Matthew Dyson

An analysis of performance evaluation for motor-imagery based BCI.

In recent years, numerous brain-computer interfaces (BCIs) based on motor-imagery have been proposed which incorporate features such as adaptive classification, error detection and correction, fusion with auxiliary signals and shared control capabilities. Due to the added complexity of such algorithms, the evaluation strategy and metrics used for analysis must be carefully chosen to accurately represent the performance of the BCI. In this article, metrics are reviewed and contrasted using both simulated examples and experimental data. Furthermore, a review of the recent literature is presented to determine how BCIs are evaluated, in particular, focusing on the relationship between how the data are used relative to the BCI subcomponent under investigation. From the analysis performed in this study, valuable guidelines are presented regarding the choice of metrics and evaluation strategy dependent upon any chosen BCI paradigm.

A Novel Design of 4-Class BCI Using Two Binary Classifiers and Parallel Mental Tasks

A novel 4-class single-trial brain computer interface (BCI) based on two (rather than four or more) binary linear discriminant analysis (LDA) classifiers is proposed, which is called a “parallel BCI.” Unlike other BCIs where mental tasks are executed and classified in a serial way one after another, the parallel BCI uses properly designed parallel mental tasks that are executed on both sides of the subject body simultaneously, which is the main novelty of the BCI paradigm used in our experiments. Each of the two binary classifiers only classifies the mental tasks executed on one side of the subject body, and the results of the two binary classifiers are combined to give the result of the 4-class BCI. Data was recorded in experiments with both real movement and motor imagery in 3 able-bodied subjects. Artifacts were not detected or removed. Offline analysis has shown that, in some subjects, the parallel BCI can generate a higher accuracy than a conventional 4-class BCI, although both of them have used the same feature selection and classification algorithms.

2009 Special Issue: Adaptive classification for Brain Computer Interface systems using Sequential Monte Carlo sampling

Adaptive classification is a key function of Brain Computer Interfacing (BCI) systems. This paper proposes robust mathematical frameworks and their implementation for the on-line sequential classification of EEG signals in BCI systems. The proposed algorithms are extensions to the basic method of Andrieu et al. [Andrieu, C., de Freitas, N., and Doucet, A. (2001). Sequential bayesian semi-parametric binary classification. In Proc. NIPS], modified to be suitable for BCI use. We focus on the inference and prediction of target labels under a non-linear and non-Gaussian model. In this paper we introduce two new algorithms to handle missing or erroneous labeling in BCI data. One algorithm introduces auxiliary labels to process the uncertainty of the labels and the other modifies the optimal proposal functions to allow for uncertain labels. Although we focus on BCI problems in this paper, the algorithms can be generalized and applied to other application domains in which sequential missing labels are to be imputed under the presence of uncertainty.

A 3-class Asynchronous BCI Controlling A Simulated Mobile Robot

We present our design and online experiments of a 3-class asynchronous BCI controlling a simulated robot in an indoor environment. Two characteristics of our design have efficiently decreased the false positive rate during the NC (no control) mode. First, three one-vs-rest LDA classifiers are combined to control the switching between NC and IC (in control) mode. Second, the hierarchical structure of our controller allows the most reliable class (mental task) in a specific subject to play a dominant role in the robot control. A group of simple rules triggered by local sensor signals are designed for safety and obstacle avoidance in the NC mode. In online experiments, subjects successfully controlled the robot to circumnavigate obstacles and reach small targets in separate rooms.

On the genetic programming of time-series predictors for supply chain management

Single and multi-step time-series predictors were evolved for forecasting minimum bidding prices in a simulated supply chain management scenario. Evolved programs were allowed to use primitives that facilitate the statistical analysis of historical data. An investigation of the relationships between the use of such primitives and the induction of both accurate and predictive solutions was made, with the statistics calculated based on three input data transformation methods: integral, differential, and rational. Results are presented showing which features work best for both single-step and multi-step predictions.

Localisation of cognitive tasks used in EEG-based BCIs

OBJECTIVE To provide candidate electrode sites and neurophysiological reference information for cognitive tasks used in brain-computer interfacing research. METHODS Six cognitive tasks were tested against the idle state. Data representing the idle state were collected with active cognitive task data during each recording session. Cross subject candidate electrode sites were obtained via a wrapper method based upon a sequential forward floating search algorithm. Source localisation results were obtained using sLORETA software. RESULTS Spatial feature distributions and localisation results are presented. Primary centres of activity for motor imagery tasks are localised to the pre- and postcentral gyrus. Auditory-based tasks show activity in the middle temporal gyrus. Calculation activity was localised to the left inferior frontal gyrus and right supramarginal gyrus. Navigation imagery produced activity in the precuneus and anterior cingulate cortex. CONCLUSIONS Spatial areas of activation suggest that arithmetic and auditory tasks show promise for pairwise discrimination based on single recording sites. sLORETA significance levels suggest that motor imagery tasks will show greatest discrimination from baseline EEG activity. SIGNIFICANCE This is the first study to provide candidate electrode sites for multiple tasks used in brain-computer interfacing.

A Comparison of Mental Task Combinations for Asynchronous EEG-Based BCIs

Aiming at developing asynchronous BCIs, we tested 21 2-class combinations of 7 mental tasks to determine whether any pair of tasks may be more suitable. The tasks under consideration were: auditory recall, mental navigation, sensorimotor attention (left hand), sensorimotor attention (right hand), mental calculation, imaginary movement (left hand), imaginary movement (right hand). Sensorimotor attention is novel in this application domain. All possible pairs were tried in 5 subjects using data from 10s periods in which subjects were free to execute the required mental task at their own pace. Recordings were done whilst the subject controlled a robot navigation simulator on a computer monitor, with the robot direction being related to the mental task. Classification of the data was done using LDA. Class-separation was estimated using the Davies-Bouldin index. Best classification results were obtained when auditory recall was followed or preceded by mental calculation. Of the possible 21 task combinations, this task pair was in the top 5 (performance-wise) for 4 of the 5 subjects. This was also the case when class-separation was used as a criterion.

Sequential Bayesian estimation for adaptive classification

This paper proposes a robust algorithm to adapt a model for EEG signal classification using a modified extended Kalman filter (EKF). By applying Bayesian conjugate priors and marginalising the parameters, we can avoid the needs to estimate the covariances of the observation and hidden state noises. In addition, Laplace approximation is employed in our model to approximate non-Gaussian distributions as Gaussians.

Bayesian inference for an adaptive Ordered Probit model

This paper proposes an algorithm for adaptive, sequential classification in systems with unknown labeling errors, focusing on the biomedical application of Brain Computer Interfacing (BCI). The method is shown to be robust in the presence of label and sensor noise. We focus on the inference and prediction of target labels under a nonlinear and non-Gaussian model. In order to handle missing or erroneous labeling, we model observed labels as a noisy observation of a latent label set with multiple classes (≥ 2). Whilst this paper focuses on the methods application to BCI systems, the algorithm has the potential to be applied to many application domains in which sequential missing labels are to be imputed in the presence of uncertainty. This dynamic classification algorithm combines an Ordered Probit model and an Extended Kalman Filter (EKF). The EKF estimates the parameters of the Ordered Probit model sequentially with time. We test the performance of the classification approach by processing synthetic datasets and real experimental EEG signals with multiple classes (2, 3 and 4 labels) for a Brain Computer Interfacing (BCI) experiment.

Sequential classification of mental tasks vs. idle state for EEG based BCIs

Results are presented from an ongoing investigation testing discrimination rates of six mental tasks against the idle state for brain computer-interfacing. An online sequential classification method is employed, results represent calculated feedback position during trial periods. Current classification rates suggest auditory imagery shows lower discrimination against the idle state. Results mirror previous work in which linear classification accuracy was maximised within a trial window.