Exploring EEG spectral and temporal dynamics underlying a hand grasp movement.

For brain-computer interfaces, resolving the differences between pre-movement and movement requires decoding neural ensemble activity in the motor cortex's functional regions and behavioural patterns. Here, we explored the underlying neural activity and mechanisms concerning a grasped motor task by recording electroencephalography (EEG) signals during the execution of hand movements in healthy subjects. The grasped movement included different tasks; reaching the target, grasping the target, lifting the object upwards, and moving the object in the left or right directions. 163 trials of EEG data were acquired from 30 healthy participants who performed the grasped movement tasks. Rhythmic EEG activity was analysed during the premovement (alert task) condition and compared against grasped movement tasks while the arm was moved towards the left or right directions. The short positive to negative deflection that initiated around -0.5ms as a wave before the onset of movement cue can be used as a potential biomarker to differentiate movement initiation and movement. A rebound increment of 14% of beta oscillations and 26% gamma oscillations in the central regions was observed and could be used to distinguish pre-movement and grasped movement tasks. Comparing movement initiation to grasp showed a decrease of 10% in beta oscillations and 13% in gamma oscillations, and there was a rebound increment 4% beta and 3% gamma from grasp to grasped movement. We also investigated the combination MRCPs and spectral estimates of α, β, and γ oscillations as features for machine learning classifiers that could categorize movement conditions. Support vector machines with 3rd order polynomial kernel yielded 70% accuracy. Pruning the ranked features to 5 leaf nodes reduced the error rate by 16%. For decoding grasped movement and in the context of BCI applications, this study identifies potential biomarkers, including the spatio-temporal characteristics of MRCPs, spectral information, and choice of classifiers for optimally distinguishing initiation and grasped movement.

1. Introduction

Patients with amyotrophic lateral sclerosis (ALS) or spinal cord injury (SCI) are reported to have significant loss of voluntary motor control and extensive dysfunction of upper and lower limbs [1-4]. Rehabilitation using robotic or functional electrical stimulation could facilitate novel therapeutic strategies for such conditions. Integration of signals could also be a potential goal for Brain-Computer Interfaces (BCI), allowing the transfer of control information from a prosthesis onto the brain tissue [5-7].

Neural correlates of voluntary grasped movement are relevant in the development of modern prosthetics and towards Brain-Computer Interface (BCI) research [8-10]. Insights from electroencephalography (EEG) recordings and their underlying neurophysiological processes involved in motor tasks can help decode movement and its functional interpretation. EEG signal components related to movement allow noninvasive measurements, making them suitable for natural BMIs [5, 6]. In addition, these correlates are measurable in paralyzed patients [3, 4, 7, 11]. For example, the assessment of grip forces [12], movement direction, and lifting paradigms [13] have allowed to objectively and quantitatively detect deficits in patients with Multiple Sclerosis (MS). Studies [14-16] have also focused on low-frequency EEG components to understand how movement trajectories [17] and grasp types [18] were encoded.

EEG-based studies have explored movement initiation or intention and execution to be associated with μ/α and β rhythms [19], although the role of higher frequency oscillations in movement execution remains to be critically explored. The modulation of β rhythms and oscillations in the context of sensorimotor state maintenance and motor function has been previously reported [20, 21]. However, there is little information related to cortical zone-dependent microstates and their movement-related cortical potential (MRCP) components and their variability during resting states for reach, grasp, and grasped movement tasks performed using a human arm. Readiness potentials, and time delays to occurrences of motor-related events while simultaneously comparing motor imagery and motor execution spectral components in EEG can extend current knowledge for developing BCIs [10]. Event-related (de)synchronisation ERD/S and movement-related cortical potentials (MRCPs) have been employed to assess and detect movement intention, execution and imagery in several studies [22-26]. Among the neural correlates of voluntary movement intention, studies had reported the role of a component of MRCP, the slow negative potential [27, 28]. MRCPs have been observed as low-frequency components within EEG signals, computed by averaging several trials commencing 2s before voluntary movement [29, 30]. The readiness potential (RP) [31], an MRCP component, was perceived with a larger amplitude at contralateral central regions around 400ms before the movement onset. In the case of conscious voluntary movement, readiness potential was known to precede 500-800ms before the onset of movement [32, 33]. MRCPs serve as reference attributed to cortical excitability [34-36] occurring before the movement. Also, MRCPs have been used as biomarkers of movement intention and movement tasks in ALS [37], stroke [25], and Parkinson’s disease [38, 39] patients.

The neural components correlated to hand-reaching that occurred within the medial parieto-frontal circuit including the medial intraparietal area (mIP) at the boundaries with area V6A and the dorsal pre-motor areas have been explored [40]. Neural activity related to grasping was reported to occur in the lateral parieto-frontal circuit involving the anterior intraparietal area (AIP) primarily and the dorsal (PMd) and the ventral (PMv) regions of pre-motor areas [41-43]. A significant role was known to be identified in the human AIP (hAIP) during grasping tasks [44-49]. It was demonstrated [50] that grasp-related neural activity was observed in the ventral premotor F5, parietal AIP, the cortical F1 areas pertaining to the hand field and in the several regions of the somatosensory cortex. An intrinsic spatial optical imaging (ISOI) study [51] had explored spatial domains that were active in M1 and S1 in response to reaching and grasping in macaque monkeys.

Assessing synchrony and desynchrony of rhythmic oscillations had allowed dissecting cortical preparation or execution of voluntary movements [52-54]. In the motor cortex, μ rhythms in the α band [22, 55], and attenuation of the α-band were observed during the preparation or execution of voluntary movements, which was also accompanied by a decrease in the β band decrease, known as event-related desynchronization (ERD). Further, a rebound of the β-band after the movement, known as event-related synchronization (ERS) [22] was used to detect changes in movement execution. From an application standpoint, ERD/ERS features allow the reuse EEG signal information within the context of a brain-machine interface (BMI) and similar methods in the anthropometric articulation of external devices [53, 54, 56–58]. Control paradigms were employed to operate wheelchairs using decoded ERD/ERS [59, 60] information. During the movement preparation and execution, ERD was explored for investigating cortical modifications and was shown the delayed ERD in Parkinson’s Disease (PD) patients compared with healthy subjects over the contralateral sensorimotor areas [22, 61]. In the S1 area, for self-paced finger or wrist-extension movement, a decrease in 10–14 Hz frequencies and a transient increase in activity (ERS) after the movement for 20 and 90 Hz frequencies were observed [62]. An increase in the 40 Hz gamma-band activity (ERS) over the central scalp region was reported during both simple response time (SRT) and complex response time (CRT) task reactions [63, 64]. Studies show correlations in the slow (6–11 Hz) and to the higher © band region of (60–120 Hz) [65, 66] corresponding to any of the 4 fingers being moved and their flexion and trajectories of finger movements during grasping.

Other than μ or β band activity, studies have reported γ band activity in the motor and premotor areas from intracranial electrode recordings and electrocorticogram recordings [67-69]. The subdural recordings of electrocorticography (ECoG) signals performing visuomotor tasks designed to activate representations of different body parts had shown © bands were distinct in the sensorimotor cortex and the ERS of γthe band has been observed during or just before the motor response and lasted for a short time and ended before the completion of the motor response. The temporal differences associated with the γ bands had implicated functional associations with motor performance [64]. During the execution phases of movement, γ band synchronisation was also reported, and γ clusters contralateral to movement have been observed [70, 71]. The findings [70] of Ramos-Murguialday and colleagues had indicated that active movement could be decoded using a low γ-band in EEG after filtering EMG artifacts from channel data.

There has been an increased demand for wearable technologies and especially, limited-channel systems for personal use. Comparing clinically graded systems and consumer systems and the applications [72-74] have been reported. The application of these consumer systems in research has not been extensively explored. It was also crucial to explore whether a limited-channel acquisition could provide interpretable data to assess the neural signatures while exploring grasped movement.

In other BCI studies [75-80], various grasp movements have been used in actual and imagined motor tasks but due to task complexity, they have not been interpreted for BCI and related neurophysiological studies. The MRCPs and spectral oscillations relationship have not been investigated in detail and remain uncertain, although both phenomena accompany sensorimotor activity, and their underlying neural circuits have been hypothesized to reside within the same anatomical structures. A few studies have explored their relationship from a neurophysiology perspective [81-84]. Through this study, the neural correlates could provide complementary information on associated grasped movement tasks where MRCPs could be related to cortical excitability shifts and ERD/ERS to the gating of task-specific thalamocortical circuits [85]. In the context of early movement prediction and detection, the number of studies addressing the relation of MRCP and ERD/ERS morphology and the classification of such data are limited [2, 24, 25].

Since lower resolution EEG was less expensive and easier to implement than the other temporally precise methods, and with the limits of EEG spatial resolution not well understood, whether specific EEG electrodes are sufficient for movement-related spatial information remains unanswered. This study explores the EEG dynamics by employing a novel protocol to understand movement intention and execution during grasped movement tasks performed in the left and right directions. While exploring the activity space underlying grasping controlled by parieto-frontal circuits [86] and central regions, this paper also attempts to relate the variations in movement initiation and grasped movement by developing classification models based on features of temporal and spectral correlations specifically with a combination of α, β, and lower γ frequency bands and readiness potentials for grasped-movement.

In our study, the rebound activity of β and γ oscillations after the movement and the slow negative deflection amplitude shift in the temporal domain were explored as these features could be potential cortical activity biomarkers for grasped movement. In this paper, we also evaluated feature-based classification of pre-movement and grasped movement using machine learning and have enumerated ranked features from these tasks that may be relevant for constructing efficient classifiers for clinically relevant movement EEG data.

2. Methods

Thirty healthy volunteers aged 18 to 30 years (mean age = 22.32 ± 1.92 years) participated in this study. This non-invasive study was reviewed and approved by the institutional ethics committee at the university. All subjects involved in this study were without any known prior medical conditions and with normal or corrected-to-normal vision. All subjects were explained the aim of the study and signed informed consent were collected before their participations in the recordings. The tasks included were grasped movement of bottle towards using left hand towards the left direction and with right-hand towards the right direction.

2.1 Experimental paradigm

The experimental setup and the task paradigm were adapted from [16] (see Fig 1). A cue-based paradigm (Fig 1B) was employed where subjects were presented with visual cues using a slide-based presentation for 45 seconds. During the recordings, the subjects were seated in a comfortable chair and the computer display was at eye level and an object to grasp was placed at a random distance of 69–82 cm (approximately the length of reaching the object by the subject’s arm) from the subject (Fig 1A). Subjects were asked to place their left or right arm parallel to the table while resting on the hand-rest of the chair. The object, a 6.5 cm diameter rigid bottle was positioned on the centre of the table.

Fig 1

Schematic representation of experimental protocol and configuration.

(A) Sitting posture of the subject during the neurophysiological recording (B) The timeline points to the temporal reference for various tasks initiated with a ‘relax’ state that started at 0s, “+” visual cue at 10s for the subject to be ‘alert’, at 15s ‘ Reach’ cue indicated the subject to reach for the object placed in the front of the subject, and at 20s ‘Object’ cue appeared to indicate the subject to grasp the object, followed by “arrow” cue at 25s to lift the bottle, consequently followed by another ‘arrow’ cue at 30s that suggested the direction in which the object was to be moved and a ‘down arrow’ cue at 35s that suggested to place the grasped object back at rest. (C) EEG Data processing analysis workflow involved the filtering of raw signal, epoch extraction, feature extraction and building a machine learning model based on processed features.

Prior to the experiment, participants were provided training to adapt to the task by progressive steps of hand movement by grasping the object. Before recordings, the subjects repeated the same procedure three or four times until they attained familiarity to perform the task.

The experimental paradigm was conducted as outlined in the following steps.

All trials commenced with a relaxation phase (blank screen), considered as a reference or baseline signal for the analysis

The subject was asked to relax for ten seconds.

The subject was alerted for the following task by a ‘+’ sign cue for 5 seconds.

The word ‘Reach’ as a cue was shown for 5 seconds, indicating the subject to reach the object.

Then an image of a bottle was shown to the subject, indicating to grasp the bottle placed in front of the table. This cue was presented for five seconds

An upward arrow cue (↑) was shown for next 5 seconds indicating to the subject to lift the bottle to the chin level.

This was followed by a leftwards arrow cue (←) for five seconds; subject was then instructed to move the bottle towards left direction using the left hand.

Following this, a down-wards arrow cue (↓) was shown for 5 seconds indicating to the subject to place the object back down on the table

A blank screen was shown as an indication of the end of the experiment trial

The same steps were repeated for right-hand right direction movement task as well.

The tasks were carried out using both hands for the two movement directions in trials defined as right-hand leftward direction (RHLD), right-hand rightward direction (RHRD), and left-hand leftward direction (LHLD), and left-hand rightward direction (LHRD). However, only LHLD and RHRD have been considered for the analysis.

2.2 EEG data acquisition

The study included the acquisition of EEG data by placing EEG sensors on the subject’s scalp. For clinically relevant data, we used a 32-electrode commercially available device (Neuroelectrics, Barcelona, Spain) positioned on the scalp according to the 10–20 international system with a sampling rate of 500Hz.

Considering commercial and limited electrode platforms, we also employed a 14-electrode device (EMOTIV EPOC+) with a sampling rate of 128 Hz. EEG signals were recorded for 10 subjects using a 32- electrode device. For each subject, four trials were performed per task. So, a total of 40 trials were performed for 10 subjects. Due to the loss of data packets, EEG data of 12 trials were discarded. Using the 14-electrode device, EEG signals were recorded for 20 subjects. In this, 10 trials were performed for 6 subjects, 5 trials for 12 subjects, 13 trials for 1 subject, and 2 trials for 1 subject have been carried out and EEG signals were recorded. Out of 175 trials, 163 were used for the data analysis, and 12 trials were rejected due to the loss of signal packets during the recording.

2.3 Offline processing of EEG data

Data analysis was performed using custom scripts written in Python incorporating the ‘mne’ package functions [87]. To obtain the spectral components initially the data was high pass filtered (1Hz) to minimise the drifts [88] and the reference (mean) was subtracted, further the data was detrended [89] and bandpass filtered using an FIR filter [90, 91] of order 20 within the range, 1Hz—60 Hz and notch filter was applied to remove line noise in the range of 50 Hz. Independent Component Analysis (ICA) [92, 93] was used to eliminate the EOG, EMG artifacts from the data. ICA considered n number of linear mixtures X, X……., Xn, n number of components in this case total number of channels have considered n = 32 number of components. signal X:

A represents mixing matrix with size n×n and s was the vector of independent components. The mixing effect, after computing the matrix generated the independent components

To obtain the independent components in this study, we used the infomax algorithm based on general optimisation principle for neural networks and other processing systems [93, 94]. The algorithm determined weights based on the maximation of output entropy of a neural network with nonlinear outputs: where: y–matrix of source estimation (y = Wx); k–number of iterations; I–the identity matrix; μ−learning rate which depended on k; g (.)–a nonlinear function. g(y) logistic function.

Further, data was segmented into non-overlapping epochs of 2s for the timelines of 0s - 10s (Relax Task), 10s-15s (premovement task), and 30s -35s (Left hand left direction movement / Right hand right direction movement). Consequently, the segmented data was used to estimate the spectral features. The spectral estimations of each rhythm were quantified using multitaper power spectral density (PSD) estimation [95, 96]. The standard multitaper PSD [97] consisted of a series of steps: multiplying each data segment by each taper, applying Fourier analysis to these products, averaging over the tapers within each segment, and averaging over the segments. The PSD was estimated for the frequency ranges of δ, θ, α/μ, β, and γ bands. The global field power (standard deviation) across the regions (central, frontal and parietal) of electrodes was obtained.

For temporal cortical potential components [29], The MRCPs typically occur at frequencies of around 0–5 Hz [98]. To minimise the drifts from the raw data reference (mean) was subtracted, further the data was detrended [89] and bandpass filtered using an FIR filter [90, 91] of order 20 within the range of 0.1Hz- 5 Hz. Further, Multiple recordings of the same trials must be taken and then averaged across the trials for extraction of MRCP from EEG traces. By averaging, the background noise is cancelled out leaving only the MRCPs, when the data from multiple trials is filtered to eliminate the higher frequency activity.

2.4 Statistical measures

Correlation analysis was carried out to determine the best electrode positions for the analysis of MRCPs. The data from channels (C3, C4, F3, F4, P3, P4) were selected and used for correlation analysis. The relation for electrode positions in different regions of cortical regions from MRCPs was analysed using Pearson correlation coefficient. For each electrode channel and N scalar observations (time points), the correlation coefficient was computed. Where x and y were two variables (electrode channels) corresponding to two electrode positions under analysis. The μx and σx values were mean and standard deviation of x (channel 1) and the μy and σy were the mean and standard deviation of y (channel 2) respectively. The goal of the correlation analysis was to identify the optimal electrode position for the relax, premovement and movement tasks. Further, the correlated channels were compared to test the significance of correlated channels for each task using multiple t-test statistics. Statistical significance was determined using the Holm-Sidak method, with alpha = 0.05. Each row (correlated pair combination of electrode) was analysed individually, without assuming a consistent SD.

The association between frequency band oscillations in the three tasks and the subject’s gender was tested using χ2 analysis. The value of χ2 was estimated using the formula, where FO was the observed frequency count and FE was expected frequency count of dependent features, the confidence interval chosen was 0.05. Similar analysis was performed to find the association between frequency band and tasks (the average count of the frequency bands was used in the contingency table for analysis).

To better understand the relationship between changes in PSD for central electrode regions across different frequency bands for all 28 trials, PSD for tasks (relax, premovement, movement) was compared. A two-way repeated measure of ANOVA with Tukey’s multiple comparison posthoc test was performed on the average of each feature extracted from the relax, pre-movement and grasped movement conditions. Similarly, we compared the frequency band features for the task’s male and female subjects for all the trials, to identify the statistical significance of the features discriminating among the male and female subjects for each task using two-way ANOVA analysis. All the statistical analysis was performed using GraphPad Prism [99].

2.5 Machine learning

Using Decision trees (DT) [100-102], support vector machines [103, 104] with different kernels, and multilayer perceptron [105, 106] with different activation functions and solvers, classification was performed on the feature-combined dataset including both spectral and temporal (RPs) information for the premovement vs movement tasks and left-hand grasped-movement vs right hand grasped movement tasks. Here, DT (criterion: ‘Gini’, splitter: ‘best’), SVM model [107-109] (regularisation parameter C = 1.0, Kernel: linear, polynomial and radial basis function, degree = 3) and for MLP (activation function: logistic sigmoid function, quasi-Newton method as a solver: ‘lbfgs’, hidden_layer_sizes: (35,2), maximum number of iterations = 300). The implementation of the algorithms was based on the scikit-learn [110] python package and was chosen for the classification of grasped movement data.

Two EEG datasets consisting of 270 samples (135 samples premovement, 135 movement trials) and 56 samples (trials or instances) (28 premovement, 28 movement trials) respectively were used in this machine learning analysis. The two datasets contained 67 and 74 features as columns which were pre-processed values from the raw data. The classifier models were trained, and accuracy was estimated using 10-fold cross-validation and the datasets had split into training and testing samples with a random choice. To explore the differences in the accuracies of the datasets, Wilcoxon signed ranked t-test [111] was employed, and the t-test had indicated that there were no significant effects (p = 0.65) in the classification accuracies when using electrodes from two different headsets data (also see S4 Fig in S1 File).

2.5.1. Pre-classification feature selection for pre-movement and grasped movement tasks

20 best-ranked features in the dataset were identified using the feature selection ranker search algorithm [112, 113], which indicated that the frequency sub-bands of μ/α (7–10 Hz), β (18, 25–29 Hz), γ (30–32 Hz) and related potential (see Fig 2).

Fig 2

Feature ranking of surface EEG signals from central electrodes C3 and C4 showed accuracy depended on α sub-band 8.62 Hz and β sub-bands 14, 15, 19, 20, 21, 22 Hz while classifying movement (gray) against premovement (light grey/white) class labels (Class 0 is premovement, Class 1 is grasped movement).

3. Results

3.1 MRCP-related time-domain variations allow to decode grasped- movement

Associated to C3-C4 electrodes (see S1 Fig in S1 File), the postcentral, precentral gyrus and the primary motor cortex (M1) has a role in the initiation and fine control of movement; and, in this study, the central electrode regions (C4-C3) reported a higher correlation measure (r = 1) for movement and “alert”/pre-movement conditions compared to “relax” or no-movement state (r = 0.3) (Fig 3A, also see S4 Table in S1 File for other electrode regions). A combination of parieto-central and frontocentral region electrodes showed a correlation measure of 0.8 and 0.4 for movement and premovement conditions respectively (Fig 3A). Parieto-frontal (P4-F4) regions reported a correlation measure of 0.5 for the premovement/ “alert” and “relax” tasks, 0.6 for the “movement” task. The correlation-based analysis recommended C3-C4 combination for premovement and movement tasks due to the higher values compared to P4-C4, P4-F4, P4-C3, C4-F4, C4-F3 combinations.

Fig 3

Temporal correlations across MRCPs from EEG electrode regions differentiate movement tasks.

(A) Correlation measures for central regions (C3-C4) electrode pair has indicated higher correlation (r = 1) for movement and ‘alert’ condition tasks whereas for a lower correlation measure (r) of 0.3 was estimated for the ‘relax’ task., P4-F4 electrode pair has indicated a measure of 0.5 for premovement and relax tasks and 0.6 for movement task. The electrode combination of C3-C4 for movement and “alert” condition had higher correlation measure compared to P4-C4, P4-F4, P4-C3, C4-F4, C4-F3 combinations. (Blue circle–“relax task”, Red square–“movement task”, Green triangle–“alert/premovement” task (B)Correlation measure of various electrode pair combinations was statistically compared for the tasks using t-test. and C3-C4 electrode combination were significantly different for “relax” and movement tasks indicating central region electrodes could be decisive in differentiating “relax” and grasped movement tasks. (C) potential amplitude decreases in 296% from premovement to movement in the central electrode regions (D) Parietal regions has shown 164% decreased in amplitude (E) Frontal regions has shown 133% decrease and (F) Occipital regions has shown 230% decrease in potential amplitude before 500ms to onset of movement. (Green line–“premovement task” and Red Line indicates “movement” task).

A t-statistic value of 70 for the “relax” and the movement tasks at the central electrode regions indicated that the two tasks were significantly different from each other for C3-C4 (Fig 3B). P4-C4 combination of electrode regions reported a t-statistic of 45 for “relax” and “movement” tasks and P3-F4 and P4-F4 regions indicated a t-statistic value of less than 10. F4-C3, F4-F3, and C3-F3-central regions showed t-statistic values of less than 10 suggesting the tasks may not be different from each other across these electrode pairs. Among the central (Fig 3C), parietal (Fig 3D), frontal (Fig 3E), and occipital regions (Fig 3F), although no change in amplitude of the MRCP was observed during “alert” or “premovement” condition, a shift from the negative to positive wave was observed for movement tasks before movement (-0.5s) until the onset of movement (Fig 3).

3.2 Increased motor cortical β and γ frequency bands act as biomarkers for detecting movement during a grasped movement task

Scalp topographies indicated the presence of the attenuated α band (ERD) and amplified β band (ERS) modulations during movement initiation in the central parietal regions over the different sub-band regions (see S1 Fig in S1 File). 15–25 Hz β sub-band regions showed activation along the precentral gyrus, postcentral gyrus (corresponding to Central C3, C4 electrode locations) (See S1 Table in S1 File for more electrode locations), and superior lateral occipital cortex (corresponding to parietal P3, P4 electrode locations) (see S1A and S1B Fig in S1 File).

3.2.1. Characterisation of μ/α oscillations

A 26% decrease in the μ/α oscillations was estimated for initiation to grasp movement, a 16% decrease was observed during grasp movement to post-movement and a 29% decrease was prominent during “relax” to grasp movement tasks across central electrode regions (Fig 4A). Compared to the central regions (corresponding to C3-C4 positions), frontal regions (corresponding to F3-F4 positions) reported a 7% increase in μ/α oscillations for grasp to grasped movement, and an 8% increase post-movement (Fig 4B).

Fig 4

Increased post movement activity by rebound β and γ oscillations.

(A) α/μ oscillations show 26% decrease from initiation to grasp movement 29% decrease from relax to grasp movement and 16% decrease from grasp movement to post movement in the central electrode regions. 14% and 10% decrease in beta and gamma oscillations from initiation to grasp movement and rebound increment of 14% beta and 20% gamma for post grasp movement. (B) Frontal region has shown 7% increase of alpha oscillations from grasp to grasped movement and 8% increase post movement (C)Parietal region electrodes has shown 21% and 20% decrease from initiation to grasp movement and rebound increment of 4% and 12% post movement. (For inter subject variability see S3 Fig in S1 File).

3.2.2. Characterisation of β and γ

There was a decrease of 14% of β and 10% of γ oscillations during movement initiation to grasped movement, and a rebound increment of 14% of β and 20% of γ frequencies post-movement were observed (Fig 4A). 9% decrease of β and 5% decrease of γ rhythms from initiation to grasped movement and rebound increment 4% and 12% for β and γ oscillations were observed in the frontal regions. In the parietal regions, 21% for β and 10% for γ decreased and rebound increment of 4% and 12% for β and γ oscillations. (Fig 4C). Also, a 2% increase for grasp to grasped movement and a 13% decrease post-movement for μ rhythms (Fig 4C) were observed.

A 7% increase in θ oscillations was observed post-movement in the middle frontal gyrus and frontal pole (corresponding to frontal lobes). During the task change from grasp to grasped movement, theta oscillations decreased by 20% in the central region (precentral gyrus and postcentral gyrus) and 30% in the parietal (superior Lateral Occipital Cortex) regions (See S2 Fig in S1 File).

The three tasks with the oscillations (α, β, and γ) were compared using the χ2- test. The χ2 value = 21.51 (degree of freedom = 4), indicated the percentage of brain oscillations varied among premovement, left movement, and right movement tasks rejecting the presumed null hypothesis and were statistically significant (p = 0.003). Among genders, with χ2 statistic as 89.43 (degree of freedom = 8), p <0.0001, the data showed significant variations among male and female subjects (see S2 Table in S1 File).

The Tukey test (see S3 Table in S1 File) among male and female subjects indicated that β oscillations of right-hand grasped movement from male participants were significantly different from β oscillations of premovement, left hand grasped movement, and right-hand grasped movement from female participants (p-value <0.0001) (See Fig 5).

Fig 5

Significant left and right movement in male and female subjects in the central region electrodes among β & γ oscillations.

(A, B, C) Variability between the male and female subjects for the tasks of premovement (PM) left movement (LM) and right movement (RM). (A) α oscillations for premovement task of male subjects (2.609 ± 2.025) female subjects (3.091 ± 2.245) have not shown any statistical significance similarly for left movement and right movement (p value >0.9999), (B) β and (C) γ oscillations have shown significant differences were presented with p value < 0.0001 (12 sample trials of Male subjects, 12 sample trials of female subjects).

3.3 Machine learning and task-specific classification of grasped movement

Classification for discriminating tasks on the EEG datasets from the 14-channel electrode and 32 channel electrodes across 170 samples and model evaluation was performed. SVM (polynomial order 3) gave the best accuracy in most classification tasks, with the best mean accuracy of 60% compared to SVM (with radial basis function), multi-layer perceptron (MLP), and decision tree (DT) algorithms (See S4A Fig in S1 File). With the test dataset of 100 samples, the SVM polynomial kernel had a higher performance rate of 54% compared to other algorithms (see S4B Fig in S1 File). When comparing left movement and right movement, only 49% training accuracy was noted with SVM (see S5 Table in S1 File for accuracy rate with other algorithms).

The premovement and movement tasks dataset acquired from central regions (50 training samples) allowed to generate a statistically significant model of classification with 70% training accuracy (see Fig 6A) and 50% (6 samples) test accuracy (see Fig 6B) using SVM and MLP algorithms. In the case of the left- and right-hand movement 76% training accuracy and 83% (F1 = 0.83; AUC = 1) test accuracy with the SVM polynomial kernel. (See S6 and S7 Tables in S1 File for accuracy, AUC and F1 scores) was observed. For the 20-features dataset generated with feature ranking methods, a 16% error rate was obtained when using decision trees and by pruning the ranked 20 features to 5 leaf nodes, and with minimum samples to 3 (see S5 Fig in S1 File for accuracy with 20 ranked features dataset).

Fig 6

Machine learning-based model performance.

(A) Training accuracies for central regions on using SVM had outperformed MLP and DT for left and right hand movement (SVM: Support Vector machine; SVM-(P3): support vector machine polynomial degree 3, SVM (rbf): Support vector machine—radial basis function, MLP: Multilayer Perceptron) (B) Testing accuracy performance with central regions SVMs has performed similarly with 50% accuracy for premovement & movement, and 80% accuracy for left and right hand movement classification DT has performed moderately for left vs right hand movement.

4. Discussion

In this study, an exploration of grasped arm movement dissecting the pre-movement and movement-specific MRCP and ERD/ERS morphology was performed in the context of multiple movement tasks. Interpreting EEG underlying left or right arm movement, allowed learning models to discriminate movement intention or execution for clinically relevant assessments and can be used for current and future brain-machine interfaces. Specifically, in this study, feature-based interpretations of grasped movement and premovement have been highlighted within EEG data using the underlying low-frequency time-domain data characteristics & from the γ oscillations observed in the central cortical regions.

The combination of the temporal peak of the readiness potential peak and the μ, β, and γ oscillations can be used to precisely represent premovement and grasped movement tasks. The signal’s temporal shift in deflection towards the positive before the onset of movement can be used as a potential biomarker to differentiate movement intention and grasped movement. Central regions’ importance in controlling arm movements may be justified as the accuracy reported among those electrodes was higher.

The main differences between premovement and movement may be observed within the first 0.5 s before the movement onset, mainly over the contralateral primary motor cortex (locations C3, C4). Our results suggest that the attenuated α oscillations and increased β oscillation topographies at central and parietal regions were indicative of hand movement states. A decrease in α, β, and γ-band activity compared to the “relax” (resting-state) task could be indicative of the progression of the grasped movement, and a rebound β and γ frequency activity after grasped movement can serve as spectral biomarkers of pre and post-movement assessments. In the C3 and C4 electrode regions, other than the μ and β bands, the γ band was also associated with grasped movement. Pre-movement and grasped movement conditions were significantly different in the central regions implying that the central regions may provide optimal features to train and predict BCI-related classification models. Although some of the frontal electrodes were evaluated, better accuracy observed with C3-C4 electrodes in the central regions corresponds to the capacity of discrimination of left- and right-hand movement by these overlapping hand regions. The β and γ-band variations across asymmetry may be crucial for discriminating left- and right-hand grasped movement and the pre-movement suppression and post-grasped movement rebound of β band in the context of the laterality of neural activity especially in the ipsilateral motor cortical areas, which could be relevant for the task-based evaluation of healthy and Parkinson’s disease patients.

Through machine learning analysis, model-based discriminators were generated focusing on premovement and grasped movement task classification. Machine learning classifiers could discriminate left and right direction movement using the same features as in premovement and movement tasks. Improved classification accuracy while discriminating pre-movement and grasped movement, was obtained using peak amplitude of readiness potential and the PSD estimates of μ, β, and γ oscillations. In this study, SVM with a polynomial kernel with a lower order provided the highest accuracy over other machine learning classifiers, and allowed faster model building and testing while discriminating peri-movement and movement data. MLP also performed relatively well, although not as optimally as the SVM compared to several other interpretable classifiers on the EEG datasets. In the case of reducing features, combining decision tree-based feature ranking and applying pruning as pre-processing of data could improve classification accuracy. Estimating across several EEG-based tasks and measurement modalities as the pruning could help generate a more generic learning model that may need to be tested on movement imagery and movement execution-based arm movement datasets.

The artifact removal in EEG involved the removal of eye-blink, eye movements, and tongue movements helped augment the classifier’s accuracy and was crucial to remove those data points that had EEG artifacts before evaluating with classifiers. The study also suggests that employing low-cost consumer-grade EEG devices, given their ease of integration and instead of and alongside clinical-grade devices, could capture critical information related to grasped movement and its execution had similar discrimination models and may be valuable in finding candidates for clinical trials.

5. Conclusion

Premovement and grasped movement may vary across spatio-temporal scales but discrimination of left and right grasped movement could be performed with temporal and spectral analysis and combining classification methods for decoding pre-movement neural activity in the case of stereotyped left or right arm movements. Interpretation relied on low-frequency time-domain signals and γ oscillations for decoding the left and right movement tasks and with minor variations across genders. As observed in neurological conditions, activations in ipsilateral sensorimotor areas can be used to interpret compensation mechanisms related to the movement implying bi-electrode C3-C4 regions may be task discriminative. Although not presented in this study, the temporal and spectral features can be the foundation for novel control strategies targeting the directional choices of an operational robotic arm.

(PDF)

Click here for additional data file.

14 Mar 2022

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Mukesh Dhamala, Ph. D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: No

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: 1- The title is too long, and can be truncated to: Exploring EEG spectral and temporal dynamics underlying a hand grasp movement

2- The abstract should be revised for grammatical, structural and typing errors, such as:

The data acquisition was done for 163 trails of 30 ...

Instead, say: The EEG data was acquired from 30 participants, where each performed 163 trials of a hand grasping movement.

3- The introduction should be grammatically re-structured, too many faults.

ex: Neural correlates to voluntary grasped movement .. should be: Neural correlates of

4- In the methods: mean age +/- std ?

Amrita Vishwa Vidyapeetham: If this is the name of the university, please write 'university'.

In this study, a 32-electrode and a 14-electrode recording systems were employed: which one exactly ? and why using 2 systems instead of either one ?

total of 173 trials were recorded for each task: for each participant ?

Every subject participated in two sessions of 4 trials and every trial was separated by two minutes of resting: 4 trials or 173 ?

what is is total duration of the experiment ?

what is the total duration of each trial ?

in line 174: the ”, for what ?

The grasp should be done in how many seconds ? how the participant knew the start of a new trial ? hwo does the system knew the end of a trial and the beginning of a new trial ?

2.3. Data Acquisition: here you say 32 electrodes!

line 188: is it the raw EEG data ?

line 224: what is the length of each sample ? what are these samples (features) ? only filtered EEG data ? spectral ? in which EEG bands ?

PLEASE re-write the methods section as many details are not present.

PLEASE re-write the discussion section for structural errors

Reviewer #2: This paper presents the analysis and classification of EEG signal recorded during grasp movement. The work consisted of an experimental task were healthy participant moved any of the arms in a synchronized experiment. One critical problem with this study is the lack of novelty, many aspects are presented in previous literature. Other problem is with analyses which are very poor and require significant improvement.

Some comments

1) What is the motivation to carry out this research? What is the scientific problem you are addressing?

2) Methods section: 1st paragraph: were the 173 trials recorded for each participant or for all of them? What are the tasks?

3) Experimental Design & Procedure: as I understand, each subject participant in two session and four trials were recorded per session.

4) In general, the description of the experimental task. For instance, “Experimental Design & Procedure:” did not presented anything about the use of both hand and two directions, but this information is subsequently presented in “Trial structure:”. Authors should improve the description of the experiment. In consequence, it seem that only two trials for hand and direction are recorded.

5) In the 1st paragraph of “Methods” you said that “In this study, a 32-electrode and a 14-electrode recording systems were employed.”, but then in “2.3. Data Acquisition:” you said that “we used a 32-electrode commercially available device”. This is highly confusing.

6) In many places there are not space between words and references, while in other places there are. Or the reference is after the end point of a sentence. Please review.

7) “The relative band power was estimated based on the brain activity bands for left hand movement and right-hand movement tasks.”: this seems to be a critical step that required further explanations. Its is not possible to replicate the procedure in its current state

8) More details and/or results about the use of ICA to remove artefactual EMG components embedded in the EEG should be given.

9) The data analysis descriptions is very plain and this is an important aspect of the research. In its current state is it not possible to replicate the procedures therein.

10) In the preprocessing of the data, why did you apply two filtering steps? I do not see the need for that since at the end only the BPF has an impact on the signals

11) The methods do not show how the MRCP are extracted from the preprocessing EEG signals.

12) It is not clear what you did in “2.5. Statistical measures:”. This needs significant rewording. In consequence, it is nor clear how the statistical analysis were done.

13) There are organization problems. One the is that you call figure 1, then figure 5.

14) “The dataset contained 270 samples has a sampling rate of 128 Hz and 56 samples 225 dataset has a sampling rate of 500 Hz.” Why are there two sampling frequencies?

15) I am not sure there you have MRCP since you are removing all frequency content below 1Hz.

16) What time windows of the trial were used to classify? What are the number of classes/categories? What is the reason to employ several classifiers? What is the dimension of the input vector? There are several open questions in the machine learning analysis.

17) Why to decode movement since in real BCI setting with final users (patients) they are not able to move or at best they only have residual movements (obviously depending on the medical condition)

18) The number of trials is very limited for each participant

19) It seems you are combining data from all participants. This is not a common strategy in the analysis of EEG signals, and they are usually done subject specific.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Bilal Alchalabi

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

27 May 2022

We thank the editor and the reviewers for their constructive and valuable comments, and we have used the same to address changes in the manuscript. We have included additions also to our supplementary material.

Reviewer 1:

1. The title is too long, and can be truncated to: Exploring EEG spectral and temporal dynamics underlying a hand grasp movement

The title has been modified to "Exploring EEG spectral and temporal dynamics underlying a hand grasp movement", as suggested by the reviewer. Thank you for the suggestion.

2. The abstract should be revised for grammatical, structural, and typing errors, such as:

The data acquisition was done for 163 trails of 30.

Instead, say: The EEG data was acquired from 30 participants, where each performed 163 trials of a hand grasping movement.

Thank you for the comment. The abstract has been revised, and the manuscript was checked for grammatical and language errors and typos. The abstract has been slightly modified (line number 16 to 38 on page number 2) to reflect these adaptations.

3. The introduction should be grammatically re-structured, too many faults.

ex: Neural correlates to voluntary grasped movement. should be: Neural correlates of

Thank you. Agreeing with the reviewer, we have revised the manuscript.

As suggested by the reviewer, the statement on page number 4, line 49 have been rewritten as "Neural correlates of voluntary grasped movement are relevant in the development of modern prosthetics, and towards Brain Computer Interface (BCI) research (Daly and Wolpaw, 2008; Lebedev and Nicolelis, 2006; Wolpaw et al., 2000)."

4. In the methods: mean age +/- std?

Thirty healthy volunteers aged 18 to 30 years (mean age = 22.32 ± 1.92 years) participated in this study. This has been added to the methods section on page 9, line number 177)

5. Amrita Vishwa Vidyapeetham: If this is the name of the university, please write 'university'.

As per Indian legal records, the university is called "Amrita Vishwa Vidyapeetham", where "Vishwa Vidyapeetham" is the Sanskrit term for university. However, for practical reasons, as indicated by the reviewer we have used the term "University". The corrections can be seen on page number 9 at line number 179 in the manuscript.

6. In this study, 32-electrode and 14-electrode recording systems were employed: which one exactly? and why using 2 systems instead of either one?

In this study both 32 -electrode medical graded system and 14 electrode consumer graded systems have been used. The consumer graded 14 electrode device is equipped with Frontal, temporal, parietal and Occipital sensors (AF3,AF4, F3,F4,F7,F8,FC5,FC6,P7,P8,O1,O2,T7,T8) and the medically graded 32 electrode device is equipped with 32 sensors from the same regions including central region (FP1,FP2,AF3,AF4,F3,F4,Fz,FC1,FC2,F7,F8,FC5,FC6,C3,C4,Cz,CP1,CP2,CP5,CP6,T7,T8,P7,P8,P3,P4,PO3,PO4,Pz,O1,O2,Oz ). Based on the study of comparison of medical and consumer EEG systems by Ratti et al., 2017 consumer systems were prone to artifacts in frontal regions, medical devices offer clear advantages in data quality, reliability, and depth analysis over consumer systems.

Since the study is more focused in clinically aspects and application of neuro prosthesis at consumer level, it was crucial to look at the accuracy from obtained features using both systems and to map neural activity that were not restricted to frontal and temporal regions only.

Though 32-electrode can provide a greater dataset compared to the 14-electrode, the 32-electrode device was more expensive (computationally). As researchers based in a developing country like India, we also are focused on using and implementing how low-cost devices can be used reliably. With this as a possible alternative for recordings outside lab settings, we have checked whether 14-electrode device provided similar or reliable accuracy like that of a 32-electrode device. The results suggest that features from central electrode regions provide reliable accuracy for premovement and movement tasks, compared to frontal electrode regions. The clear discrimination of MRCP pattern for movement initiation was observed from the central region electrodes.

In the present manuscript, introduction section on page 7, line number 135 to 140 and on page 8 line 156 to 166 we have now modified the text with description. We thank the reviewer for the comment.

7. total of 173 trials were recorded for each task: for each participant? Every subject participated in two sessions of 4 trials and every trial was separated by two minutes of resting: 4 trials or 173?

In this study, we have recorded 175 trials from 30 subjects. Off these 30 subjects, 10 subjects were recorded using 32-electrode system. For each subject, four trials per task were carried out. 40 trials were performed with 10 subjects. Due to the loss of data packets, EEG data of 12 trials were discarded. For 20 subjects, the 14-electrode system was used. With that 14-electrode device, 10 trials were performed for 6 subjects, 5 trials for 12 subjects, 13 trials for 1 subject, and 2 trials for 1 subject have been carried out and EEG signals were recorded. The recording for each task was taken in two separate sessions, for LHLD, RHRD tasks

We have included this information in the present draft on page 11, line number 233. We thank the reviewer for the comment.

8. what is total duration of the experiment? what is the total duration of each trial

Each recording employed in this study was for 45 seconds and considered as a trial of the experiment. We have reedited the methods in the manuscript (line number188, page number 10) as read below.

A cue-based paradigm (Fig 1B) was employed where subjects were presented with visual cues using a slide presentation of the duration of 45 seconds. we have recorded each trial as an experiment. EEG signals were recorded for 10 subjects using a 32- electrode device. For each subject, four trials were performed per task. So, a total of 40 trials were performed for 10 subjects. Due to the loss of data packets, EEG data of 12 trials were discarded. Using the 14-electrode device, EEG signals were recorded for 20 subjects. In this, 10 trials were performed for 6 subjects, 5 trials for 12 subjects, 13 trials for 1 subject, and 2 trials for 1 subject have been carried out and EEG signals were recorded. Out of 175 trials, 163 were used for the data analysis, and 12 trials were rejected due to the loss of signal packets during the recording.

9. in line 174: the", for what?

Thank you for pointing out a typo, we have edited the manuscript (page number 10, line number 205-220)

10. The grasp should be done in how many seconds? how the participant knew the start of a new trial? How does the system knew the end of a trial and the beginning of a new trial?

The protocol has been recorded for 45 seconds duration with different cues, The Relax cue was show for 10 seconds, alert cue was shown for 5 seconds, reach cue was shown for 5 seconds, grasp cue was shown for 5 seconds, movement cues were shown for 5 seconds and stop cue with blank screen shown for 5 seconds. The next trial was started after 3 minutes break. The blank screen cue presentation indicates the beginning and end of the trials.

This has been incorporated in the Methods sections, Experimental Paradigm on page 10, Line numbers 205-220 clarified with different steps of the protocol in the manuscript.

11. Data Acquisition: here you say 32 electrodes!

We have now edited the methods to clarify this point as well.

As indicated previously, in this study, we have recorded from two different set of experimental devices with frontal and central electrode sensors and have analyzed data separately. The results suggest that features from central electrode regions have more accuracy prediction rate for premovement and movement tasks, compared to frontal electrode regions. The clear discrimination of MRCP pattern for movement initiation was observed from the central region electrodes.

For clinically relevant data, we used a 32-electrode commercially available device (Neuroelectrics, Barcelona, Spain) positioned on the scalp according to the 10–20 international system with a sampling rate of 500Hz. Considering commercial and limited electrode platforms, we also employed a 14-electrode device (EMOTIV EPOC+) with a sampling rate of 128 Hz. EEG signals were recorded for 10 subjects using a 32- electrode device. For each subject, four trials were performed per task. So, a total of 40 trials were performed for 10 subjects. Due to the loss of data packets, EEG data of 12 trials were discarded. Using the 14-electrode device, EEG signals were recorded for 20 subjects. In this, 10 trials were performed for 6 subjects, 5 trials for 12 subjects, 13 trials for 1 subject, and 2 trials for 1 subject have been carried out and EEG signals were recorded. Out of 175 trials, 163 were used for the data analysis, and 12 trials were rejected due to the loss of signal packets during the recording.

Thank you for the question.

12. line 188: is it the raw EEG data?

We refer to the raw data acquired from the device was detrended to minimize the drifts and further band pass filtered 1- 60 Hz range. This now reads edited as line 246 on page 12. Here is the edited text from the manuscript.

To obtain the spectral components initially the data was high pass filtered (1Hz) to minimise the drifts [88]and the reference (mean) was subtracted, further the data was detrended [89] and bandpass filtered using an FIR filter [90,91] of order 20 within the range, 1Hz - 60 Hz and notch filter was applied to remove line noise in the range of 50 Hz.

13. line 224: what is the length of each sample? what are these samples (features)? only filtered EEG data? spectral? in which EEG bands?

We have used two different datasets from two different devices, the first dataset with 56 samples (no. of rows/ no. of trials) with a binary class of premovement & movement consists of 74 columns which are features (readiness potential peak (component of MRCP), alpha band, beta band, and gamma-band). The second dataset with 270 samples (no. of rows / no. of trials) consists of 67 columns which are features.

The features are readiness potential peak and spectral oscillations of alpha, beta and gamma band obtained after preprocessing the raw EEG data.

14. PLEASE re-write the methods section as many details are not present.

Thank you for the comment. We have rewritten the methods section as suggested.

We have edited the “Experimental Design & Procedure” completely restructuring it as Experimental Paradigm. The present draft in the manuscript from the line number 187 to 230 has been re-addressed with stepwise procedure to perform the experiment and has included number of subjects considered for data acquisition and number of trails recorded in total from all the subjects. We have also modified the trial structure part as a step-by-step method for ease of clarity and understanding (from line number185 to 225 at page number 11)

In the methods section the offline processing data was added and indicates our methods for power spectral analysis and movement related cortical potentials. (Page number 12, Line number 244-292).

The statistical analysis was also re-structured and elaborated (page number 14 line numbers 293- 332 in the manuscript).

15. PLEASE re-write the discussion section for structural errors

We apologize for the issues in the previous draft and thank the reviewer for the comment. We have avoided structural issues (page 21 at line numbers 457 -513) and redrafted the discussion section.

Reviewer 2:

1. What is the motivation to carry out this research? What is the scientific problem you are addressing?

We thank the reviewer for this question. Our motivation for this study is now part of the introduction in the present manuscript on page 4, line number 43 -48 and on page 8, line number 159-166

Patients with amyotrophic lateral sclerosis (ALS) or spinal cord injury (SCI) are reported to have significant loss of voluntary motor control and extensive dysfunction of upper and lower limbs [1-4]. Rehabilitation using robotic or functional electrical stimulation could facilitate novel therapeutic strategies for such conditions. Integration of signals could also be a potential goal for Brain-Computer Interfaces (BCI), allowing the transfer of control information from a prosthesis onto the brain tissue [5-7]. Neural correlates of voluntary grasped movement are relevant in the development of modern prosthetics and towards Brain-Computer Interface (BCI) research [8-10]. Insights from electroencephalography (EEG) recordings and their underlying neurophysiological processes involved in motor tasks can help decode movement and its functional interpretation. EEG signal components related to movement allow noninvasive measurements, making them suitable for natural BMIs [5,6]. In addition, these correlates are measurable in paralyzed patients [3,4,7,11]. In this study, we explored the EEG dynamics by employing a novel protocol to understand movement intention and execution during grasped movement tasks performed in the left and right directions. Exploring the activity space underlying grasping controlled by parieto-frontal circuits [86] and central regions, this paper also attempts to relate the variations in movement initiation and grasped movement by developing classification models based on features of temporal and spectral correlations specifically with a combination of α, β, and lower �  frequency bands and readiness potentials for grasped-movement.

2. Methods section: 1st paragraph: were the 173 trials recorded for each participant or for all of them? What are the tasks

We thank the reviewer for the question. We have redrafted methods to reflect the trials, questions, and procedure.

In this study, 175 trials recorded for all the participants., The tasks included left hand grasped left direction movement (LHLD), a right hand grasped right direction movement (RHRD), and a premovement. These tasks followed timelines based on the cueing paradigm and were performed by subjects for 45 seconds for each trial.

As per the reviewer’s suggestions, we have reedited the methods and have included step by step experimental paradigm in the manuscript on page 10, line number 187-225.

The experimental paradigm was conducted as outlined in the following steps.

1. All trials commenced with a relaxation phase (blank screen), considered as a reference or baseline signal for the analysis

2. The subject was asked to relax for ten seconds.

3. The subject was alerted for the following task by a ‘+’ sign cue for 5 seconds.

4. The word ‘Reach’ as a cue was shown for 5 seconds, indicating the subject to reach the object.

5. Then an image of a bottle was shown to the subject, indicating to grasp the bottle placed in front of the table. This cue was presented for five seconds

6. An upward arrow cue (↑) was shown for next 5 seconds indicating to the subject to lift the bottle to the chin level.

7. This was followed by a leftwards arrow cue (←) for five seconds; subject was then instructed to move the bottle towards left direction using the left hand.

8. Following this, a down-wards arrow cue (↓) was shown for 5 seconds indicating to the subject to place the object back down on the table

9. A blank screen was shown as an indication of the end of the experiment trial

The same steps were repeated for right-hand right direction movement task as well.

The tasks were carried out using both hands for the two movement directions in trials defined as right-hand leftward direction (RHLD), right-hand rightward direction (RHRD), and left-hand leftward direction (LHLD), and left-hand rightward direction (LHRD). However, only LHLD and RHRD have been considered for the analysis.

The same steps were repeated for right hand right direction of movement

3. Experimental Design & Procedure: as I understand, each subject participant in two sessions and four trials were recorded per session.

In this study, 175 trials recorded from 30 subjects. Out of these 30 subjects, 10 subjects were using 32-electrode system. For each subject, four trials were performed per task. So, a total of 40 trials were performed for 10 subjects. Due to the loss of data packets, EEG data of 12 trials were discarded. Yes, each participant has participated in two sessions to acquire the two separate directions of movement left-hand left direction movement and right-hand right direction movement, the recording for each task was taken in two separate sessions, (LHLD, RHRD). This has been incorporated in the manuscript methods section on page number 11 from line number 233 can be read as below paragraph.

EEG signals were recorded for 10 subjects using a 32- electrode device. For each subject, four trials were performed per task. So, a total of 40 trials were performed for 10 subjects. Due to the loss of data packets, EEG data of 12 trials were discarded. Using the 14-electrode device, EEG signals were recorded for 20 subjects. In this, 10 trials were performed for 6 subjects, 5 trials for 12 subjects, 13 trials for 1 subject, and 2 trials for 1 subject have been carried out and EEG signals were recorded. Out of 175 trials, 163 were used for the data analysis, and 12 trials were rejected due to the loss of signal packets during the recording

4. In general, the description of the experimental task. For instance, "Experimental Design & Procedure:" did not present anything about the use of both hands and two directions, but this information is subsequently presented in "Trial structure:". The authors should improve the description of the experiment. In consequence, it seems that only two trials for hand and direction are recorded.

Thank you for the suggestions, we have restructured the experimental design & procedure into a single structure called experimental paradigm and have included in the manuscript on page number 10-11 lines 187 – 225.

5. In the 1st paragraph of "Methods" you said that "In this study, a 32-electrode and a 14-electrode recording systems were employed.", but then in "2.3. Data Acquisition:" you said that "we used a 32-electrode commercially available device". This is highly confusing.

This was an important choice for us and we thank the reviewer for raising the question. Both 32 -electrode medical graded system and 14 electrode consumer graded systems have been used in this study. The consumer grade 14 electrode device was equipped with Frontal, temporal, parietal and Occipital sensors (AF3,AF4, F3,F4,F7,F8,FC5,FC6,P7,P8,O1,O2,T7,T8) and the medical grade 32 electrode device was equipped with 32 sensors from the same regions including central region (FP1,FP2,AF3,AF4,F3,F4,Fz,FC1,FC2,F7,F8,FC5,FC6,C3,C4,Cz,CP1,CP2,CP5,CP6,T7,T8,P7,P8,P3,P4,PO3,PO4,Pz,O1,O2,Oz ). Based on the study of comparison of medical and consumer EEG systems by Ratti et al., 2017 consumer systems were prone to artifacts in frontal regions, medical devices offer clear advantages in data quality, reliability, and depth analysis over consumer systems.

Since the study was focused on interpreting signal aspects for evaluating pre and movement information at consumer level, it was helpful to look at the accuracy of obtained features using both systems and in order to map neural activity that was not related to frontal and temporal regions.

The 32-electrode could yield more data compared to the 14-electrode and yet, the 32-electrode device was more expensive (computationally and financially). Apart from this paper, it was crucial to evaluate the applicability of low-cost devices especially in off laboratory settings. With this as an additional reason, we employed the 14-electrode device and tested if results were reliable as performed on a 32-electrode. Our results suggest that features from central electrode regions yielded better accuracy for premovement and movement tasks, compared to frontal electrode regions. The discrimination of MRCP patterns for movement initiation was observed from the central region electrodes.

In the present manuscript, introduction section on page 7, line number 135 to 140 and on page 8 line 156 to 166 we have now modified the text with description.

6. In many places there are not space between words and references, while in other places there are. Or the reference is after the end point of a sentence. Please review.

Thank you for pointing out this issue. We have corrected the same in the present draft of the manuscript (line numbers 59, 82,83, 94,131, 248,250, 334,335,339).

7. "The relative band power was estimated based on the brain activity bands for left hand movement and right-hand movement tasks.": this seems to be a critical step that required further explanations. It is not possible to replicate the procedure in its current state

We thank the reviewer for pointing out our short sightedness in this regard. Taking the reviewer’s suggestion, the methods section has been modified and details have been added (line number 277 on page 13).

8. More details and/or results about the use of ICA to remove artefactual EMG components embedded in the EEG should be given.

ICA analysis was performed to remove any artifacts related to EMG, EOG, from the components. AS per the author's knowledge, the authors haven't observed most of the EMG artefactual components in the data, but mostly in the data the first component has seen EOG artifacts which have been removed during the pre-processing stage. In this study Infomax ICA algorithm was used from python mne package to do ICA analysis. The detailed description of ICA method has been edited in the manuscript on page 12 lines 251- 271.

9. The data analysis descriptions is very plain and this is an important aspect of the research. In its current state is it not possible to replicate the procedures therein.

Thank you for the comment. The manuscript has been modified to reflect changes in the methods section (per say line number 244 on page 12). The present manuscript has a section on data analysis (as listed under Offline processing of EEG data).

10. In the pre-processing of the data, why did you apply two filtering steps? I do not see the need for that since at the end only the BPF has an impact on the signals.

We thank the reviewer for the comment. We have revised the methods section to address the filtering aspects (page numbers 12 & 13 at line numbers 247 & 286). In our study, we have used high pass filter above 1Hz to minimise the drifts. Spectral oscillations in the frequency range 1Hz to 60 Hz were estimated. To obtain MRCPs, the EEG data was high pass filtered for the 0.1Hz to 5 Hz range.

11. The methods do not show how the MRCP are extracted from the pre-processing EEG signals.

Thank you for the comment. We have now edited the methods to include on extracting MRCPs. (page 13, line 285).

12. It is not clear what you did in "2.5. Statistical measures:". This needs significant rewording. In consequence, it is nor clear how the statistical analysis were done.

We agree with this comment. Thank you. In the present draft of the manuscript, we have rewritten the methods section (page 14 line numbers 294 to 329).

13. There are organization problems. One the is that you call figure 1, then figure 5.

Thank you for pointing out that oversight from our side. We have modified the manuscript.

14. "The dataset contained 270 samples has a sampling rate of 128 Hz and 56 samples dataset has a sampling rate of 500 Hz." Why are there two sampling frequencies?

Two different datasets have been used for the analysis obtained from medically graded acquisition device which had 500 Hz sampling rate and other one was a consumer grade device at 128Hz sampling rate. The dataset obtained from consumer grade device contains 270 samples (135 premovement, 135 movement) and the second dataset obtained from medical graded device contains 56 samples/trials (28 premovement/ 28 movement)

15. I am not sure there you have MRCP since you are removing all frequency content below 1Hz.

We thank the reviewer for this comment. The methods section has been modified to show analysis for spectral components and for MRCPs separately (line number 285 on page 13). To obtain the MRCPs, a bandpass filter of 0.1 Hz – 5 Hz was exployed. High pass filter 1Hz was used to obtain the spectral components.

16. What time windows of the trial were used to classify? What are the number of classes/categories? What is the reason to employ several classifiers? What is the dimension of the input vector? There are several open questions in the machine learning analysis.

We thank the reviewer for the comment. In this paper, time windows of 10s(premovement), and 30s (Left hand grasped movement / right hand grasped movement) of the trial were used in the classification. We chose binary class labels (premovement, movement), (left movement, right movement) and have used two datasets from 2 devices; with the first dataset dimension is 170 * 67 and the second dataset dimension is 56 * 74.

We have reported this in our re-edited methods section, which has a Machine learning subsection (page number 16, from line number 330).

17. Why to decode movement since in real BCI setting with final users (patients) they are not able to move or at best they only have residual movements (obviously depending on the medical condition)

Our expectation here was that we targeted the classification for future tools and prostheses. We have started an ongoing study to allow BCI based transfer of control using robotic devices and within such contexts both peri-movement and movement were crucial. Sensory and Cognitive functions of the brain are only minimally affected by amyotrophic lateral sclerosis (ALS) or spinal cord injury (SCI). Rehab with robotic interfaces is also a potential goal for Brain Computer Interfaces (BCI) leading to a possible control or transfer of control signals from prosthesis to actual brain tissue.

We included some aspects of this into our introduction section (page number 4 from line 43-59).

18. The number of trials is very limited for each participant

We agree that the number of trials (175) were limited. This was also due to restrictions imposed by COVID. We have however considered all trials together from the subjects as inter subject modelling generalizing the problem using different datasets from various electrode positions for classification with binary class labels. We do hope to extend this study with other BCI components and with our 256-electrode device as we limp back into some form of normalcy after the COVID lockdowns.

The number of trials and the subjects has been mentioned in the present manuscript at methods section on page 11 from line numbers 227 – 242.

19. It seems you are combining data from all participants. This is not a common strategy in the analysis of EEG signals, and they are usually done subject specific.

Thank you for the comment. For the relevance of commonality across same tasks, we have looked into inter-subject modelling approaches of EEG using different datasets obtained from 2 different EEG devices for machine learning purpose., the analysis and feature extraction of EEG signal for each trial has been done individually and the average responses were represented in the figures (Fig 3, Fig 4). Although it was a risk, we found data from central electrodes allowed reliable discrimination across subjects performing the tasks recorded using both devices.

We are happy

Submitted filename: Bodda2022_Reviewer_comments_answers_secondversion.docx

Click here for additional data file.

9 Jun 2022

Exploring EEG spectral and temporal dynamics underlying a hand grasp movement

PONE-D-21-37461R1

Dear Dr. Diwakar,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Mukesh Dhamala, Ph. D.

Academic Editor

PLOS ONE

13 Jun 2022

PONE-D-21-37461R1

Exploring EEG spectral and temporal dynamics underlying a hand grasp movement

Dear Dr. Diwakar:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Mukesh Dhamala

Academic Editor

PLOS ONE