CAPÍTULO III. PROPUESTA DE ESTRATEGIA A SER IMPLEMENTADA
3.4 Herramientas y Canales de comunicación interna en área operativa
EEG is used to record the electrical potentials created by the brain near its surface, so electrodes are placed on the scalp, or on the cortex itself. EEG recording systems have
2.5 Electroencephalography (EEG) 25
four key parts: electrodes with conductive media, amplifiers with filters, an analogue to digital (A/D) converter, and a recording device. The electrodes interpret the signal taken from the head’s surface, where amplifiers then move the microvolt signals into a range that can be reliably digitised. The converter then shifts the signals into digital form from analogue, before finally a personal computer or similar device is used to save and display the gathered data (Teplan et al., 2002).
In 1958, the International Federation of Electroencephalography and Clinical Neu- rophysiology (IFCN) established a standard for electrode placement known as the 10-20 electrode placement system (Jasper, 1958a). This approach created a universal standard for physical placement and naming for electrodes on the scalp. In this system, the head is divided into proportional distances from important skull landmarks such as the nasion, preauricular points, and inion in order to offer enough coverage across all of the brain’s areas. Label 10-20 describes a proportional distance (in percent) from the ears and nose to where the electrode positions are established. Electrode positions are labelled based on the brain areas involved: F (frontal), C (central), T (temporal), P (posterior), and O (occipital). In addition to the letters, the left side of the head is assigned odd numbers, and the right side is assigned even numbers (Figure 2.4). The left and right sides are based on the perspective of the subject.
2.5.2
Neurophysiological signals in EEG used for BCI-based
communication
The EEG activities commonly used in BCI may be categorised into three groups, depending on the component of interest: evoked related potentials (ERPs), slow cortical potentials (SCPs), and event-related de-synchronisation (ERD/ERS). Amiri et al. (2013) compared EEG activities based on four factors: accuracy, information rate, training time, and number of required EEG channels. They considered the ERPs brain activities to be the best because they are accurate, have a high information rate, and require less training time and using fewer EEG channels.
Fig. 2.4 Electrode locations of International 10-20 system for EEG (electroen- cephalography) recording (Wiki Commons-released to the public domain)
• Evoked signals are produced because of an external stimulus.
• Spontaneous signals are voluntarily produced by the user after an internal cognitive process, with no need to any external stimulations.
Evoked signals/ Evoked related potentials (ERPs)
ERP presents changes in the activity of neuronal populations, which can be detected at specific time delays after the appearance of a stimulus (Pfurtscheller and Da Silva, 1999). To enhance the signal-to-noise ratio, averaging techniques are used to detect these signals. Such activities have been used in the literature for two purposes: for communication and for understanding brain activities related to communication. Table 2.2 summarises the ERP features that have been investigated in the literature to date. As mentioned above, some of these activities are used for the direct control of communication systems such as P300, ErrPs, SSVEP, and N200.
• P300
P300 was first described by Sutton et al. (1965). The amplitude and latency of P300 depends on a number of factors, including the inter-trial interval, the
2.5 Electroencephalography (EEG) 27
probability of target stimulus, and the user’s attention level. P300 can be detected from central electrodes such as Cz on a 10-20 international system. As a temporal pattern, the amplitude of P300 mostly falls in the range of 2–5 V, with a duration of 150–200 ms (Amiri et al., 2013). The P300-based speller is the most widely used application of P300 in BCI systems. Four paradigms are used for the design of these spellers: the row/column paradigm, the checkerboard paradigm, the single-character paradigm, and the region-based paradigm (Fazel-Rezai et al., 2012).
• Error-Related Potentials (ErrPs)
To detect ErrPs, EEG signals are recorded from the frontocentral region, from Fz to Cz in 10-20 systems (Dehaene et al., 1994). In BCI systems, ErrPs is often used as an automatic error-detection mechanism. For example, (Dal Seno et al., 2010) used ErrPs as an automatic detection tool for a P300 speller. If ErrPs are elicited after the presentation of a letter chosen by the P300 speller, then the chosen letter will be cancelled.
• Steady State Evoked Potential( SSVEP)
SSVEP usually includes the same fundamental frequency as the stimulus and a few harmonics of the fundamental frequency. For example, when a visual stimulus at a frequency ranging from 3.57 Hz is displayed, the brain generates electrical activity at in similar frequency, as well as at that frequency’s harmonics (Amiri et al., 2013). Light-emitting diodes (LEDs) are usually used in BCI systems to generate SSEVP.
SSEVP has been used to implement BCI spellers. For example, Chen et al. (2015) proposed an alphanumeric keyboard in which each key flicker targets a specific frequency and phase. The system can identify a target letter by detecting the elicited SSVEP frequency and phase in the user’s EEG signal.
• N200 using motion-onset visual response
(2009) used motion stimulus to generate the N200 signal in order to develop an N200 speller. They embedded motion stimuli into 6 x 6 matrixes that included alphabet buttons. In the study, motion was represented by vertical bars that appeared and moved leftwards for 140 ms at 200 ms intervals. The colour of the vertical bar, which could be red, green, blue, purple, yellow, or brown, was designated randomly in a protocol such that the colours of the six bars in the same row or the same column would be different from each other. The researchers compared the N200 speller with the P300 speller; their results showed that N200 could deliver performance comparable to P300 in terms of accuracy and needs less number of training trials.
• Hybrid BCI paradigms
Any BCI system has certain disadvantages that prevent some users from adopting the system (Amiri et al., 2013; Wang et al., 2015). Recently, several researchers have tried to combine different BCI technologies (known as hybrid BCI) to develop a system suitable for a large number of users, or to include several tasks so that users could select a suitable BCI activity for each task (Amiri et al., 2013). This combination can be performed when each system has its own input (simultaneously) or when the output of one of the systems serves as the input of
another system (sequentially).
One example of a hybrid BCI system is the SSVEP/P300 BCI system developed by Wang et al. (2015). In this system, the researchers used changes in shape to elicit P300 and flickering to elicit SSVEP. They found that the general users’ performance was better than that of the users who employed single-type BCI. The results also indicated that the use of shape change is preferable to the use of flashes for eliciting P300 with hybrid systems, because flashes cause a checkerboard phenomenon when presented along with flickering in the same system.
2.5 Electroencephalography (EEG) 29
Table 2.2 Example of ERP patterns that are examined in the literature ERP Fea-
tures
Description
N100 or N1 A negative deflection observed when a stimulus is presented unexpectedly. The peak occurs between 90-200 ms after the stimulus.
P200 or P2
A positive deflection. The peak occurs approximately 100 ms to 250 ms after the stimulus. It is believed that the N1/P2 component of
ERP may be characteristic of an individual’s thrill-seeking behaviour. N200 or N2 A negative deflection. The peak occurs approximately 200 ms after onset
of the stimulus.
P300 P300 is a large and positive peak amplitude that,can be detected around 300 ms after the onset of a rare but relevant,stimulus.
N400 A negative wave, related inversely to the expectancy that a given word will form the end of a sentence. N400 is seen 300 ms to 600 ms after a stimulus. It was first reported with regard to semantic incongruity. P600 An effect relevant to language processing. Sentences with syntactic
errors, with a poor syntactic structure, or with a particularly complicated syntactic structure are associated with it.
SSVEP SSVEP is evoked in the visual cortex as a response,when there is a repetitive stimulus with a constant frequency on the central,retina. ErrP /ERN
Error-Related Potentials/ Error-Related Negativity is the average amplitude of the waveform at 50 to 100 ms after the onset of an aware error.
Spontaneous signals
• Slow Cortical Potentials (SCPs)
SCPs is defined as slow changes in the voltage generated in the cortex, occurs over 0.51 seconds. SCPs is one of the lowest-frequency features recorded using EEG technologies. Negative SCPs are typically related to movement and any functions containing cortical activation, while positive SCPs are usually linked to a decrease in cortical activation (Wolpaw et al., 2000). Subjects can be trained to produce positive or negative SCPs, depending on the required task. In the literature, SCPs have been used to examine the potential to have communication systems for paralysed patients Birbaumer et al. (1999, 2003). SCPs are no longer used in BCI research, however, because of their lack of speed and their training time limitations.
• Event-Related Desynchronization/Synchronization (ERD/ERS) ERD/ERS relays on the recording of rhythmic activities over the sensorimotor cortex. In ERD/ERS, instead of having an external stimulus to generate a command as in ERPs, a user can voluntarily generate a command by controlling his or her brain activity, by imagining motor movements, or by any other activity (Rao and Scherer, 2010). ERPs have an advantage over ERDs, in that they are stable over time, while ERDs vary over time because the user can change the signals after receiving feedback. ERDs can be used for communication purposes, for example controlling BCI spellers based on hand and foot movement. The following is an overview of previous studies conducted on the use of BCI for speller control.
– Motor imagination to control BCI spellers
Guenther and Brumberg (2011) conducted a one-hour pilot study session with a single subject. The participant was asked to repeat three vowel sounds, with 20 repetitions of each sound: /AA/, /IY/, and /UW/. To represent each vowel, a different motor imagination action was linked with each sound: left-hand movement for /UW/, right-hand movement for /AA/, and foot pressing for /IY/. Limb imagery was used to ensure reliability and to obtain EEG responses. The recognised vowels were presented both visually and acoustically.
In (Perdikis et al., 2014), the authors’ main aim was to evaluate a BCI speller based on a motor-imaginary system called ’Braintree’. Left- and right-hand/ foot movement imagination was used for the selection of letters, and EMG signals represented the ’undo’ command as an error-handling mechanism. To solve the problem of the limited options for brain actions that are used compared to the 26 letters in the English alphabet, the researchers used a language model as a data-compression technique to reduce the options available to the users. The letters were represented each time as a binary tree, so the model helped to redistribute the letters each time a letter was
2.5 Electroencephalography (EEG) 31
selected. The model is called “prediction by partial matching”. Braintree was tested on 16 end-users (6 disabled users and 10 able-bodied users). The evaluation had two phases: training and spelling. In the training phase, two motor-imaginary tasks were chosen, and the classifier was trained based on the selected data. In the spelling phase, the users were asked to spell four specific words, and there was no time constraint. All users were able to complete the spelling tasks; their performance was 1.7 characters per minute (cpm) on average, with 3.6 cpm being the maximum typing speed. Dalbis et al. (2012), who developed a predictive speller controlled by motor- imaginary BCI actions, improved the speller’s performance using a predictor. Three motor-imaginary actions were selected: (1) movement of the right and left hands, (2) movement of both hands for the alphabet, and (3) feet movements to undo actions. The alphabet was divided into three groups in the system interface; each time a user chose a letter, the groups were changed based on suggestions from the language model. Three subjects tested the system, achieving spelling rates of 3 cpm, 2.7 cpm, and 2 cpm.