Page 143 - Handbook of Biomechatronics

140 Domen Novak

be as low as 0.1Hz (Novak and Riener, 2015). Most of the EEG paradigms
can then also be applied to ECoG. However, due to its higher SNR, it is
possible to use additional signal analysis paradigms that achieve much more
accurate estimation of the user’s desired motions. While EEG-based motor
imagery can only identify broad classes such as “move left arm” vs “move
right arm,” ECoG and intracortical electrodes allow “movement decoding”:
reconstruction of the detailed movement trajectory (actual or desired) from
the brain signal. Similarly to motor imagery analysis, this process usually
begins by extracting frequency features from a PSD estimated over a sliding
window. These features are transformed into an estimate of the desired
motion trajectory by means of linear regression (Chao et al., 2010) or more
advanced methods such as Kalman filters (Hochberg et al., 2012) and then
used as direct inputs to a biomechatronic device, for example, as the trajec-
tory of a BCI-controlled robotic arm.

1.3 Functional Near-Infrared Spectroscopy

Functional near-infrared spectroscopy (fNIRS) differs from EEG and ECoG
in that it measures the hemodynamic activity rather than electrical brain
activity, that is, it is a measure of blood flow. Specifically, it measures the
degree of tissue oxygen saturation and changes in hemoglobin volume using
near-infrared light (Ferrari et al., 2004). Near-infrared light (700–1000nm)
penetrates the skin, subcutaneous fat, skull, and underlying muscle/brain,
and is either absorbed or scattered within the tissue, with the degree of
absorption and scatter dependent on, among other things, the ratio of oxy-
hemoglobin to total hemoglobin within the tissue (Ferrari et al., 2004).
Since this ratio changes as a result of increased oxygen consumption due
to, for example, higher mental workload, fNIRS can be used to measure
the degree of activation of different brain regions.
A typical fNIRS sensor consists of a light source and a light detector, with
the two commonly placed on the scalp 3–5cm apart (Ferrari et al., 2004;
Naseer and Hong, 2015). The source emits a known amount of infrared
light through the scalp and skull toward the brain, and the detector measures
the amount of scattered light. Tissue oxygen saturation and brain blood flow
are then estimated from these optical density measurements via the modified
Beer-Lambert law (Naseer and Hong, 2015). While the response is slower
than EEG (often appearing a few seconds after a stimulus), it has the advan-
tage that it is less susceptible to data corruption by artifacts (e.g., blinks, mus-
cle activity) and offers better spatial resolution, allowing localization of brain