Signal Processing Application In Medical Equipment (EEG-BCI)
A Brain-Computer-Interface (BCI) is a device which enables the user to interact with its surrounding only by thought. The GRAZ-BCI is based on the detection of changes in electroencephalogram (EEG) rhythms that are modulated by motor imagery (MI). MI can be described as mental rehearsal of motor tasks without their execution. Motor tasks could be the imagination of squeezing a training ball, water paddling with both feet or playing an instrument. These thoughts can be used to generate control signal for any device, e.g. a neuro prosthesis or a communication device.
From the technical point of view, a BCI system operates in 4 phases:
· Signal Acquisition
· Signal Preprocessing
· Control and Feedback
EEG measures the electric activity happening in the brain. What is recorded is the voltage difference between minimum 2 electrodes. The EEG needs to be recorded simultaneously from multiple electrodes, in order to interpret ERP. During synaptic excitation of the dendrites in the neurons, electric currents are generated and picked up by the EEG. Because the signal detected is poor, being the electrodes far from the neurons and having the signal to travel through bones and skull, to record the electric flow is then required an amplifier.
- Electrodes-usually made of silver chloride
- A/D converters
- Recording Device
The electrodes acquire the signal from the scalp, the amplifiers process the analog signal to enlarge the amplitude of the EEG signals so that the A/D converter can digitalise the signal in a more accurate way. Finally, the recording device, which may be a personal computer or similar, stores, and displays the data.
Electrodes -The minimal configuration is composed by three electrodes: active electrode, reference electrode and ground electrode. The EEG measures the potential difference over time between signal or active electrode and the reference electrode. It is very difficult to get a reference where no electrical activity from the brain is present. Usually it is located on the mastoid, ear lobes or tip of the nose. The ground electrode is used to measure the differential voltage between the active and the reference points.
Amplifiers-The signal picked up by the electrodes is far away and attenuated by the different layers it has to travel. For this reason an amplifier is needed to bring the micro-volts to a range that can be digitized. The signal is sent to an amplifier through a cable measuring 1–2 meters. Unfortunately the cables can act as antenna and pickup signals, which would interfere with the EEG signal and cause noise to be amplified.
The A/D converter will convert the amplified signal from analog to digital form. The bandwidth for EEG signals is limited to approximately 100Hz, making 200Hz enough for sampling EEG signals.
It can be a computer or similar device, which will record, store and display the converted signal.
The preprocessing step helps to clean the data from the noise and artifacts. There are different methods and different steps in preprocessing. Often for example, filters are applied to the data. To remove the DC components of the signal and the drifts are employed high-pass filters, where usually a frequency cut-off of 1Hz is enough. Often also low pass filters can be applied to remove the high frequencies of the signal, because in EEG usually frequencies over 90Hz are not studied. Other methods are used to remove artifacts as the eyeball movements or eye blinking.
After different steps of preprocessing, when the signal is clean from most of the artifacts and noise, the recording is cut in epoch of few seconds: this allows us to have a large number of features from a single EEG recording, and to use them for statistics or to apply classifiers, as we will see in the next sections.
The next step is feature extraction: the analysis of the signal and extraction of information. As the EEG signal is very complex, it is impossible to find meaningful information just looking at it. It is needed then to apply processing algorithms which allows to find content (such as a person’s intent, for example) which would be hidden at a naked eye.
There are many methods for feature extraction, some of them are: Band powers (BP) Cross-correlation between EEG band powers frequency representation (FR) time-frequency representation (TFR) Hjorth parameters, parametric modelling inverse model and specific techniques used for P300 and VEP such as Peak picking (PP) and Slow cortical potentials calculation (SCPs)
Another step which can be applied to the signal, now mostly clean from artifacts, is to apply classification algorithms. Using machine learning techniques it is possible to train a classifier to recognize which features, for example, belongs to one or another class. mathematical Again, the classification helps to find out which kind of mental task the subject is performing
Translation and Feedback device
After the signal has been classified, the result is passed to the feature translation algorithm. At this point the features need to be translated in the corresponding action required. “For example, a P3 potential could be translated into the selection of the letter that evoked it”.
The feedback device receives the command from the translation step. For example it can be the computer, where the signal will be used to move a cursor, or it could be a robotic arm where the data are used to allow movement.
Transforms based techniques for EEG signal analysis
The wavelet packet method is utilized for finding the steady state visual evoked potential (SSVEP). The normalized power from special sub-wavebands was detected by this method. This is considered as a feature vector for the linear classifier. The SSVEP is an input signal of brain computer interface (BCI). The wavelet packet decomposition method is the extension of the wavelet transform. It splits the bands into two namely, low and high frequencies. The sub-waveband width for scale 4 is expressed in
where Fs is the sampling frequency and n is the number of sub-waveband.
The BCI system was used to detect the non-evoked and evoked EEG signals. After preprocessing method, the data was handled by wavelet packet decomposition to obtain the normalized power of 16 sub-wavebands. Then they compared the method with conventional FFT for performance measures. It is stated that the phase locking method and Hilbert-Huang transformation of SSVEP are also better than the FFT method.
Algorithms for EEG signal analysis
To detect the different facial expressions, EEG was recorded with facial electromayography. Early posterior negativity was enhanced at the range of 200–280ms. Spontaneous zygomatic activity was enhanced at the range of 500–750ms. At the range of 250ms the EMG changes occurred for faces, and then for every 500ms in scenes. It is stated that faster response occurred in facial expressions and weaker response to neural activity.
The hemodynamic changes in interictal epileptic discharges (IED) were represented by the fusion of EEG and fMRI analysis as shown in Fig. 2. To identify the irritative area in IED, a method EEG source imaging (ESI) was used. This method used to improve the blood oxygen level dependent (BOLD) signal. Then, the region of interest (ROI) is defined which consists of five ESI neighboring solution points nearby having maximum density. This is implemented in LAURA source localization algorithm because of the local dependency constraint. Finally, the average ROI density was identified using the continuous ESI function analysis.
Fusion process of EEG signal and fMRI imaging for localization of irritative zone in pre-surgical epilepsy cases. Interictal epileptic discharges (IED). Electrical source imaging (ESI). Continuous electrical source imaging (cESI). Hemodynamic Response Function (HRF).
The self-regulation of slow cortical potentials (SCP) method is proposed that aim to prevent the epileptic seizures in addition to link with paralyzed patients. The fusion of BCI and fMRI were used to the investigation of SCP. The four step processes were used in this study includes; training of SCP, SCP feedback in fMRI environment, functional imaging and artifact handling. The self-regulation of SCP monitored by the EEG and fMRI which allows to correlates the local BOLD responses and SCP changes.
The wavelet, shear-let and contour-let transforms can be utilized to detect epilepsy and other brain irregularities The transforms decompose the EEG signals into frequency sub-bands. The features are extracted from these sub-bands and fed to classifiers for detecting irregularities. The epileptic seizure has been detected by tunable-Q wavelet transform using kraskov entropy.