Leonardo Rufiner

Genetic feature selection to optimally detect P300 in brain computer interfaces

A Brain Computer Interface is a system that provides an artificial communication between the human brain and the external world. The paradigm based on event related evoked potentials is used in this work. Our main goal was to efficiently solve a binary classification problem: presence or absence of P300 in the registers. Genetic Algorithms and Support Vector Machines were used in a wrapper configuration for feature selection and classification. The original input patterns were provided by two channels (Oz and Fz) of resampled EEG registers and wavelet coefficients. To evaluate the performance of the system, accuracy, sensibility and specificity were calculated. The wrapped wavelet patterns show a better performance than the temporal ones. The results were similar for patterns from channel Oz and Fz, together or separated.

Evaluation of LDA Ensembles Classifiers for Brain Computer Interface

The Brain Computer Interface (BCI) translates brain activity into computer commands. To increase the performance of the BCI, to decode the user intentions it is necessary to get better the feature extraction and classification techniques. In this article the performance of a three linear discriminant analysis (LDA) classifiers ensemble is studied. The system based on ensemble can theoretically achieved better classification results than the individual counterpart, regarding individual classifier generation algorithm and the procedures for combine their outputs. Classic algorithms based on ensembles such as bagging and boosting are discussed here. For the application on BCI, it was concluded that the generated results using ER and AUC as performance index do not give enough information to establish which configuration is better.

Genetic Feature Selection for a P300 Brain Computer Interface

A Brain Computer Interface (BCI) provides a direct form of communication between a person and the outside world using brain signals, either to increase his/her integration in society or to provide a way to control the environment where he/she lives. BCIs are communication systems based on electroencephalographic (EEG) signals, such as event-related evoked potentials (ERP). P300 is one of there ERP. It is a peak that usually appears in the EEG signals around 300 ms in response to an infrequent stimulus. The BCI based on P300 is usually composed by different blocks: input (data acquisition), feature selection/extraction, classification, output (e.g. control commands) and, eventually, feedback. In this work, a Genetic Algorithm (GA) is proposed as a feature selection method before the classification stage, implemented using Fisher’s Linear Discriminant Analysis (LDA). A dataset of input patterns was generated from a database of EEG recordings of healthy people, in order to train and test the proposed configuration. The addition of the GA as a feature selection method resulted in a significant improvement in classification performance ( p < 0.001 ) and in a reduction of the amount of features needed to reach such performance ( p < 0.001 ). The results of this work suggest that this configuration could be implemented in a portable BCI.

A comparison of feature extraction strategies using wavelet dictionaries and feature selection methods for single trial P300-based BCI

AbstractThe P300 component of event-related potentials (ERPs) is widely used in the implementation of brain computer interfaces (BCI). In this context, one of the main issues to solve is the binary classification problem that entails differentiating between electroencephalographic (EEG) signals with and without P300. Given the particularly unfavorable signal-to-noise ratio (SNR) in the single-trial detection scenario, this is a challenging problem in the pattern recognition field. To the best of our knowledge, there are no previous experimental studies comparing feature extraction and selection methods for single trial P300-based BCIs using unified criteria and data. In order to improve the performance and robustness of single-trial classifiers, we analyzed and compared different alternatives for the feature generation and feature selection blocks. We evaluated different orthogonal decompositions based on the wavelet transform for feature extraction, as well as different filter, wrapper, and embedded alternatives for feature selection. Accuracies over 75% were obtained for most of the analyzed strategies with a relatively low computational cost, making them attractive for a practical BCI implementation using inexpensive hardware. Graphical AbstractExperiments performed for P300 detection