Brain-Computer Interface - site visit 2

Online Adaptive Brain-Computer Interface, a project of the Laboratory of Computational Engineering under Academy Professor Mikko Sams' leadership at the Helsinki University of Technology.

Site visit in Espoo on 29 October 2004 

The aim of the project is to recognise and classify different brain activation patterns associated with selected mental tasks. In a typical test situation, the subject is asked to think about movement, the mental activity is measured by sensors attached to the skin of the skull, a classifier classifies and recognises the electrical signals associated with the thinking, and some activity is performed based on this classifcation.

Such a brain-computer interface (BCI) is intended for helping severely motor-disabled as well as healthy people to operate electrical devices. The group uses an approach based on artificial neural networks to classify EEG or MEG signals measured from the brain. The task is challenging as the signal-to-noise ration is high and the signal might change substantively in a very short time period. 

The first approach to build a BCI has been to robustly locate and classify finger movements. Test subjects were asked to think of either lifting their right or their left index finger. The task includes preprocessing of the data, feature extraction and selection, and feature classification. The group has performed tests with both healthy and tetraplegic (quadriplegic) subjects that are paralysed from their neck down. Results show that two types of features are easily recognised in healthy subjects, a premovement contralateral desynchronization in time (the subject is starting to think about to move his or her finger), and a postmovement contralateral rebound activity about 1 to 1.5 seconds after the movement (the movement is over, everything goes back to normal).

The group used an offline study to find the right tools for spectrum estimation and classification related to post-movement features extracted from MEG recordings. The study showed that MEG postmovement features can be used to effectively classify movements. In a test case with five healthy subjects, the BCI had an accuracy of 83% in recognising the movement (which hand, left or right, was moved). The classification was done by a using a radial basis function (RBF). 

The features, however, vary strongly with time. By using particle filters for sequentially updating the parameters of the RBF classifier, the group was able to improve classification with postmovement features from 85% to 88% and with premovement features from 89% to 92%. When uncertain classifications are rejected the improvement doubles with both pre- and postmovement features. It is to be noted though, that these test results were achieved with healthy subjects. Disabled subjects did about fifteen percentage units less well in all cases.

The group is now looking into to making the classification more efficient. Preliminary results with a Rao-Blackwellisation of the sequential process show a 40% increase in computational efficiency with no loss in performance. The group also uses a dynamic autoregressive (AR) model to detect the onset of finger movement; in the above tests, the user issued commands when the system was ready to receive them. In an asynchronous BCI, the system should be able to detect when the movement starts. Also here, results show that detecting the point of onset does not affect classification accuracy compared to marked onset. Also, the system has so far produced no false positives (detected movement when there is none); however, more tests still need to be performed. 

Disabled subjects (patients) differ from healthy subjects. The BCI system seems to perform less well with them. For example, patients do not seem to have the rebound activity of healthy subjects, at least not on their first try. Patients that have been longer paralysed also seem to do less well than recently injured patients. Patients, however, show activity in both sides of their brain, as opposed to healthy subject, who have contralateral activity (left brain half shows activity when lifting right finger and vice versa). In dealing with disabled subjects, the group works closely with the Käpylä Rehabilitation Centre.

The group has built a Brain-Computer interface platform for measuring signals online. The platform transfers the data from measurement devices to Matlab where different models can be selected for classification  

Some recent publications related to the project

• Tommi Nykopp, Laura Kauhanen(née Laitinen), Jukka Heikkonen, and Mikko Sams Statistical Methods for MEG based finger movement Classification. IEEE Transactions on Neural Systems and Rehabiliteering Engineering (in press)
• Tommi Nykopp, Laura Kauhanen, and Mikko Sams: Statistical Analysis of Neuromagnetic Activity of Healthy and Tetraplegic Subjects. In Proceedings of the Proactive Computing Workshop, PROW 2004, Helsinki, November 25-26, 2004, pp. 72-76
• Laura Kauhanen (née Laitinen), Pekka Rantanen, Janne Lehtonen, Iina Tarnanen, Hannu Alaranta, and Mikko Sams: Sensorimotor cortical activity of tetraplegics during attempted finger movements. Biomedisinische Tecknik, Proceedings of the 2nd International Brain-Computer Interface Workshop and Training Course, Graz, Austria, September, 17-18, 2004, 49, 59-60. This Poster was given the poster award of the conference.
• J.A. Lehtonen, T. Nykopp, J. Heikkonen, M. Sams. Brain Computer Interface platform, Biomedisinische Tecknik, Proceedings of the 2nd International Brain-Computer Interface Workshop and Training Course, Graz, Austria, September, 17-18, 2004, 49, 67-68
• T. Nykopp, J. Heikkonen, M. Sams: Sequentical classification of finger movements from MEG recordings, Biomedisinische Tecknik, Proceedings of the 2nd International Brain-Computer Interface Workshop and Training Course, Graz, Austria, September, 17-18, 2004, 49, 77-78
• Laura Laitinen: Neuromagnetic sensorimotor signals in brain computer interfaces. MSc Thesis (Tech), Department of Electrical and Communications Engineering, HelsinkiUniversity of Technology, 2003.
• Janne Lehtonen and Mikko Sams: Ajatus ohjaa tietokonetta. Tiede 2/2003. 

Research group

HelsinkiUniversity of Technology (HUT), Laboratory of Computational Engineering, Research Centre for Computational Science and Engineering:

• Academy Professor Mikko Sams, Academy Research Fellow Jukka Heikkonen, and PhD students Laura Kauhanen, Janne Lehtonen and Tommi Nykopp 

For more information, please see the web page of the group at www.lce.hut.fi/research/bci/