Single-Trial Analysis of Bioelectromagnetic Signals: The Quest for Hidden Information - Modeling in Computational Biology and Biomedicine - page 244

7.4.1 Co-variations Between EEG and fMRI Signals

Although functional MRI and EEG have very different characteristics and res-
olutions (Sect. 7.1 ), EEG single-trial analysis can make a bridge between these
complementary measures. The general idea is to use parameters extracted from the
EEG at the level of single trials, and test whether the fMRI signal from each brain
region is co-varying with the EEG -derived information (parametric analysis). The
existence of such co-variation can indicate that the region as defined by fMRI is at
the source of the EEG signal, or that both phenomena are modulated by the same
processes (e.g., attention). Parameters can represent the amplitude or latency of a
given wave on the evoked activity. This method has been used successfully with
two well-known evoked potentials: the error negativity [ 8 ], and the oddball-related
P300 [ 4 , 9 ](Fig. 7.9 ). A recent study has pushed even further the principle and used
information arising from high frequency gamma oscillations [ 18 ].

7.4.2 Distinction Between Latency and Amplitude Effects

in Evoked Potentials

As mentioned in the previous section, a classical marker used in EEG evoked
activity is the amplitude of a given wave. For example, the amplitude of the
P300 (a positive wave peaking around 300 ms after presentation of an unexpected
stimulus) has been used as a marker in many clinical applications [ 19 ]. However,
the amplitude of a wave on the average signal can in principle be influenced by two
factors at the single trial level (Fig. 7.10 ):

• The actual amplitude of the waves;

• An increased jitter in latency across trials.

There was a classical view in the literature that the amplitude of the P300
(response to rare events) depends on the number of frequent stimuli preceding the
rare event. However, in [ 13 ], it has been shown that in fact this could be attributed
to ISI (inter-stimulus interval) effects: when considering the single-trial amplitude,
there is no effect of latency. This can also be seen on Fig. 7.9 , where trials are sorted
by increasing reaction time. Long intervals between rare events tend to induce a
faster response: with cumulating frequent events, the probability of having a rare
event increases, and the subject is more prepared to respond. Therefore, the ISI is
inversely correlated with the reaction time. In addition, the P300 latency is very
highly correlated with the reaction time (as can be visualized on the sorted trials
figure, see also [ 4 ]). As a consequence, long ISI correspond to low variability in
latency of P300, and therefore to higher amplitude on average, similarly to what is
illustrated in Fig. 7.10 .