Wednesday, September 17, 2008



The experiments with a new ABI system were performed with three subjects in the Laboratory of Computational Engineering in February 2002. In the beginning of this chapter the hardware and software of the new ABI system are reviewed. The experimental methods and the online results are then presented along with the discussion. In addition, the subject reports about the mental task strategies and the feedback experiences are also presented in this chapter.
4.1 The ABI system:
The ABI system consists of the hardware, which is used to measure the EEG and four Microsoft Windows based programs: EEP H11 Server, ABI Learning and online programs and ABI Visualization.
4.1.1 Hardware:
The hardware used in these experiments was different from the hardware previously used with the ABI (see the section 3.2). The EEG system here was Electro Encephalo Processor model H11 (EEP H11) manufactured by Fase Sistemi, one of the partners of the ABI project. The whole system consists of EEG device, battery, battery charger and electrode cap. The EEG device is small, its dimensions are only 160 _ 110 _ 55 mm and it weights 700 g.
The system has 38 analogical single ended channels and 2 electrical references. The analogical signal from each electrode is converted to digital signal with 16 bit resolution and sampling rate of 6800 Hz. The EEG device is battery powered. The data is transmitted to the computer via fiber-optic cable. This makes the EEG device isolated from the surrounding electrical fields. The electrode cap of medium size (54-58 cm) is ECI Electro-Cap Electrode System manufactured by Electro-Cap International, Inc. The cap has 32 electrodes for scalp EEG and 6 additional electrodes, which can be used as reference electrodes and measuring the EOG. Cap, EEG device and battery can be seen in Figure 4.1.
4.1.2 Software:
Three Microsoft Windows based programs are needed in order to operate the ABI system: EEP H11 Server, ABI Learning program, and ABI online program. The EEG signal can be visually inspected with the ABI Visualization program.
EEP H11 Server and ABI Visualization:
This program acts as a server (application programming interface (API)) for the ABI applications. In other words, it sends the EEG signal coming from the EEP H11 system to the ABI online and the ABI Visualization programs. The electrodes used with the ABI are selected with this program. The EEG signals can be seen in the main program window.
It is possible to filter the EEG signals, for example, if they contain the 50-Hz component (see the section 2.4.4).
ABI Learning:
This program is used to train a set of prototypes from the recorded data. Each prototype corresponds to one particular mental task as described in the section 3.2.1. The classifier will then use these prototypes in the online classification of the EEG signals. In order to train a new set of prototypes, the processed data of one recording is loaded up into the program. Usually, the data without transitions is used. The data does not contain the initial resting period. The mental tasks are labeled in the data. The data is then divided into training and validation sets in such a way that three consecutive samples go to the training set and every fourth one goes to the validation set.
A new set of prototypes can be trained using the training and validation sets. It is also possible to load old prototypes or use a clustering method, the Self-Organizing Map (SOM), to create a set of prototypes from the training set. The old prototypes or those created by the SOM can then be used as a starting point for training of new prototypes.
ABI online program:
This program acts as a client for the EEP H11 Server, i.e., the program gets the EEG data from the server. This data can be be saved in two formats: Raw data and processed data. The raw data is the data coming from the EEP H11 Server. The processed data goes through the feature extraction methods explained in the section 3.2.1 expect that in these experiments a spline Laplacian method [4] was used instead of the surface Laplacian. The classifier uses the processed data. The raw data is saved for later evaluation.
The program can also be used to give feedback for the subject. In order to do this, the prototypes trained in the Learning program must be loaded for each of the mental tasks.
The probability and the distance thresholds of the classifier can be adjusted. It is also possible to choose either positive or total feedback. Positive feedback means that only the correct classifications are shown for the subject. If total feedback is used, all the classifications are shown for the subject.
4.2 Experimental methods:
This section describes the methods used in the experiments. The experiments were done in the Laboratory of Computational Engineering, in the Helsinki University of Technology, during five consecutive days starting from 4th of February and ending 8th of February 2002. The experiments were made in 2_4 meters room (see Figure 4.4).
4.2.1 Subjects:
Two healthy male subjects (subjects TN and JL) and one healthy female subject (subject LL) in their mid-twenties were the test subjects. Both male subjects were familiar with the ABI system (subjects TN and JL). They had used it a couple of times before these tests, but no actual training were done with them. The ABI was new to subject LL.
4.2.2 Mental tasks:
For the subject unfamiliar with the ABI (subject LL), the first step was to introduce the system to her. She was able to try the mental tasks (form the list presented in the section 3.2.2) by herself and decide which three were the most suitable for her. Before the actual first measurement a test run was made with these tasks in order to introduce the whole procedure to the subject.
Two of the subjects (TN and JL) used the mental tasks they were familiar with. Subject JL used relax, subtraction and right hand. Subject TN used relax, cube and left hand. The third subject, LL used three different combinations: Relax, subtraction and right (days 1 and 2), relax, left and right (days 2 and 3) and relax, right and words (days 3, 4, and 5). Three different combinations were used because the first two did not work.
4.2.3 Data acquisition:
The same electrodes were used in these tests as with the ABI before (F3, F4, C3, Cz, C4, P3, Pz and P4). The signals were referenced to the right ear. The ground electrode was in the left ear. The signals were checked with the ABI Visualization program and if they were noisy, a bandstop filter of order 24 and the frequency band from 45 Hz to 55 Hz was selected in the EEP H11 Server. Instead of the local surface Laplacian method used previously with the ABI, the spline Laplacian method was applied to the raw data. In addition, before each recording the EEG signals were visually checked with the ABI Visualization program to ensure that all electrodes were giving good signals. Each recording lasted about 300 seconds, including the initial resting period (60 s). After the resting period, the operator instructed the subject which mental task to perform first.
The operator then changed the mental tasks so that each mental task lasted between 10-25 seconds. The order and length of the mental tasks were random. Three to five recordings were done in one session (day). There was about 5-10 minutes break between each recording.