Limits...
Independent Component Analysis of Gait-Related Movement Artifact Recorded using EEG Electrodes during Treadmill Walking.

Snyder KL, Kline JE, Huang HJ, Ferris DP - Front Hum Neurosci (2015)

Bottom Line: The remaining 1% of sources had locations within the brain volume and low residual variances, but had topographical maps, power spectra, time courses, and event related spectral perturbations typical of non-neural sources.Caution should be exercised when interpreting ICA for data that includes semi-periodic artifacts including artifact arising from human walking.Separating true brain signals from motion artifact could clear the way for EEG brain computer interfaces for assistance during mobile activities, such as walking.

View Article: PubMed Central - PubMed

Affiliation: School of Kinesiology, University of Michigan Ann Arbor, MI, USA ; Department of Mathematics and Statistics, University of Minnesota Duluth Duluth, MN, USA.

ABSTRACT
There has been a recent surge in the use of electroencephalography (EEG) as a tool for mobile brain imaging due to its portability and fine time resolution. When EEG is combined with independent component analysis (ICA) and source localization techniques, it can model electrocortical activity as arising from temporally independent signals located in spatially distinct cortical areas. However, for mobile tasks, it is not clear how movement artifacts influence ICA and source localization. We devised a novel method to collect pure movement artifact data (devoid of any electrophysiological signals) with a 256-channel EEG system. We first blocked true electrocortical activity using a silicone swim cap. Over the silicone layer, we placed a simulated scalp with electrical properties similar to real human scalp. We collected EEG movement artifact signals from ten healthy, young subjects wearing this setup as they walked on a treadmill at speeds from 0.4-1.6 m/s. We performed ICA and dipole fitting on the EEG movement artifact data to quantify how accurately these methods would identify the artifact signals as non-neural. ICA and dipole fitting accurately localized 99% of the independent components in non-neural locations or lacked dipolar characteristics. The remaining 1% of sources had locations within the brain volume and low residual variances, but had topographical maps, power spectra, time courses, and event related spectral perturbations typical of non-neural sources. Caution should be exercised when interpreting ICA for data that includes semi-periodic artifacts including artifact arising from human walking. Alternative methods are needed for the identification and separation of movement artifact in mobile EEG signals, especially methods that can be performed in real time. Separating true brain signals from motion artifact could clear the way for EEG brain computer interfaces for assistance during mobile activities, such as walking.

No MeSH data available.


Related in: MedlinePlus

Cortical component characteristics. Cortical locations, topographic maps, residual variance (RV) values for the full set and for each half set for the split-half comparisons (in parentheses), and average distance between analogous components for the full-set and each half set in Talairach coordinates are shown for the nine components with RV < 15% and neural locations for the full set. The word “None” appears when there were no components with neural locations and RVs < 15% for a given set. Components that were reliably identified for all three sets are labeled with a red “R.” Alone, these characteristics are not enough to declare all of these components non-neural.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4664645&req=5

Figure 3: Cortical component characteristics. Cortical locations, topographic maps, residual variance (RV) values for the full set and for each half set for the split-half comparisons (in parentheses), and average distance between analogous components for the full-set and each half set in Talairach coordinates are shown for the nine components with RV < 15% and neural locations for the full set. The word “None” appears when there were no components with neural locations and RVs < 15% for a given set. Components that were reliably identified for all three sets are labeled with a red “R.” Alone, these characteristics are not enough to declare all of these components non-neural.

Mentions: Seven of the nine components with RVs below 15% and locations in the brain shared similar locations and topographical characteristics (Figure 3). However, two revealed differences in location and topographical map characteristics from the other seven. These seven sources were generally located along the midline in the sensorimotor and parietal areas (Figures 2, 2, 3). They displayed topographical maps that appeared somewhat dipolar, but possessed asymmetries and abnormalities that are not typical of true neural sources.


Independent Component Analysis of Gait-Related Movement Artifact Recorded using EEG Electrodes during Treadmill Walking.

Snyder KL, Kline JE, Huang HJ, Ferris DP - Front Hum Neurosci (2015)

Cortical component characteristics. Cortical locations, topographic maps, residual variance (RV) values for the full set and for each half set for the split-half comparisons (in parentheses), and average distance between analogous components for the full-set and each half set in Talairach coordinates are shown for the nine components with RV < 15% and neural locations for the full set. The word “None” appears when there were no components with neural locations and RVs < 15% for a given set. Components that were reliably identified for all three sets are labeled with a red “R.” Alone, these characteristics are not enough to declare all of these components non-neural.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4664645&req=5

Figure 3: Cortical component characteristics. Cortical locations, topographic maps, residual variance (RV) values for the full set and for each half set for the split-half comparisons (in parentheses), and average distance between analogous components for the full-set and each half set in Talairach coordinates are shown for the nine components with RV < 15% and neural locations for the full set. The word “None” appears when there were no components with neural locations and RVs < 15% for a given set. Components that were reliably identified for all three sets are labeled with a red “R.” Alone, these characteristics are not enough to declare all of these components non-neural.
Mentions: Seven of the nine components with RVs below 15% and locations in the brain shared similar locations and topographical characteristics (Figure 3). However, two revealed differences in location and topographical map characteristics from the other seven. These seven sources were generally located along the midline in the sensorimotor and parietal areas (Figures 2, 2, 3). They displayed topographical maps that appeared somewhat dipolar, but possessed asymmetries and abnormalities that are not typical of true neural sources.

Bottom Line: The remaining 1% of sources had locations within the brain volume and low residual variances, but had topographical maps, power spectra, time courses, and event related spectral perturbations typical of non-neural sources.Caution should be exercised when interpreting ICA for data that includes semi-periodic artifacts including artifact arising from human walking.Separating true brain signals from motion artifact could clear the way for EEG brain computer interfaces for assistance during mobile activities, such as walking.

View Article: PubMed Central - PubMed

Affiliation: School of Kinesiology, University of Michigan Ann Arbor, MI, USA ; Department of Mathematics and Statistics, University of Minnesota Duluth Duluth, MN, USA.

ABSTRACT
There has been a recent surge in the use of electroencephalography (EEG) as a tool for mobile brain imaging due to its portability and fine time resolution. When EEG is combined with independent component analysis (ICA) and source localization techniques, it can model electrocortical activity as arising from temporally independent signals located in spatially distinct cortical areas. However, for mobile tasks, it is not clear how movement artifacts influence ICA and source localization. We devised a novel method to collect pure movement artifact data (devoid of any electrophysiological signals) with a 256-channel EEG system. We first blocked true electrocortical activity using a silicone swim cap. Over the silicone layer, we placed a simulated scalp with electrical properties similar to real human scalp. We collected EEG movement artifact signals from ten healthy, young subjects wearing this setup as they walked on a treadmill at speeds from 0.4-1.6 m/s. We performed ICA and dipole fitting on the EEG movement artifact data to quantify how accurately these methods would identify the artifact signals as non-neural. ICA and dipole fitting accurately localized 99% of the independent components in non-neural locations or lacked dipolar characteristics. The remaining 1% of sources had locations within the brain volume and low residual variances, but had topographical maps, power spectra, time courses, and event related spectral perturbations typical of non-neural sources. Caution should be exercised when interpreting ICA for data that includes semi-periodic artifacts including artifact arising from human walking. Alternative methods are needed for the identification and separation of movement artifact in mobile EEG signals, especially methods that can be performed in real time. Separating true brain signals from motion artifact could clear the way for EEG brain computer interfaces for assistance during mobile activities, such as walking.

No MeSH data available.


Related in: MedlinePlus