Biomedical Engineering Reference
In-Depth Information
In contrast to single-unit activity (SUA) recordings, ECoG measures population
activity, which offers a better prospect for long-term recording stability (Chao
et al.
2010
). Furthermore, as ECoG recording does not penetrate the cortex, signal-
prohibitive encapsulation, an obstacle in chronic SUA, multiunit activity (MUA),
and local fi eld potential (LFP) recordings, is less likely to occur during long-term
implantation (Vetter et al.
2004
; Szarowski et al.
2003
; Bjornsson et al.
2006
).
Compared with electroencephalography (EEG), ECoG offers higher spatial reso-
lution, broader bandwidth, and higher amplitude and is less sensitive to artifacts
such as electromyogram (EMG) (Freeman et al.
2003
; Schwartz et al.
2006
).
Compared with functional magnetic resonance imaging (fMRI) and near-infrared
spectroscopy (NIRS), which are based on the blood oxygenation level, ECoG
offers direct measures of neural activity with signifi cantly higher temporal resolu-
tion. In conclusion, ECoG provides a great balance between signal fi delity, tempo-
ral and spatial resolutions, long-term durability and stability, and capability to
cover multiple brain regions. Therefore, ECoG may be the optimal choice for mining
spatio-spectro- temporal cortical dynamics.
Successes in ECoG research have accumulated greatly during the past decade in
the areas of neuroscience and neuroengineering and especially in the fi eld of brain-
machine interfaces (BMIs) [see reviews in (Donoghue
2002
; Mussa-Ivaldi and
Miller
2003
; Nicolelis
2003
; Lebedev and Nicolelis
2006
; Patil and Turner
2008
)].
One recent development of ECoG recording technology enables 288-channel
recordings with submillimeter and submillisecond resolution (Viventi et al.
2010
).
Our laboratory also developed a 256-channel ECoG system that can cover most of
the cortex, including structures in the medial wall (Nagasaka et al.
2011
). These
advances pose a challenge to data analysis regarding how to extract relevant infor-
mation from terabytes of data with satisfactory thoroughness and effi ciency.
Depending on the goal of specifi c studies, ECoG analysis can be generally clas-
sifi ed into two main categories: offl ine and online analyses. Offl ine analysis aims to
identify statistically relevant features in ECoG signals underlying the neural pro-
cesses of interest (Sect.
3.2
). Conversely, online analysis focuses on establishing a
decoding model that can translate ECoG signals to specifi c sensory inputs, motor
outputs, or cognitive processes in real time (Sect.
3.3
). Online analysis is usually,
but not exclusively, used in BMI applications, where real-time interpretation of
brain activity is used for either controlling external devices or estimating cognitive
states. Both offl ine and online analyses can provide insights into how the brain
encodes information, and both require the mining of high-dimensional data to
extract relevant characteristics in the signals.
Offl ine and online analyses are not unique to ECoG signals. Many tools used in
ECoG analysis originate from EEG analysis and are shared with analyses of fMRI
and other electrophysiological technologies, such as magnetoencephalography
(MEG). Here, we will not focus on detailed theoretical backgrounds. Instead, our
goals are to provide the following:
• Routines for mining multichannel ECoG signals, particularly in the frequency
domain, as frequency bands in ECoG signals have distinctive functional
interpretations
