Classifying Event-Related Desynchronization in EEG, ECoG and MEG Signals

Abstract

We present the results from three motor-imagery-based Brain-Computer Interface experiments. Brain signals were recorded from 8 untrained subjects using EEG, 4 using ECoG and 10 using MEG. In all cases, we aim to develop a system that could be used for fast, reliable preliminary screening in the clinical application of a BCI, so we aim to obtain the best possible classification performance in a short time. Accordingly, the burden of adaptation is on the side of the computer rather than the user, so we must adopt a machine-learning approach to the analysis. We introduce the required machine-learning vocabulary and concepts, and then present quantitative results that focus on two main issues. The first is the effect of the number of trials-how long does the recording session need to be? We find that good performance could be achieved, on average, after the first 200 trials in EEG, 75-100 trials in MEG, or 25-50 trials in ECoG. The second issue is the effect of spatial filtering-we compare the performance of the original sensor signals with that of the outputs of Independent Component Analysis and the Common Spatial Pattern algorithm, in each of the three sensor types. We find that spatial filtering does not help in MEG, helps a little in ECoG, and improves performance a great deal in EEG. The unsupervised ICA algorithm performed at least as well as the supervised CSP algorithm in all cases-the latter suffered from poor generalization performance due to overfitting in ECoG and MEG, although this could be alleviated by reducing the number of sensors used as input to the algorithm.

Model based generalization analysis of common spatial pattern in brain computer interfaces

“…For overfitting, it is reported that small training set has poor generalization [16]. The overfitting effect is worse when there are a larger number of channels related to the number of trials [15]. We have no idea about how the other factors can influence the generalization.…”

“…With the development of CSP algorithm, the issue of generalization has been gradually noticed [15,16]. The generalization of CSP refers to how far the common spatial filter performs on the testing data from the training data.…”

“…In Ref. [15], Hill et al compared the performance of ICA and CSP algorithm in EEG, ECoG (electrocorticography) and MEG (magnetoencephalography) signals, and found that spatial filtering does not help in MEG, helps a little in ECoG, and improves performance a great deal in EEG. It was found that the supervised CSP algorithm suffers from poor generalization performance due to overfitting in ECoG and MEG.…”

Handcrafted Electrocorticography Electrodes for a Rodent Behavioral Model

“…Furthermore, identifying neural functions by means of cortical maps can ease the establishment of proper links for neural interfaces that can offer disabled patients an alternative solution for their lost sensory and motor functions through the use of brain-computer interface (BCI) technology [6][7][8][9]. Several recording methods such as electroencephalography (EEG), electrocorticography (ECoG), magnetoencephalography (MEG), positron emission tomography (PET), and functional magnetic resonance imaging (fMRI) have proven to be potentially useful in the implementation of BCI [10][11][12][13][14]. However, due to not having considerable temporal resolution and being large systems, MEG, fMRI, and PET are not practical for continuous recording of electrical signals from neurons.…”

“…This method measures the activity from a large population of neurons by attaching electrodes to the scalp of a subject. As a consequence, the impedance of the distance between the cortical tissue and the electrodes force the EEG to have a low spatial resolution [10,15]. Another type of recording technology is the Utah Intracortical Electrode Array (UIEA) which has allowed implanting a large number of microelectrodes into a small area of the cortex due to its micromachined structure.…”