A Multifaceted Approach to Covert Attention Brain-Computer Interfaces
Over the last years, brain-computer interfaces (BCIs) have shown their value for assistive
technology and neurorehabilitation. Recently, a BCI-approach for the rehabilitation of hemispatial
neglect has been proposed on the basis of covert visuospatial attention (CVSA).
CVSA is an internal action which can be described as shifting one's attention to the visual
periphery without moving the actual point of gaze. Such attention shifts induce a lateralization
in parietooccipital blood flow and oscillations in the so-called alpha band (8-14 Hz),
which can be detected via electroencephalography (EEG), magnetoencephalography (MEG)
or functional magnetic resonance imaging (fMRI). Previous studies have proven the technical
feasibility of using CVSA as a control signal for BCIs, but unfortunately, these BCIs could not
provide every subject with sufficient control. The aim of this thesis was to investigate the
possibility of amplifying the weak lateralization patterns in the alpha band - the main reason
behind insufficient CVSA BCI performance.
To this end, I have explored three different approaches that could lead to better performing and
more inclusive CVSA BCI systems. The first approach illuminated the changes in the behavior
and brain patterns by closing the loop between subject and system with continuous real-time
feedback at the instructed locus of attention. I could observe that even short (20 minutes)
stretches of real-time feedback have an effect on behavioral correlates of attention, even when
the changes observed in the EEG remained less conclusive. The second approach attempted
to complement the information extracted fromthe EEG signal with another sensing modality
that could provide additional information about the state of CVSA. For this reason, I firstly
combined functional functional near-infrared spectroscopy (fNIRS) with EEG measurements.
The results showed that, while the EEG was able to pick up the expected lateralization in
the alpha band, the fNIRS was not able to reliably image changes in blood circulation in the
parietooccipital cortex. Secondly, I successfully combined data from the EEG with measures
of pupil size changes, induced by a high illumination contrast between the covertly attended
target regions, which resulted in an improved BCI decoding performance. The third approach
examined the option of using noninvasive electrical brain stimulation to boost the power of
the alpha band oscillations and therefore render the lateralization pattern in the alpha band
more visible compared to the background activity. However, I could not observe any impact of
the stimulation on the ongoing alpha band power, and thus results of the subsequent effect
on the lateralization remain inconclusive.
Overall, these studies helped to further understand CVSA and lay out a useful basis for further
exploration of the connection between behavior and alpha power oscillations in CVSA tasks, as well as for potential directions to improve CVSA-based BCIs.
EPFL_TH9052.pdf
openaccess
11.07 MB
Adobe PDF
35f4501857fcb6d9497fe1a471d7359e