EEG response varies with lesion location in patients with chronic stroke

Twelve patients with chronic stroke (9 males, 3 females; mean ± SD age, 54.0 ± 6.6 years) participated in this study. All of the participants had a single stroke, exhibited unilateral motor problems in the upper extremities/limbs that continued for at least 3 months after their stroke, and were aged between 45 and 70 years old. Patients with cognitive disorders that rendered them unable to understand the task instructions and/or those with orthopedic disorders that led to amputation or joint contraction were excluded. The mean ± SD Fugl-Meyer Assessment scores were 47.3 ± 9.2 and 64.8 ± 9.2 for the affected and unaffected sides, respectively. Lower scores indicate more severe impairment. The patients did not have any history of neurological illness. The characteristics of these patients are provided in Table 1. Grasp strength, Purdue Pegboard Test, and Fugl-Meyer Assessment (FMA) have been used to evaluate motor functions of patients with stroke during rehabilitations phases [27‐29]. More specifically, grasp strength shows the physical strength of the hand (clinical norms for the 55–59 years age group: men: right hand, 45.8; left hand, 37.7; women: right hand, 25.9, left hand 21.4 [kg]), Purdue Pegboard Test indicates the delicate control ability of the hand function (norms for the 55–59 years age group: men: right hand, 19.2; left hand, 21.0; women: right hand, 17.8, left hand 19.4), and FMA is generally used to evaluate the upper-limb functions for volitional movement ranges and reflex activities (scored on a scale of 0 and 66). Patients have significant differences in hand function between affected and unaffected hands in Grasping strength, Purdue Pegboard Test, and FMA (rank sum test, p < 0.01). These scores were used as exclusion criteria for patients with severe impairment (0 to 20 FMA score), and all participants in the moderate (21 to 50 score) or mild (51 to 66 score) categories, who were able to perform the motor tasks, were included [29]. A radiologist assessed and categorized lesion location based on the MRI data: (1) supratentorial lesions that included M1 (hereafter, SM1+), (2) supratentorial lesions that excluded M1 (SM1-), and (3) infratentorial (INF) lesions. SM1+ indicates a cortico-subcortical lesion and damaged M1, whereas SM1- indicates a subcortical lesion without M1 damage. The lesions of the SM1+ and SM1- groups are located in the supratentorial area while those of the INF group are in the infratentorial area. In addition to the patients with stroke, twelve age- and sex-matched HCs (8 males, 4 females; 57.8 ± 4.7 years) served as controls. No subjects had previously participated in an EEG experiment. The Institutional Review Boards of the Samsung Medical Center (Application Number: SMC 2013-02-091) and Korea Institute of Science and Technology (Application Number: KIST 2013–009) approved this study. The participants were informed about the study’s purpose, experimental procedures, and their right to withdraw at any time. Written informed consents were obtained from all of the participants. All of the research data were collected and analyzed under Institutional Review Board guidance.

In this study, the subjects were asked to conduct grasping and supination movements with the affected hand; these are two basic hand functions involved in activities of daily living. They performed each movement with active, passive, and motor imagery (MI) tasks. In the active task, subjects were asked to perform a given movement with motor intention by themselves. A robotic device performed the movement in the passive task. In the MI task, each subject was asked to imagine the movement with motor intention, but he or she did not perform the physical movement. The experimental protocol consisted of three motor tasks, each composed of three blocks (nine blocks in total). Each block consisted of 14 repeated trials, and each trial consisted of four time periods: relax, motor task, stay, and return. A fixation appeared on the screen during the relax period with a random duration between 2 and 3 s. Participants performed a motor task in the 2-s motor task period, which started with auditory and visual cues. The 1-s stay period is necessary in order to prevent the risk of a sudden movement change. Then, the robotic device was reset to its original handle position during the return period in the case of active and passive motor tasks. Therefore, each participant performed 42 sequential trials (14 trials for each of the three blocks) for each of the three motor tasks (active, passive, and MI), accounting for a total of 126 trials; EEG data were recorded during the entire experimental protocol.

EEG signals were acquired with a 64-channel EEG active electrode system (sampling rate: 2,048 Hz; Active-two, BioSemi S.V., Amsterdam, Netherlands). The acquired EEG signals were preprocessed using the following steps: downsampling, 1–80 Hz band-pass and 60 Hz notch filtering, trial epoching, independent component analysis (ICA) for electrooculographic and muscle artifacts removal [30], and common average reference (CAR) [31]. In our study, the CAR was used for re-reference with the average of whole EEG channels for each individual EEG channel. Alternatively, the Laplacian montage method can be used when the local average surrounding a target EEG channel is adopted to adjust the bias of the target channel [32]. After preprocessing, spectral power was computed using short-time Fourier transform with a 500-ms hamming window, and sliding by 50 ms for each of the 64 EEG channels. The baseline of each epoch was defined as the 1 s before the motor task cues. The spectral power was normalized by subtracting the baseline mean from each data point in an epoch and by dividing the resulting value by the baseline SD. The ERD/ERS was defined as the spectral power changes in the motor task period relative to the baseline. Two frequency bands selected in our study include the mu (8–13 Hz) and beta (13–32 Hz) bands, both of which reflect sensorimotor rhythms. Detailed information on the experimental protocol and the EEG processing method can be found elsewhere [20].

The quantitative analyses of the EEG data were based on the LC and topographic mapping of the EEG spectral power. The hemispheric asymmetries for ERD/ERS, LC was calculated as follows:

where C denotes the ERD/ERS of the contralateral motor cortex and I denotes the ERD/ERS of the ipsilateral motor cortex [23, 33].

We compared the LC values across different combinations of the frequency bands (mu and beta bands), motor tasks (active, passive, and MI tasks), movements (supination and grasping movements), and participants (SM1+, SM1-, INF, all patients, and HCs). We observed the LC pattern in the mu and beta bands because these bands are known to be associated with motor movement.

In the analysis of this study, we focused on the active and MI motor tasks because the passive motor task using a robot-guided device would lack of the subject’s motor intention, a key factor in effective rehabilitation [34, 35]. ERD/ERS patterns on the active and MI motor tasks was compared between the subgroups of patients and the HCs. More specifically, Pearson’s linear correlation analysis was performed using the ERD/ERS power changes in the bilateral motor cortex during the motor task period. For the topographical analysis, we selected 28 EEG channels around the bilateral motor cortex and supplementary motor area (SMA), both of which are associated with motor movement. For each of the 28 channels, the ERD/ERS power changes were averaged across all HCs or each of the patient subgroups. In addition, we compared the EEG topographies from all possible combinations across the two frequency (i.e., mu and beta) bands, three motor tasks (i.e., active, passive and MI tasks), two different movements (i.e., grasping and supination), and four subject groups (i.e., SM1+, SM1-, INF and HCs). A Pearson’s correlation coefficient was calculated using the ERD/ERS power changes between the HCs and each of the three patients subgroups for each of the 28 channels. Then, one-way ANOVA was performed across the three patient subgroups using the 28 correlation coefficients across the 28 channels from each subgroup.

Figure 1 illustrates a comparison of the LC values in the beta band of patients and controls. The five bars indicate the LC values of SM1+, SM1-, INF, all patients, and HCs. The LC pattern in the mu band is displayed Additional file 1: Figure S1 in the additional material. In HCs, the ERD in the contralateral motor cortex was stronger than that in the ipsilateral motor cortex regardless of the movement and task types, which resulted in positive LC values.

The difference in the LC values between all patients and HCs was not significant, even though all the patients represented lower LC values compared to the controls in the active and MI tasks.

Figure 1 shows that the SM1+ subgroup had a negative LC value in both of the movements in the active and MI motor tasks. Especially in the active task, there were significant differences between the SM1+ subgroup and HCs (rank-sum test, p < 0.05). The SM1- and INF subgroups had positive values in the same condition. For the passive task, LC values were very small values. It indicates the brain activation in bilateral motor cortex. The SM1+ subgroup exhibited negative LC values while they performed the MI task; however, these values were not significantly different from those of the other groups.

Figure 2 shows the average power patterns of the beta band of the three patient subgroups and HCs during the 2 s supination movements in the active and MI tasks. The average power patterns of the beta band showed marginal differences between the ipsilateral and contralateral sides of the motor cortex and between the active and MI tasks. The ERD in the contralateral motor cortex was generally stronger than that in the ipsilateral motor cortex. The ERD of the HCs appeared stronger than those of the patient subgroups, except in the ipsilateral motor cortex during the active task.

Table 2 lists Pearson’s linear correlation coefficients of the average beta band power (shown in Fig. 2) calculated between each patient subgroup and the HCs. In most cases, the correlation coefficients are statistically significant. Moreover, the correlation coefficients consistently decreased in the following order: INF > SM1- > SM1 + .

A topographical analysis was implemented based on the 28 EEG channels around the SMA and bilateral motor cortex. Figure 3 shows the average beta band power distributions across the subjects in each group during the supination movement in the active task. The topographies in the MI task are shown in Additional file 2: Figure S2 in the additional material. The upper three rows display the topography patterns that corresponded to the three patient subgroups. For the SM1+ subgroup (first row), the ERD of the ipsilateral side was stronger than that of the contralateral side. For the SM1- subgroup (second row), the ERD of the contralateral side was stronger than that of the ipsilateral side, and it was particularly widespread. The INF subgroup (third row) showed that the ERD of the contralateral side was stronger than that on the ipsilateral side, and, in particular, the ERD distribution was focused on the motor cortex and parietal area on the contralateral side. For all of the patient subgroups in the fourth row, the ERD distribution was located in the bilateral motor cortex. In the case of the HCs in the last row, the ERD of the contralateral side was stronger than that of the ipsilateral side, and the strong ERD distribution was focused on the contralateral motor cortex.

Figure 4 shows the similarities of the beta band power changes across the 28 channels between the HCs and each of the three patient subgroups. The INF group showed similar ERD/ERS power changes in comparison to HCs, whereas the SM1+ group was represented a deviated ERD/ERS power changes compared to the HCs. The correlation coefficients differed significantly between the three subgroups (one-way ANOVA test, **p < 0.01). In the case of MI task, the similarities of the beta band power changes across the 28 channels between the HCs and each of the three patient subgroups are shown in Additional file 3: Figure S3 in the additional material.