The movement-related cortical potential (MRCP) is a low-frequency negative shift in the electroencephalography (EEG) recording that takes place about 2 seconds prior to voluntary movement production. MRCP replicates the cortical processes employed in planning and preparation of movement. In this study, we recapitulate the features such as signal’s acquisition, processing, and enhancement and different electrode montages used for EEG data recoding from different studies that used MRCPs to predict the upcoming real or imaginary movement. An authentic identification of human movement intention, accompanying the knowledge of the limb engaged in the performance and its direction of movement, has a potential implication in the control of external devices. This information could be helpful in development of a proficient patient-driven rehabilitation tool based on brain-computer interfaces (BCIs). Such a BCI paradigm with shorter response time appears more natural to the amputees and can also induce plasticity in brain. Along with different training schedules, this can lead to restoration of motor control in stroke patients.
The idea of predicting the motor tasks was initially presented by Helmholtz in 1867. Later on, in the fifties Sperry and Von Holst expressed that motor commands make an internal replica which uncovers the anticipated movement and its subsequent sensations [
The brain’s current motor activity can be understood in real time through EEG, which can be further employed for prediction of the next voluntary motor task. Real-time EEG might present novel nonmuscular control channel Brain Computer Interfaces (BCIs) for delivering messages and commands to the external world [
Studies have shown that EEG comprises enough real-time information to be utilized for different purposes/tasks such as internet browsing, controlling environment (e.g., light, television, and temperature), word processing, controlling a two-dimensional cursor movement on screen, or even operating neuroprosthesis [
The concept of “premovement” or “before the movement” indicates the time when no muscle movement is evident or is unrelated if it occurs, but the subject is fully familiar with the action he is going to perform in the near future. This is also referred to as planning/preparation of the movements. In this time interval (i.e., 0.5–2 s prior to the movement onset), the cortex is adapted for implementation of action [
This paper aims to review the different studies which have used movement-related cortical potentials (MRCPs) to predict the upcoming movements. In the next section, we illustrate the key modifications in the EEG data reported prior to the voluntary movement and how the knowledge of these variations can be used to extract information about the forthcoming movement. In each case, we discuss the main foundations of the study and evaluate the EEG setup and protocols. Finally, in the Conclusion, we recapitulate the key ideas with the hope to bring more consideration to the affluence of premovement and premotor imagery EEG.
In this section, we summarize the reported changes in EEG prior to the onset of the actual or imagery movement. All the following phenomena have been delineated both when the movement is imagined and when it is actually executed. One or an amalgamation of these progressions is the fundamental spotlight of the studies acquiring features from premotor imagery or premovement period, discussed in Section
The implementation of a motor task in humans measured over the primary motor cortex is preceded by a slow decrease in the EEG amplitude (within at least 500 ms) and this potential is known as an MRCP [
MRCPs of a healthy subject for real and imaginary right ankle dorsiflexion. Each wave is an average of 50 large Laplacian spatial filtered EEG trials recorded from sites F3, Fz, F4, C3, Cz, C4, P3, Pz, and P4. Time 0 s is defined as the movement onset. BP1 is early BP, BP2 is late BP, MP is motor potential, and MMP is movement-monitoring potential. For more information on experiment protocol, see [
BP or RP is a negative cortical potential which starts to grow around 1.5 to 1 s prior to the onset of a voluntary movement [
CNV is a slow negative wave that originates in the interval (1–1.5 s) between a “Warning” and a “Go” stimulus [
Several studies reported that the BP might be recorded from subcortical structures such as basal ganglia and thalamus [
In order to elucidate the exact area and timing of the motor cortical activation in voluntary movement, dipole source analysis incorporating multiple constraints was applied for MRCP. The work [
To a certain degree, automatic movements such as blinking of eyelids, spontaneous eye movements, swallowing, chewing, and respiration are also controlled by volitional factors; therefore, BP is recorded when these movements are reiterated at a self-paced rate [
The dorsal premotor cortex (PMd) is believed to play substantial role in cued movement preparation rather than in self-initiated movements [
Comparing the MRCP for a foot movement with hand movement showed interesting differences across some movement components [
For the study of BP in individual subjects against hand movements, it is vital to record EEG from multiple electrodes, including C1 and C2, for identifying the abrupt increase of the gradient [
The MRCP can easily be masked by activity in the higher frequency bands because its amplitude typically lies between 5 and 30
Components of MRCP can be inspired by various factors such as preparatory state, level of intention, movement selection, pace of movement repetition, speed and precision of movement, praxis movement, perceived effort, force exerted, discreteness and complexity of movement, learning and skill acquisition, and pathological injuries of various brain structures. The review by [
In this section, we extracted information of the premovement or preimageries from different studies. Some studies display the usefulness of data obtained in the real-time BCIs. Studies utilized different EEG data acquisition techniques including different electrode montages and signal enhancement methods, since these are related to the results reported. Studies mentioned in Tables
Experiment protocols of studies reviewed.
Reference | Number of subjects | Number of electrodes | Movement type | Self-paced or cue-based | Brain signals |
---|---|---|---|---|---|
(Yom-Tov and Inbar, 2003) [ |
5 (healthy) | 9, 4 out of 9 channels were used | Executed finger movement (button press) | Self-paced | MRPs |
|
|||||
(Haw et al., 2006) [ |
5 (not mentioned) | 1 | Executed finger movements | Cue-based | BP |
|
|||||
(Bai et al., 2007) [ |
12 (healthy) | 122 | Executed hand movement | Self-paced | MRCPs and ERD (event-related desynchronization) |
|
|||||
(Boye et al., 2008) [ |
1 (not mentioned) | 9 | Executed and imagined foot movement (isometric plantar-flexion), but only imaginary task was further analyzed | Cue-based | MRCP |
|
|||||
(Kato et al., 2011) [ |
7 (not mentioned) | 1 | Executed and imagined finger movements (button press) | Cue-based | CNV |
|
|||||
(Niazi et al., 2011) [ |
19 (healthy) and 5 (stroke patients) | 10 | Executed and imagined foot movement (ankle dorsiflexion) | Self-paced | BP |
|
|||||
(Lew et al., 2012) [ |
8 (healthy), 2 (control), and 2 (stroke patients) | 64, 34 out of 64 channels were used | Executed arm movements (reaching task) | Self-paced | BP |
|
|||||
(Niazi et al., 2012) [ |
16 (healthy) | 10 | Imagined foot movements (dorsiflexion) | Self-paced | MRCP |
|
|||||
(Niazi et al., 2013) [ |
20 (healthy) and 5 (stroke patients) | 10 | Executed and imagined foot movements (dorsiflexion) | Self-paced | MRCP |
|
|||||
(Ahmadian et al., 2013) [ |
3 (healthy) | 128 channels | Finger movement (button press) | Self-paced | BP |
|
|||||
(Jochumsen et al., 2013) [ |
12 (healthy) | 10 | Executed foot movement (isometric dorsiflexion) | Cue-based | MRCP |
|
|||||
(Jiang et al., 2015) [ |
9 (healthy) | 9 | Executed foot movements (stepping) | Self-paced | MRCP |
|
|||||
(Xu et al., 2014) [ |
9 (healthy) | 9 | Executed and imagery foot movements (dorsiflexion) | Self-paced | MRCP |
Techniques used for prediction of onset of movement and main findings of the studies reviewed.
Reference | Preprocessing techniques | Classifiers | Performance | Latency (ms) | Offline or online system | Single-trial analysis | Limitations |
---|---|---|---|---|---|---|---|
(Yom-Tov and Inbar, 2003) [ |
Low-pass filter (10 Hz) using 8th-order Chebyshev | Simple threshold element, support vector machine (SVM), and linear vector quantiser 3-feature reduction with 1-nearest neighbor (1-NN) | Using hybrid detector 25% improvement in performance was achieved as compared to Mason-Birch low frequency asynchronous detector (LFASD) | 25 decisions s−1 | Offline | — | Detector fails to work correctly partly due to MRPs related to other limbs and imagined movements |
|
|||||||
(Haw et al., 2006) [ |
Building a specific template during 3 or 4 training sessions for each subject | Thresholding based on correlation and error | Accuracy was 70% with a false positive rate (FPR) of (5/24) | — | — | Yes | Variability in performance between users |
|
|||||||
(Bai et al., 2007) [ |
Low pass filter (100 Hz) using 3rd-order Butterworth filter | Linear Mahalanobis Distance (MD), Quadratic MD, Bayesian Classifier (BC), Multilayer Perceptron (MLP) Neural Network, Probabilistic Neural Networks, and SVM | Accuracy was 75% | — | Offline | Yes | Large number of electrodes (122) |
|
|||||||
(Boye et al., 2008) [ |
Downsampling from 500 Hz to 20 Hz, with antialiasing prefiltering (0–5 Hz) and PCA and Locality Preserving Projection (LPP) | A variation of |
Sensitivity for SVM = 96.3 ± 2.0% for |
— | — | Yes | Method was tested on segmented data rather than ongoing EEG traces with only 1 subject |
|
|||||||
(Kato et al., 2011) [ |
Low pass filter (35 Hz) and high pass filter (0.05 Hz) for EEG and 0.1 Hz for EOG | SVM | Detection rate (intention to switch = 99.3% and (not to switch = 2.1%) | — | Both | Yes | Online system cannot differentiate between intend to switch and do not intend to switch |
|
|||||||
(Niazi et al., 2011) [ |
Band pass filter (0.05–10 Hz) with Optimized Spatial Filter (OSF) | Neyman Pearson Lemma | For healthy subject’s movement execution TPR = 82.5 ± 7.81% and for movement imagination TPR = 64.5 ± 5.33% | −66.6 ± 121 | Offline | Yes | Small sample size (patients) and no online detection due to instrumentational limitation |
For stroke patients TPR = 55.01 ± 12.01% | −56.8 ± 139 | ||||||
|
|||||||
(Lew et al., 2012) [ |
Narrow band zero phase noncausal IIR filter with cutoff frequencies of 0.1 and 1 Hz | Linear Discriminant Analysis (LDA) | TPR = 76 ± 7% (healthy) | −167 ± 68 (healthy) | Offline | Yes | Large number of electrodes (34) |
For stroke and control subjects TPR = 81 ± 11% (left hand) versus (right hand) TPR = 79 ± 12% | Right hand = −140 ± 92 versus left hand = −162 ± 105 | ||||||
|
|||||||
(Niazi et al., 2012) [ |
Band pass filter (0.1–100 Hz) and OSF | Matched Filter | TPR = 67.15 ± 7.87% and FPR = 22.05 ± 9.07% | −125 ± 309 (offline) | Online | — | Different aspects of triggered stimulations were not fully considered |
|
|||||||
(Niazi et al., 2013) [ |
Band pass filter (0.05–10 Hz) and OSF to maximize SNR | Matched Filter | For motor execution (healthy) TPR = 69 ± 21% and FPR = 2.8 ± 1.7 | −196 ± 162 | Offline | Yes | — |
For stroke patients TPR = 58 ± 11% and FPR = 4.1 ± 3.9 | 152 ± 239 | ||||||
For motor imagery (healthy) TPR = 65 ± 22% and FPR = 4.0 ± 1.7 | — | ||||||
|
|||||||
(Ahmadian et al., 2013) [ |
Filtering data between 0.1 Hz and 70 Hz | Independent component analysis (ICA) | Computation time for constraint blind source extraction (CBSE) algorithm was 0.26 s and blind source separation (BSS) algorithm took 51.90 s | 260 | — | Yes | Large number of electrodes (128) with small number of subjects |
|
|||||||
(Jochumsen et al., 2013) [ |
Band-pass filter (0.05–10 Hz) using 2nd-order Butterworth in forward and reverse direction with three spatial filters, large Laplacian spatial filter (LLSF), OSF, and common spatial patterns (CSP) | SVM | TPR = ~80% and FPR <1.5 accuracy = 80 ± 10% (speed) and 75 ± 9% (force) | 317 ± 73 | Offline | Yes | Inclusion of only healthy subjects |
|
|||||||
(Jiang et al., 2015) [ |
ICA followed by LSF to enhance SNR | ICA | TPR = 76.9 ± 8.97% and FPR = 2.93 ± 1.09 per minute | −180 ± 354 | Offline | Yes | Prediction of gait initiation was not done |
|
|||||||
(Xu et al., 2014) [ |
Band-pass filter (0.05–3 Hz) and large LSF to enhance SNR | LPP followed by LDA | LPP-LDA TPR = 79 ± 12% FPR = 1.4 ± 0.8 per minute | 315 ± 165 | Online | — | Inclusion of only healthy subjects and classifier did not work for training trials less than 15 |
This section briefly describes the classification algorithms used in studies mentioned in this paper. The classification algorithms include Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Neural Networks (NN), Multilayer Perceptron (MLP), Bayesian Classifier (BC),
The Support Vector Machine (SVM) is a pattern recognition algorithm that has been successfully applied to wide variety of classification problems. It learns to distinguish among various classes of objects by some complex data transformations and then separate the data based on the defined labels for classes. For example, the data for a two-class problem consist of objects labeled corresponding to two classes, for example, +1 (data belong to class 1) or −1 (data belong to class two). The system then automatically identifies the input points and uses them to represent the solution [
The purpose of Linear Discriminant Analysis (LDA) (also called Fisher’s LDA) is to use hyperplanes to isolate the data into different classes [
Neural Networks (NN) can be thought as circuits of immensely interconnected units with flexible interconnection weights, which allow us to yield nonlinear decision boundaries. They can be classified by architecture, algorithm for calibrating the weights, and the kind of units utilized as a part of the circuit [
Multilayer Perceptron (MLP) is composed of several layers of neurons: an input layer, perhaps one or many hidden layers, and an output layer [
Bayesian Classifier (BC) depends on Bayes’ theorem and can anticipate class membership probabilities, for example, the likelihood that a given sample fits into a specific class. In order to classify a feature vector, it learns the way of computing the probability of each class. BC assumes that estimation of a specific feature does not rely on value of any other feature, which provided the class variable. Being a generative classifier, it produces nonlinear decision boundaries and performs more efficient rejection of uncertain samples as compared to discriminative classifiers [
The objective of this method is to allocate to an unseen point the dominant class amongst its
Mahalanobis Distance assigns a feature vector to a class according to its nearest neighbor(s) from a class prototype. It assumes a Gaussian distribution
A spatial filter amalgamates data from two or more locations (electrodes). Spatial filtering techniques comprise common spatial patterns (CSP), common average referencing (CAR), surface Laplacian (SL), independent component analysis (ICA), and principle component analysis (PCA). This section briefly describes some spatial filters; for more details please refer to [
ICA is a method intended to find a linear illustration of non-Gaussian data in the form of statistically independent constituent components [
The performance of studies is computed using sensitivity, specificity, and detection error. Sensitivity (also known as true positive rate (TPR)) quantifies the fraction of actual positives (movements) which are precisely recognized. Specificity (also called the true negative rate (TNR)) assesses the fraction of negatives (no motion or noise) which are exactly detected. Sensitivity and specificity are calculated using the following equations, respectively, where TP and TN represent number of true positives and number of true negatives, respectively [
For the development of self-paced closed loop BCIs, the robust detection of motor intention is a vital and critical issue. In the past decades, sensory motor rhythms have been used for detection of motor intention in studies comprising BCIs to control visual feedback [
In recent years, slow cortical potentials captured the attention in the rehabilitation field. Several studies have been reported, which concentrated their attention intended for communication purposes.
Yom-Tov and Inbar [
To detect movement planning, [
From single trial EEG, [
The validity of OSF on imagination of isometric plantar-flexion was confirmed in a study conducted by [
Kato et al. [
The detection of movement intention from single trial MRCPs of movement imagination and movement execution was performed by [
Lew et al. [
Ahmadian et al. [
Motor intention could be detected from MRCP using the Matched Filter, with small latency and satisfactory accuracy. The same task was performed in [
Jochumsen et al. [
Xu et al. [
To further improve the results, [
In a recent study by [
A complexity confronted in this paper involves the absence of similar studies in terms of purpose of detector, signal acquisition, limb movement, and number of electrodes. It should be noted that movements executed in different studies were not similar leading to variances in signal morphology and SNR. Ideally studies should be compared within the same context, that is, with similar protocol, users, and similar extraction of features.
Although the mentioned studies have delivered a valuable insight into the prediction of MRCPs using different signal acquisition techniques, the framework of research is not without its impediments. One limitation relevant to most of the studies mentioned is the absence of clear ecological validity in the research, that is, “the extent to which an experimental situation mimics a real world situation” [
EEG signals are notably nonstationary so training sets acquired from different sessions are probable to be quite different. Consequently, a low variance (sensitivity to training set) can be a solution to tackle with the variability issue in some studies. Unstable classifiers tend to have a low bias (deviation between the estimated mapping and the superlative mapping) and a high variance, while stable classifiers have a high bias and a low variance [
The classifier will probably give bad performance if the number of training data is lesser matched to the size of the feature vectors. Usage of at least five to ten times training samples per class as the dimensionality is recommended [
Furthermore, combinations of classifiers also seem to be very efficient in some studies [
MRCP has been employed as a control signal in BCI technology. It is mainly beneficial for neuromodulation applications in which the delay between the intention of action and the feedback from the system is crucial to induce plasticity [
EEG data collected prior to imminent movement which associates with motor preparation and planning period of the brain present substantial prediction potentials. Illustrating the intention to move through MRCP can be employed in rehabilitation protocols. Depending on the purpose of the BCI system, a higher TPR could be achieved increasing the number of false positives, while some studies tend to give a priority to a low FPR rather than high TPRs [
The authors declare that there is no conflict of interests regarding the publication of this paper.