Difference between revisions of "Integrating Motor Imagery and Error Potentials in a Brain-Computer Interface"

From AIRWiki
Jump to: navigation, search
(Part 3: Project tracking)
(Part 2: Project description)
Line 52: Line 52:
  
 
A method taking into account four MI classifications for each of the six seconds of a trial (i.e., a singular motor imagery recognition) and an equivalent number of ErrP detections has been developed. The idea is to provide a feedback after each second of the motor imagery activity, and to verify the possible presence of an error potential for each feedback presentation. The results collected from the two analysis are used to infer the most probable user motor intent. A Bayesian classifier employing data from training sessions is applied to produce the desired result.  
 
A method taking into account four MI classifications for each of the six seconds of a trial (i.e., a singular motor imagery recognition) and an equivalent number of ErrP detections has been developed. The idea is to provide a feedback after each second of the motor imagery activity, and to verify the possible presence of an error potential for each feedback presentation. The results collected from the two analysis are used to infer the most probable user motor intent. A Bayesian classifier employing data from training sessions is applied to produce the desired result.  
 +
 +
 +
[[Image:Pipe.png]]
 +
  
 
Motor imagery classification proved to perform particularly well over a subject, achieving accuracies up to 75%. Other subjects achieved inferior but still interesting accuracies of 55%, 59% and 62%. The error potential detection reached an overall accuracy of 65% for one test subject. The proposed method to integrate the results of the two classifications processes lead to an increase in performance of 10\% in the subject achieving the 62% accuracy in the motor imagery paradigm. This result is encouraging as applied to a spelling interface leads to an increase of the spelling rate from 0.7 to 2.2 characters per minute.
 
Motor imagery classification proved to perform particularly well over a subject, achieving accuracies up to 75%. Other subjects achieved inferior but still interesting accuracies of 55%, 59% and 62%. The error potential detection reached an overall accuracy of 65% for one test subject. The proposed method to integrate the results of the two classifications processes lead to an increase in performance of 10\% in the subject achieving the 62% accuracy in the motor imagery paradigm. This result is encouraging as applied to a spelling interface leads to an increase of the spelling rate from 0.7 to 2.2 characters per minute.

Revision as of 19:09, 7 December 2009

Part 1: Project profile

Project name

Integrating Motor Imagery and Error Potentials in a Brain-Computer Interface based on Motor Imagery: A Case Study

Project short description

This project is aimed is to control an external device through the analysis of brain waves measured on the human scalp.

Dates

Start date: 01/12/2008

End date: 01/12/2009

Website(s)

http://airlab.elet.polimi.it/index.php/airlab/theses_lab_projects/brain_computer_interfaces_based_on_motor_imagery

People involved

Project head(s)
Other Politecnico di Milano people
Students ivolved in the project

Laboratory work and risk analysis

Laboratory work for this project will be mainly performed at AIRLab-IIT/Lambrate. The main activity consists in the acquisition of brain signals through an EEG amplifier for on-line or off-line processing. This is a potentially risky activity since there is an electrical instrumentation that is in direct contact with the human body. It is thus important to keep the system isolated from the power line. The EEG amplifier (as all biomedical instrumentations) is certified by the vendor to be isolated and the acquired data are transferred to the PC using an optic fiber connection . Anyhow for increased safety the PC and any other electronic device connected to the system must be disconnected from the power line. Standard safety measures described in Safety norms will be followed.


Part 2: Project description

Brain-computer interfaces (BCIs) make use of brain signals to transform specific patterns in subject's brain into control for an external device such as a wheelchair, a computer or a software application. Among the different technologies available to acquire the neural signal, the electroencephalography (EEG) has been chosen. At present, only EEG allows indeed to easily record and process the brain signal with inexpensive equipment, leading to a practical possibility of a new non-muscular and, mostly important, non-invasive communication channel. The most relevant drawback of using this type of approach is the strong presence of noise in the acquired signal appearing as the sum of potentials produced from various locations of the cortex and easily propagating through the head and being affected by muscular and ocular activity.

Several paradigms have been developed in literature to exploit specific patterns in brain activity and to transform them into control signals for the considered device. In this thesis, the attention has been focused on the motor imagery (MI) and on the error potential (ErrP) paradigm.

Motor imagery is a well studied technique of detecting changes in the electrical charge of cortical areas normally dedicated to the control of body movement. The imagination of specific body parts movement (e.g., a hand, a foot or the tongue) causes some precise areas of the motor cortex to change their electrical activity, increasing or decreasing their activity over some well defined frequency. The activities detectable at those frequency have been called rhythms. In particular, the so called \textitmu and beta rhythms which are located respectively in the bands 8-12 Hz and 15-22 Hz. By means of a spectral analysis over these bands the possibility to infer the imagined movement has been demonstrated.

Event related potentials (ERPs) is a family of stimulus responses observable in acquired EEG signals as a common pattern happening just after some interesting, rare or unexpected event. Error potentials are a subset of ERPs generated by the awareness of making a mistake or by receiving unexpected results from the interface. In these situations a particular shape composed by a sequence of a negative potential and a subsequent positive one can be detected in the EEG recordings a few instants after receiving the stimulation. The detection of such a pattern can provide useful information to improve the overall performance of the BCI.

The aim of this thesis is that of developing an interface able to provide both spectral and temporal analysis of the EEG signal to allow its classification into one of four possible classes (i.e., left, right, up, down) basing the analysis on the mentioned paradigms. The attention is focused on the individuation of a method capable to classify the user's motor imagery activity and, at the same time, to detect error potentials on the presented result to assess its validity.

A method taking into account four MI classifications for each of the six seconds of a trial (i.e., a singular motor imagery recognition) and an equivalent number of ErrP detections has been developed. The idea is to provide a feedback after each second of the motor imagery activity, and to verify the possible presence of an error potential for each feedback presentation. The results collected from the two analysis are used to infer the most probable user motor intent. A Bayesian classifier employing data from training sessions is applied to produce the desired result.


Pipe.png


Motor imagery classification proved to perform particularly well over a subject, achieving accuracies up to 75%. Other subjects achieved inferior but still interesting accuracies of 55%, 59% and 62%. The error potential detection reached an overall accuracy of 65% for one test subject. The proposed method to integrate the results of the two classifications processes lead to an increase in performance of 10\% in the subject achieving the 62% accuracy in the motor imagery paradigm. This result is encouraging as applied to a spelling interface leads to an increase of the spelling rate from 0.7 to 2.2 characters per minute.

Part 3: Project tracking

  • literature about motor imagery in BCI field has been read
  • initial sessions for signal analysis have been performed
  • a feature extractor and a classifier have been trained to interpret brain's motor imagination
  • some utilities to deal with large amount of data have been deployed
  • everything has been ported from Matlab to BCI2000
  • feedback sessions have proven the functionality of the previous work
  • a BCI speller has been tested with the output of the classifier
  • Start studying material from course of Methodologies for Intelligent Systems
  • Start studying of ERP for applications in motor imagery
  • Read "Toward An Integrated P300 And ErrP-Based Brain-Computer Interface" Dal Seno B.
  • Analyzing ErrP caused by classification results in a MI task
  • Deploying a MI interface with visual synchronization signal
  • Interface with discrete feedback presentation ready for use
  • Interface with early classification feedback ready for testing
  • Started testing of newly created interfaces
  • Analizing P300+Errp code for merging
  • Brand new interface for better stimulation of potentials developed
  • Major improvement in file management for Matlab offline analysis
  • Set up of acquisition software for the new interface
  • Development of scripts to approximatively determine the end of training phase
  • Merging and adapting code of P300+ErrP to motor imagery tasks
  • Achieving a method for pipe branching in BCI2000
  • Beginning acquisition of data with ErrPs
  • Ended deployment of execution pipe, from acquisition to extraction and collection of ErrPs
  • Focusing on MIS to learn methods for smart use of error probability in class selection
  • Bayesian classifier developed to merge Motor Imagery and Error Potentials
  • Exhaustive mapping of solutions implemented for motor the imagery only case
  • Added error potentials to mapping
  • Tree organization of results and result explorer implemented
  • Testing online performances
  • Performing offline analysis

Part 5: References

  • Control of two-dimensional movement signals by a noninvasive brain-computer interface in humans, Wolpaw J.R., McFarland J., PNAS, vol. 101, no. 51, december 2004, pages 17849-17854.
  • Brain Computer interfaces for communication and control, Wolpaw J.R., Birbaumer N., McFarland D., Pfurtsheller G., Vaughan T., Clinical Neurophysiology 113, 2002, 767-791
  • EEG based communication: prospects and problems, Vaughan T., Wolpaw J.R., Donchin E., IEEE transactions on rehabilitation engineering, vol. 4, no. 4, december 1996, pages 425-430.


Part 6: Links