EEG-BCIs have been well studied in the past decades and implemented into several famous applications, like P300 speller and wheelchair controller. However, these interfaces are indirect due to low spatial resolution of EEG. Recently, direct ECoG-BCIs attract intensive attention because ECoG provides a higher spatial resolution and signal quality. This makes possible localization of the source of neural signals with respect to certain brain functions. In this article, we present a realization of ECoG-BCIs for finger flexion prediction provided by BCI competition IV. Methods for finger flexion prediction including feature extraction and selection are provided in this article. Results show that the predicted finger movement is highly correlated with the true movement when we use band-specific amplitude modulation.
Deep Dive into Decoding Finger Flexion using amplitude modulation from band-specific ECoG.
EEG-BCIs have been well studied in the past decades and implemented into several famous applications, like P300 speller and wheelchair controller. However, these interfaces are indirect due to low spatial resolution of EEG. Recently, direct ECoG-BCIs attract intensive attention because ECoG provides a higher spatial resolution and signal quality. This makes possible localization of the source of neural signals with respect to certain brain functions. In this article, we present a realization of ECoG-BCIs for finger flexion prediction provided by BCI competition IV. Methods for finger flexion prediction including feature extraction and selection are provided in this article. Results show that the predicted finger movement is highly correlated with the true movement when we use band-specific amplitude modulation.
A brain-computer interface (BCI) is a new communication and control option for those with severe motor disabilities [1]. A BCI system translates brain signals into commands for a computer or other devices. Electroencephalography (EEG) based BCIs are the most studied non-invasive interfaces mainly due to the fine temporal resolution, ease of use, portability and low set-up cost of EEG recordings [2]. Unfortunately, non-invasive implants produce noisy signals because of the damping effect introduced by the skull. Further, its spatial resolution is low due to the size of the electrode. Thus, it is difficult to use EEG signals for directly decoding the task events.
Electrocorticography (ECoG) has recently emerged as a promising recording technique for use in brain-computer interfaces (BCI) and has been extensively investigated [3,4,5,6]. ECoG electrodes arrays were initially implanted (under the skull but over the surface of the cortex) for severe epileptic patients for presurgical planning, in order to identify the sources generating epileptic seizures [7]. The spatial resolution of ECoG signals is higher than EEG with a better signal-to-noise ratio (SNR). Typically, the diameter of one ECoG electrode is of 4 mm with 1 cm inter-electrode distance. Therefore ECoG can provide a spatial resolution of approximately 1 cm [7]. Spatial resolution plays an important role in BCI [4]. The fine spatial resolution of ECoG provides an opportunity for decoding brain functions directly and therefore possibly leads to the implementation of direct neural interfaces, which is difficult to be accomplished through EEG-based BCIs.
In order to study the application of ECoG in BCIs, several research groups have recorded ECoG signals when the participants performed certain kind of tasks related to the brain functional areas that the implanted electrode arrays had covered. The testing tasks include center-out reaching or pointing task [4], finger flexion [3] and cursor trajectory [6]. Some of these groups are interested in the use of frequency features as the inputs of decoder [6], and others use amplitude modulation [4], or both [3]. For features extracted from frequency domain, it was found that different frequency bands have different correlation coefficients to the event tasks [3,5]. Further, the implanted ECoG electrode array is usually a grid with a size from 66 to 88. It covers a cortical zone involving several different functions of the brain. Thus, frequency band and electrode selection are necessary for identifying good descriptors.
In this paper, we first present the material used, the ECoG data set adopted from BCI competition IV in section 2. The methods for feature extraction and selection are presented in 3.1. The construction of a linear regression model for decoding is done in 3.2. Simulation results on this competition data set are reported in section 4. We draw conclusions with discussions on further work in section 5.
Recently several paradigms of ECoG-BCIs have been realized focusing on motor tasks. Here, we just adopt the ECoG-BCI data set provided by BCI competition IV [3,8]. Three subjects participated in the recording were epileptic patients. Each subject was implanted a subdural electrode array with the arrangement of 86 or 88 electrodes. The ECoG signals were recorded when the subjects performed a finger movement task. The subjects were instructed to move one certain finger by the corresponding word displayed on a computer screen. The execution of finger movement lasted 2 seconds and it was followed by a 2-second resting period. There were 30 movement stimulus for each finger resulting in 600-second recording for each subject. The ECoG signals were recorded through the general-purpose BCI system BCI2000 [9], bandpass filtered between 0.15 to 200 Hz and sampled at 1000 Hz. The finger movements were recorded using a dataglove sampled at 25 Hz. Figure 1 shows an example of the visualization of the ECoG signals and the finger movement time course from subject 1. Due to space limitation, only a subset of electrodes is displayed.
The evidence of sensorimotor ECoG dynamics has been reported in several specific frequency bands including sub-bands (1-60 Hz), gamma band (60-100 Hz), fast gamma band (100-300 Hz) and ensemble depolarization (300-6k Hz) [4]. Being inspired by the firing rate coding scheme used in spike train decoding, Sanchez proposed the band-specific amplitude modulation (AM) as the descriptor for ECoG signal decoding, which is defined as the sum of the power of the voltage of the band-specific ECoG signals in a time bin ∆t:
where t n+1 = t n + ∆t. We simply let ∆t = 40ms so the resulting bandspecific AM feature inputs have the same sampling rate (i.e. 25Hz) as that of the dataglove position measurements. It seems that the sensitivity profile of ECoG for each frequency band is specific to a given task [4]. For our study, we also found that each finger movement was correlated to two or three s
…(Full text truncated)…
This content is AI-processed based on ArXiv data.