[ARTICLE] Classification of EEG signals for wrist and grip movements using echo state network – Full Text


Brain-Computer Interface (BCI) is a multi-disciplinary emerging technology being used in medical diagnosis and rehabilitation. In this paper, different techniques of classification and feature extraction are applied to analyse and differentiate the wrist and grip flexion and extension for synchronized stimulation using sensory feedback in neuro-rehabilitation of paralyzed persons. We have used an optimized version of Echo State Network (ESN) to identify as well as differentiate the wrist and grip movements. In this work, the classification accuracy obtained is greater than 96% in a single trial and 93% in discrimination of four movements in real and imagination.


The popularity of analysing brain rhythms and its applications in healthcare is evident in rehabilitation engineering. Motor disabilities as a consequence of stroke require rehabilitation process to regain the motor learning and retrieval. The classification of EEG signals obtained by using a low cost Brain Computer Interface (BCI) for wrist and grip movements is used for recovery. Using Movement Related Cortical Potential (MRCP) associated with imaginary movement as detected by the BCI, an external device can be synchronized to provide sensory feedback from electrical stimulation [1]. The timely detection, classification of movement and the real time triggering of the electrical stimulation as a function of brain activity is desirable for neuro-rehabilitation [2,3]. Thus, BCI has an active role in helping out the paralyzed persons who are not able to move their hand or leg [4]. Using BCI system, EEG data is recorded and processed. The acquired data should have the least component of environmental noise and artifacts for effective classification [5]. EEG signals acquired from the invasive method are found to exhibit least noise components and higher amplitude. However, in most applications, a non-invasive method is preferred. The human brain contains a number of neuron networks. EEG provides a measurement of brain activity as voltage fluctuations which are recorded as a result of ionic current within neurons present inside the brain [6]. Many people have motor disabilities due to the nerve system breakdown or accidental failure of nerve system. There are different methods to resolve this problem, e.g. neuro-prosthetics (neural prosthetics) and BCI [3,79]. In neuro-prosthetics, a solution of the problem is in the form of connecting brain nerve system with the device and in BCI connecting brain nerve system with computer [2]. BCI produce a communication between brain and computer via EEG, ECOG or MEG signals. These signals contain information of any of our body activity [10]. Moreover, in addition to neuro-rehabilitation, assistive robotics and brain control mobile robots also utilizes similar technologies as reported recently [11,12]. The signal processing of these low amplitude and noisy EEG signals require special care during data acquisition and filtering. After recording EEG measurements, these signals are processed via filtration, feature extraction, and classification. Simple first or second order Chebyshev or Butterworth filter can be used as a low pass, high pass or a notch filter. Some features can be extracted by using one of the techniques from time analysis, frequency analysis, time-frequency analysis or time-space-frequency analysis [13,14]. Extracted EEG signal further classify by using one of the techniques like LDA, QDA, SVM, KNN etc. [15,16].

We aim to classify the wrist and grip movements using EEG signals. This research will be helpful for convalescence of persons having disabilities in wrist or grip. Our work is based on offline data-sets, in which the EEG data is collected multiple times from 4 subjects. We present the following major contributions in this paper: First, the differentiation between the wrist and grip movements has been performed by using imaginary data as well as the real movements. Secondly, we have tested multiple algorithms for feature extraction and classification and used ESN with optimized parameters for best results. This paper is organized as follows: section 2 describes a low-cost BCI setup for EEG, section 3 deals with the DAQ protocol, section 4 explains the echo state network and its optimization while section 5 discusses results obtained in this research. Section 6 concludes the paper.

Brain Computer Interface Design

Brain-Computer Interface (BCI) design requires a multi-disciplinary approach for engineers to observe EEG data. Today, a number of sensing platforms are available which provide a low-cost solution for high-resolution data acquisition. Developing a BCI interface requires a two-step approach namely the acquisition and the real-time processing. In off-line processing, the only requirement is to do the acquisition. The data is acquired via a wireless network from the pick-off electrodes arranged on the scalp of the subjects [17]. One such available system is Emotiv, which is easy to install and use. Emotiv headset with 14 electrodes and 2 reference electrodes, CMD and DRL, is used to collect data as shown in Figure 1. All electrodes have potential with respect to the reference electrode. Emotiv headset is a non-invasive device to collect the EEG data as preferred in most of the diagnosis and rehabilitation applications [18].


Figure 1. Emotiv EEG acquisition using P-300 standard.

It is important to understand the EEG signal format and frequency content for pre-processing and offline classification. Table 1 shows some of the indications of physical movements and mind actions associated with different brain rhythms in somewhat overlapping frequency bands. It is obvious that the motor imagery tasks are associated with the μ-rhythm in 8-13 Hz frequency band [19].

Rhythm Frequency
Indication Diagnosis
Δ 0-4 Deep sleep stage Hypoglycaemia, Epilepsy
υ 4-7 Initial sleep stage
α 8-12 Closure of eyes Migraine, Dementia
β 12-30 Busy/Anxious thinking Encephalopathies, Tonic seizures
γ 30-100 Cognitive/motor function
µ 8-13 Motor imagery tasks Autism Spectrum Disorder

Table 1. Brain frequency bands and their significance.



Continue —> Classification of EEG signals for wrist and grip movements using echo state network


, , , , , , , , ,

  1. Leave a comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: