we demonstrate that the algorithm achieves the same eye blink artifact removal performance as the equivalent centralized CCA algorithm, which is at least as good as other state-of-the-art multi-channel algorithms that require a transmission of all channels. Significance: Due to their potential for extreme miniaturization, WESNs are viewed as an enabling technology for chronic EEG monitoring. However, multi-channel analysis is hampered in WESNs due to the high energy cost for wireless communication. This paper shows that multi-channel eye blink artifact removal is possible with a significantly reduced wireless communication between EEG modules.

A generic EEG artifact removal algorithm based on the multi-channel Wiener filter

“…A factor γ is used to scale the artifacts d to generate EEG data with different artifact SNR values. We define the artifact SNR as the ratio of artifact power and the clean EEG power [21], i.e. SNR = 10 log 10 E{(γd…”

“…In fact, the estimated artifact signal in the artifact-free segments of channel i, given byd i , should be as close to zero as possible. To assess this, we use the Signal-to-Error Ratio (SER) [19], [21] in a single channel i, computed as…”

“…the residue d i −d i should be as small as possible. This leads to the Artifact-to-Residue Ratio (ARR) [21] in a single channel i, computed as…”

“…The MWF was applied to the EEG recordings with eye blink artifacts, and to the EEG recordings containing both eye blink and muscle artifacts. The same recordings are processed by three component-based methods: infomax ICA [8], [30], which is implemented in EEGLAB, the FastICA algortihm [31] from the FastICA MATLAB toolbox, and CCA [9], [21]. Artifact removal is performed semi-automatically by selecting artifact components to reject after visualization.…”

Optimal Versus Approximate Channel Selection Methods for EEG Decoding With Application to Topology-Constrained Neuro-Sensor Networks

“…Considerable research is ongoing to make wearable miniature-EEG (mini-EEG) devices which allow to record EEG 24/7 in daily-life activities [3]- [8]. Although these mini-EEG devices only cover small skin areas due to their far-driven miniaturization, the concept of neuro-sensor networks enables the simultaneous use of multiple such mini-EEG devices connected wirelessly thereby increasing the spatial resolution [10], [19]. Such a collection of wirelessly interconnected mini-EEG devices is also known as wireless EEG sensor networks (WESNs).…”