Skip to main content
Download PDF
Download ePub

Investigating the cortical effect of false positive feedback on motor learning in motor imagery based rehabilitative BCI training

Journal of NeuroEngineering and Rehabilitation volume 22, Article number: 61 (2025) Cite this article

Abstract

Background

Motor imagery-based brain–computer interface (MI-BCI) is a promising solution for neurorehabilitation. Many studies proposed that reducing false positive (FP) feedback is crucial for inducing neural plasticity by BCI technology. However, the effect of FP feedback on cortical plasticity induction during MI-BCI training is yet to be investigated.

Objective

This study aims to validate the hypothesis that FP feedback affects the cortical plasticity of the user’s MI during MI-BCI training by first comparing two different asynchronous MI-BCI paradigms (with and without FP feedback), and then comparing its effectiveness with that of conventional motor learning methods (passive and active training).

Methods

Twelve healthy volunteers and four patients with stroke participated in the study. We implemented two electroencephalogram-driven asynchronous MI-BCI systems with different feedback conditions. The feedback was provided by a hand exoskeleton robot performing hand open/close task. We assessed the hemodynamic responses in two different feedback conditions and compared them with two conventional motor learning methods using functional near-infrared spectroscopy with an event-related design. The cortical effects of FP feedback were analyzed in different paradigms, as well as in the same paradigm via statistical analysis.

Results

The MI-BCI without FP feedback paradigm induced higher cortical activation in MI, focusing on the contralateral motor area, compared to the paradigm with FP feedback. Additionally, within the same paradigm providing FP feedback, the task period immediately following FP feedback elicited a lower hemodynamic response in the channel located over the contralateral motor area compared to the MI-BCI paradigm without FP feedback (p = 0.021 for healthy people; p = 0.079 for people with stroke). In contrast, task trials where there was no FP feedback just before showed a higher hemodynamic response, similar to the MI-BCI paradigm without FP feedback (p = 0.099 for healthy people, p = 0.084 for people with stroke).

Conclusions

FP feedback reduced cortical activation for the users during MI-BCI training, suggesting a potential negative effect on cortical plasticity. Therefore, minimizing FP feedback may enhance the effectiveness of rehabilitative MI-BCI training by promoting stronger cortical activation and plasticity, particularly in the contralateral motor area.

Background

Cortical plasticity, also known as neuroplasticity, is the brain’s ability to heal itself after an injury, enabling patients to regain brain function including motor skills [1]. This occurs through new neural connections, reorganization of pathways, and sometimes new neuron generation [2, 3]. Broadly, conventional motor learning techniques used to induce cortical plasticity after a stroke can be categorized into active and passive training methods. Passive training is beneficial in patients who are unable to move their bodies voluntarily; however, continuous repetitive movements devoid of the user’s intention are problematic. Passive training, while crucial for muscle maintenance, lacks the motor planning and cognitive engagement necessary to stimulate neural plasticity, leading to reduced neural activation, motivation, and meaningful feedback for functional recovery [4,5,6]. The motor imagery-based brain–computer interface (MI-BCI) is a novel solution to this problem; it provides users with the intended movement feedback based on users’ motor intention, using the modalities such as electroencephalogram (EEG) [7,8,9,10,11,12,13]. Delivering well-timed and correct movement feedback is critical in making MI-BCI training analogous to active training, which is the ultimate goal of rehabilitative MI-BCI [12,13,14].

Typically, MI-BCI systems can be classified into synchronous and asynchronous systems. Asynchronous MI-BCI is a self-paced system in which detection is continuous, whereas synchronous MI-BCI is a cue-based technique in which detection is restricted within a predetermined time frame following the cue to attract the attention of users. Several studies have claimed that asynchronous MI-BCI systems are more closely related to activities of daily living [15, 16]; however, a major disadvantage of the asynchronous system is the possibility of false positive (FP) feedback, i.e., incorrect feedback not originating from user intention, even though false positive (FP) can be present in both synchronous and asynchronous systems [17,18,19]. Many studies assumed that FP feedback in MI-BCI training could undermine user training and induce negative neuroplasticity [18,19,20,21]; nevertheless, the issue remains outstanding [20].

Several attempts have been made to investigate the effects of FP feedback on MI-BCI training [21,22,23,24,25]. Previous studies have primarily described FP feedback in BCI systems as a factor that degrades performance and causes user frustration by triggering unintended action [21, 22, 24, 25], While FP feedback can sometimes promote greater learning effects during initial learning [22, 23]; however, these studies focused either on the FP-related potential (e.g., error-related potential [25]) or the relationship between FP feedback and BCI operation ability (e.g., accuracy and performance of BCI [22,23,24]). The studies primarily utilized FP feedback itself in terms of BCI performance or investigated the effect of FP feedback in the viewpoint of MI-BCI accuracy, not neuroplasticity. Hence, there is still a lack of clarity about the impact of FP feedback on inducing neuroplasticity during task training. A recent study compared the corticomotor excitabilities of several MI-BCI systems using transcranial magnetic stimulation [26]; an asynchronous FP-free MI-BCI system was implemented using the cue-based approach. The results demonstrated that higher cortical plasticity was induced using this approach compared with other MI-BCIs; however, it was unable to differentiate whether the effects were due to the cues or the FP feedback because the experiment was not designed to specifically verify the effects of FP feedback. Thus, the cortical effects of MI-BCI training must be investigated under different feedback conditions.

Cortical plasticity is directly related to motor recovery after stroke, making it a critical factor in illustrating the effectiveness of MI-BCI training for motor recovery [27,28,29]. Numerous studies have shown that oxygenation levels are a key indicator of this plasticity [27, 29]. To assess cortical effects, brain imaging techniques such as functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS), which evaluate the hemodynamic response to neural activity, can be employed [29,30,31,32,33,34,35,36,37]. However, real-time fMRI scanning during MI-BCI training presents significant challenges [37]. fNIRS, which has low sensitivity to body movements and is easily combined with EEG, is the optimal method for examining the effects of MI-BCI training on motor recovery [31,32,33,34,35,36]. Based on the findings from a novel work on an event-related design using fNIRS [34], we believe it is possible to investigate the effects of FP feedback on cortical activation during MI-BCI training using fNIRS.

In this study, we used a simultaneous EEG-fNIRS measurement system. While there is neurovascular coupling between EEG and fNIRS [34, 36, 38], fNIRS-based MI-BCI systems exhibit an inherent delay due to the nature of hemodynamic responses, whereas EEG is highly sensitive to electrical noise from robotic feedback. Therefore, we used an EEG to implement the MI-BCI system and the fNIRS to measure cortical activation during MI-BCI training and validated the results by comparing the effectiveness of two different cue-based MI-BCI systems—with and without FP feedback. We hypothesized that FP feedback significantly affects cortical activation during the MI-BCI training period. Furthermore, for quantitative comparisons, the MI-BCI system was compared with two conventional motor learning methods (passive and active training).

Methods

Participants

Twelve healthy volunteers (six men; mean age = 21.92 ± 3.34 years) and four people with a stroke (three men; mean age: 54.75 ± 9.03 years) participated in this study. The people with a stroke were recruited from Yeungnam University Medical Center, South Korea, based on the following characteristics: (1) hemiplegia caused by stroke, (2) age < 70 years, (3) normal cognitive function, and (4) able to follow simple verbal commands and freely communicate with researchers. Table 1 presents the characteristics of the people with a stroke. All the people with stroke were right-handed; whereas ten healthy participants were right-handed, while one was left-handed and another was ambidextrous. The healthy volunteers had no history of neurological or psychiatric disorders.

All volunteers received full instructions about the experiment and voluntarily agreed to participate. The study was approved by the Institutional Review Board of Daegu Gyeongbuk Institute of Science and Technology (approval number: DGIST-170721-HR-025-08).

Table 1 Characteristics of people with a stroke (n = 4)
Full size table

Study design

To examine the study hypothesis, we compared the results of two MI-BCI paradigms (using asynchronous MI-BCI systems) implemented under different feedback conditions, with the Raw paradigm incorporating FP feedback and the Match paradigm without FP feedback (Fig. 1). The cued paradigm was applied to define FP clearly. In this study, FP feedback was defined as feedback resulting from a positive classification by the asynchronous MI-BCI classifier during the rest period, when the green light (cue) was off [18, 19]. Conversely, true feedback was defined as feedback from a positive classification by the asynchronous MI-BCI classifier during the task period, when the green light (cue) was on. We compared Raw and Match paradigms with two conventional methods of motor learning (active and passive training). In active training, participants voluntarily performed physical movements without any external assistance, while passive training utilized a hand exoskeleton robot that provided full assistance to participants, guiding and facilitating their movements (Fig. 1).

Fig. 1
figure 1

Conceptual diagram of paradigms and comparison groups (PT: passive training, AT: active training, EEG: electroencephalogram, fNIRS: functional near infrared spectroscopy)

Full size image

Protocol

The experiment spanned two days. On day 1, motor execution and motor imagery tasks were sequentially performed to collect training data for implementing the MI-BCI (Fig. 2a). The instructed task involved performing a single open/close movement of the right hand (or the affected hand in patients with stroke) during the task. On the second day, four different paradigms (Raw paradigm, Match paradigm, active training, and passive training) were carried out; in all paradigms other than active training, movement feedback for the MI-BCI was provided through a customized hand exoskeleton robot, which provided hand opening and closing movements [39, 40]. For this experiment, we used a modified version of the MI-BCI described in previous works [18, 19]. The original MI-BCI had a two-phase classifier to minimize FP for rehabilitative BCI [18], but we modified the classifier by deleting the second phase because appropriate amount of FP is needed for this experiment. Note that all the other procedures for training and test were same to the previous works. The same experimental protocols were applied to both healthy participants and patients.

On both days, participants were seated in a chair with backrest which can fix their torso, with their right arm resting on the armrest throughout all sessions (Fig. 2b). Following a brief sound cue (lasting 0.25 s), a green light was displayed on the monitor for 2 s (or 4 s for people with stroke). Participants were specifically directed to perform the instructed task within the period when the green light was illuminated, and to relax once the green light had turned off, which lasts for 10 s (or 14 s for people with stroke). The interstimulus interval was set based on the previous studies to allow participants to sufficiently relax in the event-related design [31, 34, 41]. The sound cue for each paradigm was repeated 30 times, with 5 min of rest between each paradigm to minimize potential influence of fatigue. Any cognitive activities unrelated to the task were prevented through verbal instructions. The participants initially performed motor execution exercises to familiarize themselves with the kinesthetic sensation associated with the tasks. Following ample adaptation, passive training, Raw paradigm, and Match paradigm were randomly administered; active training was conducted last to ensure that the experience did not affect other paradigms by making the user’s MI substantially easier [34]. Patients with stroke who experienced difficulties with motor execution did not perform the task on day 1; for the same reason, they did not engage in active training on day 2 (Fig. 2a).

Data acquisition

EEG data were recorded using the Active Two system (Biosemi, Netherlands) using the sintered Ag/AgCl electrodes on both days. A total of 29 channel electrodes were positioned on the scalp near the motor cortex, following the international 10–20 system (Fig. 2c) [42]. In addition, fNIRS data were collected alongside the EEG signals using the LABNIRS (Shimadzu, Japan) on day 2. To ensure synchronization, the digital trigger signal from the fNIRS device was transmitted to the EEG device whenever the sound cue was presented. Participants wore a head cap for EEG-NIRS simultaneous measurement (FLASH; Shimadzu Co. Ltd., Japan) [43]. The optode probe set, comprising 14 sources and 14 detectors, was embedded within the EEG electrode cap, and the neighboring source and detector pairs were spaced approximately 3 cm apart, forming a total of 45 channels (Fig. 2c) [44].

Fig. 2
figure 2

Schematic representation of the experimental protocol. (a) Experimental protocol; yellow boxes indicate protocols for people with a stroke (b) Experimental setting (c) Diagrammatic illustration of the placement of the EEG electrodes and the fNIRS optodes [43]

Full size image

Data analysis

EEG data from day 1 was used to train the classifier for the MI-BCI system, while the data from day 2 was employed to test the classifier, following the same methodology as outlined in the previous study [18]. Additionally, EEG data from the Match paradigm for healthy participants on day 2 was analyzed to assess brain activation during task periods with and without robotic feedback (i.e., true feedback and no feedback), in order to verify whether the MI-BCI system was functioning as expected. The EEG signals with a frequency of 2048 Hz underwent preprocessing steps, including band-pass filtering between 8 and 30 Hz and removal of independent components identified through independent component analysis, to reject the electrical noise originating from the robot [45, 46]. After epoching the EEG data (− 5 to 5 s based on the cue), abnormal epochs (> 75 µV) were rejected for noise attenuation [47,48,49]. A time-frequency map was visualized as an event-related spectral perturbation (ERSP) map using the EEGLAB toolbox (Swartz Center for Computational Neuroscience, USA) [50,51,52]; the ERSP was normalized on the baseline from − 4 to − 2 s before the cue and was displayed from − 4 to 4 s across 4–30 Hz [53, 54]. For each healthy participant, we examined the ERSP within the channel placed on the contralateral primary motor cortex (C3 for right hand) in relation to the hand motor tasks [50, 51]. In addition, to investigate the channel distribution, the averaged ERSP from 8 to 30 Hz for 2 s after the cue was calculated in each channel. ERSP and channel distribution maps were compared between two kinds of task sets, task periods with true feedback and task periods without feedback (no feedback). For statistical analysis, the averaged ERSP from the C3 channel was compared between two task sets using the Wilcoxon signed-rank test. It should be noted that EEG data was analyzed to validate whether our MI-BCI was delivering the true feedback to the users when event-related desynchronization (ERD) was detected in motor-related frequency range and area.

fNIRS data from the four paradigms were compared to test our hypothesis about FP feedback. The Raw paradigm specifically enabled a more direct investigation of the influence of FP feedback on the user’s task periods because it included task trials right after getting feedback (Raw+) and task trials following no FP feedback in rest periods just before (Raw−) (Fig. 3). Consequently, Raw + and Raw − were extracted to explore the direct effect of FP feedback compared to the Match paradigm. The hypothesis was validated in the region of interest (ROI) channel; additionally, all channels were analyzed to determine the effect of FP feedback.

Fig. 3
figure 3

Conceptual diagram of different datasets for data analysis

Full size image

For fNIRS data analysis, we mainly used oxyhemoglobin (oxy-Hb) because it is known to be sensitive to task-related changes [55, 56]. NIRS_SPM (KAIST, South Korea), a MATLAB-based software package, was utilized for this purpose [57]. The hemodynamic data of oxy-Hb underwent preprocessing, including Gaussian smoothing (full width at a half-maximum of 2 s) and wavelet minimum description length detrending, to compensate for the noise and motion artifacts caused by the participant’s movements [58, 59]. Preprocessed hemodynamic data was segmented from − 2 to 12 s from the cue (for the data from patients, − 2 to 18 s from the cue). Baseline normalization was conducted to temporal hemodynamic response by subtracting the averaged value of the 2 s before the cue. and averaging were then applied to estimate the cortical hemodynamic response. We examined the hemodynamic responses of oxy-Hb in each channel and generated topographic distribution maps using the average hemodynamic responses within the vascular response (6–8 s after the cue) [34, 60]. Since we recruited patients with both right- and left-sided hemiplegia, the topographical distribution map for patients with left-side affected was flipped left and right to calculate the average [61]. The peak points for each paradigm were calculated in each channel and normalized by dividing them into the first motor execution task to reduce individual differences, as described in (1).

$$\:normalized\:peak=\frac{peak\:of\:motor\:learning\:method}{peak\:of\:motor\:execution}$$
(1)

The normalization was implemented only in healthy participants who performed motor execution tasks. For people with stroke, we cannot implement motor execution task for normalization. Since the patients’ peak values were not normalized, it was not feasible to make a direct quantitative comparison with the healthy participants’ normalized peak values.

For statistical analysis, a nonparametric Friedman test was employed to assess the ranks between parameters, followed by post hoc testing using the Wilcoxon signed-rank test as a post-hoc test [62, 63]. Most statistical analyses were conducted using IBM SPSS Statistics (version 25.0; IBM Corp, USA) [64]. After the post-hoc test for peak oxy-Hb concentration comparison, the false discovery rate (FDR) was controlled using the procedure introduced by [65], and FDR-corrected p-values were calculated with fixed lambda (0.5) in MATLAB. In the further analysis results, data from one outlier of healthy participants were excluded using the interquartile range method to ensure the accuracy of the overall tendency [24, 66, 67].

Results

Validation of the asynchronous MI-BCI system

Since asynchronous MI-BCI feedback was delivered to the user through a robot, we must ensure that the correct feedback is provided to support fNIRS results. Figure 4 presents the ERSP and topographic distribution maps for trials with and without true feedback used to verify whether the MI-BCI system was functioning as expected.

Trials with true feedback showed a stronger ERD of ROI channels compared with those without feedback in the mu and beta band (8–30 Hz) after the cue (p < 0.05). Similarly, in the topographic distribution, trials with true feedback showed a strong ERD pattern in the bilateral motor area compared to a weak ERD pattern for trials without feedback. The number of true feedback between Raw and Match paradigms was compared to guarantee a fair interparadigm comparison, and there was no between-group difference (p > 0.05).

Fig. 4
figure 4

Averaged ERSP map and topographic distribution maps across all the healthy participants (dB scale). True feedback indicated the trial sets with robotic feedback for task periods, whereas no feedback indicated the trial sets without robotic feedback for task periods

Full size image

Match vs. raw paradigms

To investigate the effect of FP feedback during MI-BCI training, the Raw and Match paradigms were compared with the conventional passive and active training motor learning methods in the ROI channel. Figure 5 illustrates the hemodynamics in the ROI channel for all participants.

Figure 5a depicts the extent to which a paradigm elicited the maximum hemodynamic response based on the primary motor execution task. For oxy-Hb, there were statistically significant differences among the four paradigms (p = 0.012); post hoc testing revealed significant differences between active and passive training (p = 0.019), active training and the Raw paradigm (p = 0.029), Match paradigm and passive training (p = 0.038), and Match paradigm and Raw paradigm (p = 0.038). In Fig. 5c and d, solid lines represent hemodynamic responses for all participants.

Among healthy volunteers, active training induced the strongest average activation, followed by the Match paradigm, Raw paradigm, and passive training (Fig. 5a). In contrast, for people with a stroke, the Match paradigm elicited the highest peak of oxy-Hb concentration, followed by the Raw paradigm, and passive training; however, there were no statistically significant differences between the paradigms (p = 0.174; Fig. 5b). Owing to the event-related design of this experiment, statistical analysis could be performed on a time sample between different paradigms. As shown in Fig. 5c, significant differences were observed for healthy volunteers between passive and active training, active training and the Raw paradigm, the Match and Raw paradigms, and the Match paradigm and passive training. For people with stroke (Fig. 5d), the Match paradigm showed statistically significant differences from the Raw paradigm and passive training.

Apart from the ROI channel, the hemodynamic response was calculated and compared for all other channels as well. In healthy participants, strong activation for oxy-Hb was observed near C3 in all paradigms (Fig. 6a and b). The overall cortical activation, including C3, was highest during active training, followed by the Match paradigm, the Raw paradigm, and passive training. Further statistical analysis was done to compare the hemodynamic response peaks across different paradigms. Figure 6c and d illustrate the topographical distribution of the statistical results obtained through the Friedman test for all paradigms and the post hoc analysis between the Raw and Match paradigms.

Fig. 5
figure 5

Results of the analysis of hemodynamic response between paradigms in the region of interest (ROI) channel for all study participants (PT: passive training, AT: active training). The data represent the mean values of each index, where the indices from 30 trials per participant were averaged. (a, b) Boxplots for the peak oxyhemoglobin (oxy-Hb) between each paradigm; *indicate statistically significant differences (p < 0.05) for the normalized peak oxy-Hb in (a) healthy volunteers and (b) patients with stroke. (c, d) Comparison of grand mean across all participants and statistical results for oxy-Hb concentrations between each two paradigms; for (c) healthy volunteers and (d) patients with stroke (red solid lines: Raw paradigm; purple solid lines: Match paradigm; black dash-dot lines: AT; brown dash-dot lines: PT). The gray-shaded area indicates a statistically significant difference in the concentrations between two different paradigms (*p < 0.05)

Full size image
Fig. 6
figure 6

Results of the analysis of hemodynamic responses in all channels. (a, b) Topographic distribution of the average oxyhemoglobin (oxy-Hb) for different paradigms in (a) healthy participants and (b) patients with stroke (PT: passive training, AT: active training) (c, d) Topographical distribution of results of statistical comparison between all paradigms and between Match and Raw paradigms for (c) healthy volunteers and (d) patients with stroke. The color axis indicates a − log10(p-value); channels marked with an asterisk have a p-value < 0.05

Full size image

Healthy volunteers exhibited significantly different oxy-Hb concentrations in bilateral brain areas among all paradigms (channels 3, 4, 7, 9, 11, 15, 19, 21, 25, 31, 34, 36, 37, 40, and 44; p < 0.05; Fig. 6c). Among these, three channels showed statistically significant differences in the Match and Raw paradigms (channels 3, 19, and 31; p < 0.05). Conversely, people with a stroke had statistically significant differences in the contralateral area among all paradigms (channels 14 and 20; p < 0.05). There were no statistically significant differences for any channel between the Match and Raw paradigms; however, in the motor area, people with a stroke had relatively high significance between hemodynamic responses of the Match and Raw paradigms compared to other channels (Fig. 6d).

Raw + vs. Raw

The trial sets Raw + and Raw– were extracted from the Raw paradigms and compared to investigate the effectiveness of FP feedback (Fig. 7). For healthy volunteers, the normalized peak oxy-Hb concentration showed that Raw + had significantly lower activation than the Match paradigm (p = 0.022), whereas there were no significant differences between the Raw − and Match paradigms (p = 0.099), and Raw − and Raw+ (p = 0.116; Fig. 7a). For people with a stroke, the peak oxy-Hb concentration was not significantly different between the Match paradigm and both Raw+ (p = 0.079) and Raw− (p = 0.084) (Fig. 7b). Further statistical analysis using time samples revealed that healthy volunteers had statistically significant differences between the Match paradigm and Raw+, and the early phase of Raw + and Raw− (Fig. 7c). In contrast, in people with stroke, the Match paradigm showed a higher hemodynamic response than the early phase Raw + and Raw − in periods after the peak (Fig. 7d). Raw − showed higher activation than Raw + in the early phase.

Figure 8 depicts the comparison results for all the channels between the Match paradigm and the Raw + and Raw–. For healthy volunteers, the Match and Raw– showed a higher hemodynamic response in the contralateral area, whereas Raw + showed weak overall activation compared to the Match paradigms and Raw– (Fig. 8a). Furthermore, Raw + showed significantly lower activation than the Match paradigm in channels 19, 21, 24, 35, and 41 (*p < 0.05), while there were no significantly different channels between the Raw– and Match paradigms (Fig. 8c). In contrast, people with stroke showed bilateral activation in the Raw– and Match paradigms, with stronger activation in the Match paradigm compared to Raw– (Fig. 8b). Raw + showed strong activation in the ipsilateral motor area compared to Raw– but not in the contralateral area. There were no significantly different channels between the Match paradigm and the two Raw paradigm trial sets (Fig. 8d).

Fig. 7
figure 7

Results of the analysis of hemodynamic response between the Raw+, Raw–, and Match paradigms in the region of interest (ROI) channel for study participants. The data represent the mean values of each index, calculated by averaging the indices from all trials per participant (a, b) Boxplots for normalized peak oxyhemoglobin (oxy-Hb) between the three paradigms for (a) healthy volunteers and (b) patients with stroke; *indicate statistically significant differences (p < 0.05). (c, d) Comparison of grand mean and time sample data for oxy-Hb concentrations between two paradigms for (c) healthy volunteers and (d) patients with stroke (dark red solid lines: Raw+; pink solid lines: Raw–; purple solid lines: Match paradigm). The gray-shaded area indicates a statistically significant difference in the concentrations between the two paradigms (*p < 0.05)

Full size image
Fig. 8
figure 8

Results of the comparison between the Raw+, Raw–, and Match paradigms for all channels. (a, b) Topographic distribution for different paradigms from (a) healthy volunteers and (b) patients with stroke (c, d) Topographical distribution of results of statistical analysis between the three paradigms for (c) healthy volunteers and (d) patients with stroke. The color axis indicates the -log10(p-value); the channels marked with an asterisk showed statistically significant differences (p < 0.05)

Full size image

Discussion

This study investigates whether FP feedback affects the user’s motor learning during rehabilitative MI-BCI training. We used asynchronous MI-BCI paradigms with and without FP feedback (Raw and Match paradigms), along with two motor learning paradigms of conventional rehabilitation training. Additionally, the trials after the rest periods with and without FP feedback (Raw + and Raw–) were extracted separately from the MI-BCI with FP feedback (Raw) to demonstrate the effect of FP feedback in the same paradigm and compare it with the MI-BCI paradigm without FP feedback (Match). To ensure adequate feedback of the asynchronous MI-BCI implemented in this study, we compared the EEG results in the task trials with true robot feedback and without robot feedback from Match paradigm. A mu and beta ERD pattern after the sound cue was shown in the ERSP map when the true feedback was provided, whereas a weak ERD pattern appeared in no feedback trials. According to previous studies, mu and beta ERD pattern in bilateral activation indicated the hand MI related signal [34, 50, 51]. Considering the activation occurring immediately after the sound cue to represent the activation originating from the participants’ MI, we believe that the asynchronous MI-BCI sufficiently reflected the participant’s movement intention.

To validate the effect of FP feedback on cortical plasticity induction, we compared the Raw and Match paradigms, of which the Raw paradigm tended to induce significantly lower cortical activation compared to active training. In contrast, no significant difference was observed between the Match paradigm and active training, implying that training without FP feedback had similar cortical activation as active training, which is notably advantageous for inducing neuroplasticity. In addition, the Match and Raw paradigms showed significant difference in normalized peak comparison, and the Match paradigm induced a significantly higher oxy-Hb concentration than the Raw paradigm in the 4–6 s period after the cue. Considering that Match paradigm showed higher peak in average than Raw paradigm, Match paradigm can induce earlier and more time-locked response compared to Raw paradigm. The people with stroke in this study also demonstrated a similar tendency for hemodynamic response in the ROI as healthy volunteers; however, we could not compare the two MI-BCI paradigms with active training since some of these participants had difficulties in performing their hand movements voluntarily. Even though there were no significant differences in the peak hemodynamic response between the Match paradigm, Raw paradigm, and passive training, the Match paradigm showed significantly higher hemodynamic response in the specific time range of 8–11 s after the cue compared to the Raw paradigm and passive training. These results suggest that the Match paradigm, which induced a more precisely time-locked response, also showed higher oxy-Hb concentration near the average peak points compared to the Raw paradigm. Unlike the results observed in healthy individuals, the peak of the hemodynamic response was not delayed in stroke patients, despite the delayed response observed in the healthy group. The absence of a delayed response in the Raw paradigm for stroke patients can likely be attributed to the difference in the number of FP feedback instances provided to each group, even though efforts were made to control for the number of FP feedback. This observation led us to perform a detailed analysis of the individual conditions within the Raw paradigm (Raw + vs. Raw–) to further investigate this phenomenon.

We also conducted a comprehensive analysis of the hemodynamic response across all channels, encompassing the C3 area corresponding to the left postcentral gyrus. Irrespective of the participant’s health status, the Match paradigm consistently elicited stronger hemodynamic responses near the contralateral motor area compared to the Raw paradigm (Fig. 6). In healthy volunteers, channel distribution maps for hemodynamic response from all the channels focused on the contralateral motor cortex, with the strongest in our designated ROI (i.e., C3; channel 21); however, there were some differences between the channels in different tasks. The Match paradigm showed significantly higher peak hemodynamic response in the channels corresponding to the frontal cortex (channel 3) and the ipsilateral motor cortex (channels 19 and 31) than the Raw paradigm. It means that even though the Match paradigm couldn’t induce significantly stronger activation of the contralateral motor area, it showed significantly higher cortical activity in some areas of the ipsilateral motor and frontal cortices compared to the Raw paradigm. For all channel analyses in people with a stroke, we could not obtain statistically significant differences between Raw and Match paradigms due to the small sample size; nevertheless, the pattern was similar to healthy volunteers, with significant differences for the motor area channels among passive training and the two MI-BCI paradigms. Specifically, the differences between the three paradigms were most notable in the left primary motor cortex (channels 14 and 20) near the left postcentral gyrus. Also, all patients showed a higher peak hemodynamic response in the contralateral primary motor cortex (channels 14 and 20) adjoining the contralateral postcentral gyrus during task performance in the Match paradigm compared to the Raw paradigm. Considering a history of insult to the brain on the contralateral postcentral gyrus, it is noteworthy that asynchronous MI-BCI without FP feedback promoted neuroplasticity during task periods more than the MI-BCI with FP feedback.

Further analysis of individual trial sets in the MI-BCI with FP feedback revealed a more direct effect of FP feedback in the same paradigm. In healthy volunteers, while the Match paradigm could not show significantly higher normalized peaks in the ROI compared to the Raw–, the normalized peaks were significantly higher than the Raw+ (Fig. 7a and c). Additionally, Raw– showed a significantly higher hemodynamic response than Raw + in the early phase. These results indicate that within the same paradigm, cortical activation is more induced during MI trials without FP feedback just before, compared to MI trials immediately after FP feedback. In terms of peak hemodynamic response in people with stroke, both the Match paradigm and each trial set from the Raw paradigm, Raw + and Raw–, were comparable results within the limitations of our small sample of four stroke patients. (Fig. 7b). However, the Match paradigm had a significantly higher hemodynamic response in a specific range compared to Raw+ (1–9 s after the cue), which was comparable to healthy volunteers (Fig. 7d). Also, Raw– had significantly different hemodynamic response in 5–7 s after the cue. In contrast to the results from healthy participants, Raw– demonstrated a significantly lower hemodynamic response compared to the Match paradigm during the 8–12 s window following the cue. Raw + showed slower cortical activation compared to the other conditions, Raw– and the Match paradigm, which is consistent with the results from healthy volunteers. However, unlike in healthy participants, FP feedback attenuated the hemodynamic response throughout all time periods of the Raw paradigm in stroke patients in this study.

From the viewpoint of overall channel distribution, Raw– was more similar to the Match paradigm, and Raw + showed more chaotic distribution compared to both Raw − and Match paradigms. In healthy volunteers, Raw − was similar to the Match paradigm in terms of distribution and activation level, and there were no different channels between the Raw– and Match paradigms. Alternatively, Raw + showed a more chaotic distribution with weaker activation compared to the Match paradigm. The increased oxy-Hb concentrations in the bilateral motor cortex (channels 19, 21, and 24), including the left postcentral gyrus (channel 21), and ipsilateral parietal-occipital area (channels 35 and 41) during the Match paradigm compared to Raw + highlight the distinct neurovascular responses linked to successful right hand open/close MI [34, 68,69,70]. Particularly noteworthy is the activation in the left postcentral gyrus (channel 21), implying a strong correlation between this region and the effective execution of the task [34, 68, 70]. The people with a stroke in this study also showed similar results, with the Match and Raw– encouraging cortical activation in bilateral motor areas and stronger activation in the contralateral motor area, whereas Raw + encouraged cortical activation in the ipsilateral motor area. Additionally, Raw + had differences in the contralateral area compared to the Match paradigm, whereas Raw– had differences in the ipsilateral brain area. Considering the contralateral representation of motor functions in the brain, activation of the left postcentral gyrus highlights a direct correlation with improvements in the right-hand MI performance during the Match paradigm [34, 68, 70]. Furthermore, when compared with Raw+, trials with FP feedback in the last rest period induced lower and slower hemodynamic responses than the Raw paradigm. Therefore, we can presume that the previous FP feedback interfered with the user’s motor learning and resulted in not only a low and slow hemodynamic response in the ROI but also a chaotic cortical activation in the overall distribution. We expect FP feedback would be a major disturbance to cortical plasticity induction during rehabilitative BCI; hence, the overall results highlight the importance of minimizing FP feedback to induce cortical plasticity in most MI-BCI users for rehabilitation purposes.

In the data analysis, we excluded one healthy outlier from the analysis by anomaly detection using the interquartile range [24, 66, 67]. The outlier volunteer showed higher activation induced by the Raw paradigm compared with the Match paradigm, particularly with Raw + than Raw–. This volunteer showed higher hemodynamic response for asynchronous MI-BCI paradigms, especially for Raw paradigm, compared to any other healthy participants. The volunteer, who had strong activation, may have been motivated by the FP feedback and attempted a stronger MI, suggesting that motivational tendencies can vary from person to person. Previous studies have shown that FP feedback does not always interfere with the BCI operation or induce wrong-directed cortical plasticity in all participants [22,23,24, 26]. Thus, FP feedback typically weakly interrupts users to induce cortical plasticity. This typical correlation, also observed in the present study, emphasizes the specific impact of paradigm design, especially the intentional avoidance of FP feedback, on targeted motor task performance. Nevertheless, we must consider individual differences in the effect of FP feedback to optimize rehabilitative MI-BCI training and enhance its impact. A detailed understanding of these correlations will offer valuable insights into the neural mechanisms underlying MI tasks and highlight the potential for optimizing rehabilitative BCI paradigms for enhanced motor learning and cortical plasticity induction. According to related theories, FP feedback can distort key principles of reinforcement learning and reward-based learning [71, 72], thereby impeding effective learning processes. From the perspective of reinforcement learning, accurate feedback is crucial for proper behavior modification and the development of appropriate neural circuits [71]. However, FP feedback introduces incorrect rewards, leading users to adopt inefficient motor imagery strategies. In reward-based learning, the congruence between expected and actual rewards plays a vital role; FP feedback disrupts this reward prediction error mechanism, which in turn negatively affects dopamine pathways involved in learning [72]. Repeated exposure to such erroneous feedback can diminish user trust in the system, further impairing cortical plasticity and overall learning performance. From this viewpoint, we believe minimizing the FP feedback is a typical solution to maximize cortical plasticity induction for most MI-BCI users; however, tailoring FP feedback to individual differences can enhance the effectiveness of rehabilitative BCI training for cortical plasticity induction.

In the present study, the experiments were carefully designed to verify the effects of FP feedback on the user’s MI during MI-BCI training. As we used MI-BCI training, the possible contaminants from the experimental setup need to be attenuated for valid comparison. Regarding EEG results, we performed additional pre-processing, such as independent component analysis and abnormal epoch rejection, to attenuate the effect of the contaminants [45,46,47,48]. Moreover, the use of fNIRS in our setup offered several advantages because of its robustness against electrical noise from robotic feedback and low sensitivity to body movements with pre-processing techniques [36, 58, 59, 73]. Since the oxygenation level obtained from fNIRS is an important index of cortical plasticity [27,28,29], our results could appropriately show the influence of MI-BCI training on cortical plasticity. Most fNIRS studies used a block design for their experiments, which is based on the accumulated response to repeated stimuli. However, because FP feedback in MI-BCI is an unpredictable event that can be modulated when repeated, the block design was not suitable for our study [30, 31, 34]. In contrast, an event-related design can demonstrate a single stimulus response and can investigate the feedback effect independently. Despite its advantages, only a few fNIRS studies have reported results based on an event-related design in MI-BCI owing to their low significance. We recently showed that an event-related design can be applied to investigate the cortical effects of MI-BCI and different motor learning methods [34]. Several studies have investigated the effects of FP feedback on MI-BCI training. While most of these studies have primarily focused on the impact of FP feedback on BCI performance, such as accuracy, or brain activation during FP feedback events (e.g., error-related potentials), a recent study has demonstrated that an asynchronous MI-BCI paradigm without FP feedback results in greater cortical excitability compared to one incorporating FP feedback. In contrast, the study did not examine the effects of FP feedback during MI-BCI training on users because it only investigated and compared corticomotor excitability pre- and post-intervention [26]. Moreover, there were differences in the existence of cues, as well as the FP feedback between the two asynchronous BCIs used, making it difficult to associate the difference between the two paradigms solely with FP feedback. In contrast, this study applied fNIRS with event-related design to investigate motor performance during the MI-BCI training with and without FP feedback, instead of comparing the motor performance pre- and post-intervention. Furthermore, as both asynchronous MI-BCI methods were designed in cue-based systems, the difference between the two methods in this study was only the existence of FP feedback. Hence, this study can investigate the effect of FP quite clearer than previous studies.

This study had several limitations. First, our event-related design was based on a fixed cue-based paradigm, which is an approach with a fixed interstimulus interval [41]. As fixed cues can be a predictable event for some participants, a random interstimulus interval would be beneficial in reducing the influence of cue prediction by the volunteer. Second, although we attempted to reduce individual differences by normalizing the peak hemodynamic responses, we could not strictly control the speed and force of MI [74]. Future studies may aim to better regulate these individual differences by providing proper verbal and visual instructions, thereby obtaining data with minimal participants-related variations. Furthermore, we believe our results for the people with stroke are an interesting finding but it could be concluded carefully. First, our participants with stroke have age difference compared to healthy participants in this study. Some researches have shown that the BCI proficiency during MI change with age, being easier to understand and use in younger people [75, 76]. Second, we recruited chronic stroke patients, who have lesser plasticity, in comparison to acute or subacute patients, which could be affected more from a FP feedback; hence, if we could recruit acute or subacute patients, the effect of FP feedback could be exaggerate compare to our results from chronic stroke patients. Third, normalization was restricted to healthy participants, thereby making it difficult to compare peak values between healthy participants and stroke patients. Although a direct comparison is not feasible, the grand mean of the hemodynamic response (Figs. 5c and d and 7c and d) allows for an indirect assessment of the differences in oxy-Hb concentration. In addition, people with stroke cannot effectively perform MI because they tend to forget their kinesthetic sensations long after the cerebrovascular accident, thereby causing diminished brain activation. To overcome this limitation, MI-enhancing methods, i.e. such as the rubber/virtual hand illusion approach [53, 77, 78], and/or motor-learning-enhancing methods, i.e. transcutaneous auricular vagus nerve stimulation [79], may help patients imagine their motor tasks to acquire more reliable and improved MI data. Lastly, further validation with a larger, more diverse population is needed. Future research with larger sample sizes should also account for handedness as a potential factor influencing functional performance, as some evidence suggests that neuroplasticity may be more readily induced in the dominant hand compared to the non-dominant hand, even in cases of stroke [80].

Conclusions

We found that the MI-BCI paradigm without FP feedback induced significantly higher brain activation, focusing on the contralateral motor area, for the user’s MI compared with the FP feedback paradigm. Overall results illustrated that FP feedback reduced cortical activation for users during MI-BCI training, suggesting a potential negative effect on cortical plasticity. Interestingly, these findings were not limited to healthy volunteers; similar patterns were observed in some patients who had experienced a hemorrhagic or ischemic stroke. This observation opens up new avenues for investigation, as it suggests the potential for reducing FP feedback to improve the effectiveness of rehabilitative MI-BCI training in stroke patients. While these findings were consistent across healthy participants and some patients with a stroke, the small sample size and the inclusion of only chronic stroke patients limit the generalizability of the results. Further investigations with larger and more diverse populations are needed to refine these approaches and validate their broader applicability.

Data availability

No datasets were generated or analysed during the current study.

Abbreviations

MI:

Motor imagery

BCI:

Brain computer interface

fNIRS:

Functional near infrared spectroscopy

FP:

False positive

EEG:

Electroencephalogram

oxy-Hb:

Oxyhemoglobin

ERD:

Event-related desynchronization

ERSP:

Event-related spectral perturbation

References

  1. Schaechter JD. Motor rehabilitation and brain plasticity after hemiparetic stroke. Prog Neurobiol. 2004;73(1):61–72.

    Article  PubMed  Google Scholar 

  2. James W. The principles of psychology; 1890.

  3. Jerzy, Konorski. Conditioned reflexes and neuron organization. CUP Archive; 1948.

  4. Wolpaw JR, Birbaumer N, McFarland DJ, Pfurtscheller G, Vaughan TM. Brain–computer interfaces for communication and control. Clin Neurophysiol. 2002;113(6):767–91.

    Article  PubMed  Google Scholar 

  5. Beets IA, Macé M, Meesen RL, Cuypers K, Levin O, Swinnen SP. Active versus passive training of a complex bimanual task: is prescriptive proprioceptive information sufficient for inducing motor learning? PLoS ONE. 2012;7(5).

  6. Perez-Ibarra JC, Siqueira AA, Silva-Couto MA, de Russo TL, Krebs HI. Adaptive impedance control applied to robot-aided neuro-rehabilitation of the ankle. IEEE Rob Autom Lett. 2018;4(2):185–92.

    Article  Google Scholar 

  7. Várkuti B, Guan C, Pan Y, et al. Resting state changes in functional connectivity correlate with movement recovery for BCI and Robot-Assisted Upper-Extremity training after stroke. Neurorehabilit Neural Repair. 2013;27(1):53–62. https://doiorg.publicaciones.saludcastillayleon.es/10.1177/1545968312445910.

    Article  Google Scholar 

  8. Bai ZF, Fong KNK, Zhang JJ, Chan J, Ting KH. Immediate and long-term effects of BCI-based rehabilitation of the upper extremity after stroke: a systematic review and meta-analysis. J Neuroeng Rehabil. 2020;17(1).

  9. Teo W-P, Chew E. Is motor-imagery brain-computer interface feasible in stroke rehabilitation? PM&R. 2014;6(8):723–8.

    Article  Google Scholar 

  10. Sitaram R, Ros T, Stoeckel L, et al. Closed-loop brain training: the science of neurofeedback. Nat Rev Neurosci. 2017;18(2):86–100.

    Article  CAS  PubMed  Google Scholar 

  11. Leamy DJ, Kocijan J, Domijan K, Duffin J, Roche RAP, Commins S et al. An exploration of EEG features during recovery following stroke - implications for BCI-mediated neurorehabilitation therapy. J Neuroeng Rehabil. 2014;11.

  12. Mane R, Chouhan T, Guan C. BCI for stroke rehabilitation: motor and beyond. J Neural Eng. 2020;17(4):041001.

    Article  PubMed  Google Scholar 

  13. Nojima I, Sugata H, Takeuchi H, Mima T. Brain–Computer interface training based on brain activity can induce motor recovery in patients with stroke: A Meta-Analysis. Neurorehabilit Neural Repair. 2022;36(2):83–96. https://doiorg.publicaciones.saludcastillayleon.es/10.1177/15459683211062895.

    Article  Google Scholar 

  14. Lotte F, Larrue F, Mühl C. Flaws in current human training protocols for spontaneous brain-computer interfaces: lessons learned from instructional design. Front Hum Neurosci. 2013;7:568.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Aricò P, Sciaraffa N, Babiloni F. Brain–computer interfaces: toward a daily life employment. Brain Sci. 2020;10(3):157. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/brainsci10030157

  16. An H-S, Kim J-W, Lee S-W. Design of an asynchronous brain-computer interface for control of a virtual avatar. IEEE; 2016.

  17. Nooh AA, Yunus J, Daud SM. A review of asynchronous electroencephalogram-based brain computer interface systems. 2011:55–9.

  18. Song M, Jeong H, Kim J, Jang S-H, Kim J. An EEG-based asynchronous MI-BCI system to reduce false positives with a small number of channels for neurorehabilitation: A pilot study. Front Neurorobotics. 2022;16:971547. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fnbot.2022.971547.

    Article  Google Scholar 

  19. Song M, Oh S, Jeong H, Kim J, Kim J. A novel movement intention detection method for neurorehabilitation brain-computer interface system. IEEE; 2018. pp. 1016–21.

  20. Grosse-Wentrup M, Mattia D, Oweiss K. Using brain–computer interfaces to induce neural plasticity and restore function. J Neural Eng. 2011;8(2):025004.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Zhou L, Tao X, He F, Zhou P, Qi H. Reducing false triggering caused by irrelevant mental activities in Brain-Computer interface based on motor imagery. IEEE J Biomedical Health Inf. 2021;25(9):3638–48. https://doiorg.publicaciones.saludcastillayleon.es/10.1109/jbhi.2021.3066610.

  22. Barbero Á, Grosse-Wentrup M. Biased feedback in brain-computer interfaces. J Neuroeng Rehabil. 2010;7(1):1–4.

    Article  Google Scholar 

  23. González-Franco M, Yuan P, Zhang D, Hong B, Gao S. Motor imagery based brain-computer interface: a study of the effect of positive and negative feedback. IEEE; 2011. pp. 6323–6.

  24. Alimardani M, Nishio S, Ishiguro H. Effect of biased feedback on motor imagery learning in BCI-teleoperation system. Front Syst Neurosci. 2014;8:52.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Faller J, Vidaurre C, Solis-Escalante T, Neuper C, Scherer R. Autocalibration and recurrent adaptation: towards a plug and play online ERD-BCI. IEEE Trans Neural Syst Rehabil Eng. 2012;20(3):313–9.

    Article  PubMed  Google Scholar 

  26. Niazi IK, Navid MS, Rashid U, et al. Associative cued asynchronous BCI induces cortical plasticity in stroke patients. Ann Clin Transl Neurol. 2022;9(5):722–33. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/I3.51551.

    Article  PubMed  PubMed Central  Google Scholar 

  27. Yang M, Yang Z, Yuan T, Feng W, Wang P. A systemic review of functional Near-Infrared spectroscopy for stroke: current application and future directions. Front Neurol. 2019;10:58. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fneur.2019.00058. Published 2019 Feb 5.

    Article  PubMed  PubMed Central  Google Scholar 

  28. Pillai JJ. Insights into adult postlesional Language cortical plasticity provided by cerebral blood oxygen level-dependent functional MR imaging. AJNR Am J Neuroradiol. 2010;31(6):990–6. https://doiorg.publicaciones.saludcastillayleon.es/10.3174/ajnr.A1896.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  29. Cramer SC, Bastings EP. Mapping clinically relevant plasticity after stroke. Neuropharmacology. 2000;39(5):842–51. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/s0028-3908(99)00258-0.

    Article  CAS  PubMed  Google Scholar 

  30. Tie Y, Suarez RO, Whalen S, Radmanesh A, Norton IH, Golby AJ. Comparison of blocked and event-related fMRI designs for pre-surgical Language mapping. NeuroImage. 2009;47:T107–15.

    Article  PubMed  Google Scholar 

  31. Ozawa T, Aihara T, Fujiwara Y et al. Detecting event-related motor activity using functional near-infrared spectroscopy. IEEE; 2013. pp. 1529–32.

  32. Maidan I, Nieuwhof F, Bernad-Elazari H, et al. The role of the frontal lobe in complex walking among patients with Parkinson’s disease and healthy older adults: an fNIRS study. Neurorehabilit Neural Repair. 2016;30(10):963–71.

    Article  Google Scholar 

  33. Oh S, Song M, Kim J. Validating attentive locomotion training using interactive treadmill: an fNIRS study. J Neuroeng Rehabil. 2018;15(1):1–11.

    Article  Google Scholar 

  34. Jeong H, Song M, Oh S, Kim J, Kim J. Toward comparison of cortical activation with different motor learning methods using event-related design: EEG-fNIRS study. IEEE; 2019. pp. 6339–42.

  35. Kim DH, Lee KD, Bulea TC, Park HS. Increasing motor cortex activation during grasping via novel robotic mirror hand therapy: a pilot fNIRS study. J Neuroeng Rehabil. 2022;19(1).

  36. Li R, Potter T, Huang W, Zhang Y. Enhancing performance of a hybrid EEG-fNIRS system using channel selection and early Temporal features. Front Hum Neurosci. 2017;11:462.

    Article  PubMed  PubMed Central  Google Scholar 

  37. Simon J, Fishbein P, Zhu L, Roberts M, Martin I. Functional magnetic resonance imaging-based brain computer interfaces. Neural Interf Eng Link Phys World Nerv Syst. 2020:17–47.

  38. Wang Z, Yang L, Zhou Y, Chen L, Gu B, Liu S, et al. Incorporating EEG and fNIRS patterns to evaluate cortical excitability and MI-BCI performance during motor training. IEEE Trans Neural Syst Rehabil Eng. 2023;31:2872–82.

    Article  PubMed  Google Scholar 

  39. Bae SJ, Jang SH, Seo JP, Chang PH. The Optimal Speed for Cortical activation of passive wrist movements performed by a rehabilitation robot: a functional NIRS study. Front Hum Neurosci. 2017;11:194. Published 2017 Apr 20. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fnhum.2017.00194

  40. Lee DJ, Bae SJ, Jang SH, Chang PH. Design of a clinically relevant upper-limb exoskeleton robot for stroke patients with spasticity. IEEE Int Conf Rehabil Robot. 2017;2017:622–7. https://doiorg.publicaciones.saludcastillayleon.es/10.1109/ICORR.2017.8009317.

    Article  PubMed  Google Scholar 

  41. Plichta MM, Heinzel S, Ehlis AC, Pauli P, Fallgatter AJ. Model-based analysis of rapid event-related functional near-infrared spectroscopy (NIRS) data: A parametric validation study. NeuroImage. 2007;35(2):625–34.

    Article  CAS  PubMed  Google Scholar 

  42. Cobb W, London GB, Gastaut H, Hess R Jr, Jung R, Magnus O, Terzian H. Report of the committee on methods of clinical examination in electroencephalography. Electroencephalogr Clin Neurophysiol. 1958;10(2):370–5.

    Article  Google Scholar 

  43. Takeuchi M et al. Brain cortical mapping by simultaneous recording of functional near infrared spectroscopy and electroencephalograms from the whole brain during right median nerve stimulation. Brain Topography. 2009;22(3):197–214.

  44. Gagnon L, Yücel MA, Dehaes M, et al. Quantification of the cortical contribution to the NIRS signal over the motor cortex using concurrent NIRS-fMRI measurements. NeuroImage. 2012;59(4):3933–40.

    Article  PubMed  Google Scholar 

  45. Meirovitch Y, Harris H, Dayan E, Arieli A, Flash T. Alpha and beta band event-related desynchronization reflects kinematic regularities. J Neurosci. 2015;35(4):1627–37.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Iriarte J, Urrestarazu E, Valencia M, et al. Independent component analysis as a tool to eliminate artifacts in EEG: a quantitative study. J Clin Neurophysiol. 2003;20(4):249–57.

    Article  PubMed  Google Scholar 

  47. Barkoukis TJ, Matheson JK, Ferber R, Doghramji K. Therapy in sleep medicine E-book. Elsevier Health Sciences; 2011.

  48. Foldvary-Schaefer N, Krishna J, Budur K. A case a week: sleep disorders from the Cleveland clinic. Oxford University Press; 2010.

  49. Kerick SE, Oie KS, McDowell K. Assessment of EEG signal quality in motion environments; 2009.

  50. Pfurtscheller G, Neuper C. Motor imagery activates primary sensorimotor area in humans. Neurosci Lett. 1997;239(2–3):65–8.

    Article  CAS  PubMed  Google Scholar 

  51. Pfurtscheller G, Da Silva FL. Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin Neurophysiol. 1999;110(11):1842–57.

    Article  CAS  PubMed  Google Scholar 

  52. Brunner C, Delorme A, Makeig S. Eeglab–an open source matlab toolbox for electrophysiological research. Biomedical Engineering/Biomedizinische Technik. 2013;58(SI–1–Track–G):000010151520134182.

    Google Scholar 

  53. Song M, Kim J. Motor imagery enhancement paradigm using moving rubber hand illusion system. IEEE; 2017. pp. 1146–9.

  54. Weersink JB, Gefferie SR, van Laar T, Maurits NM, de Jong BM. Pre-Movement Cortico-Muscular dynamics underlying improved Parkinson gait initiation after instructed arm swing. J Parkinsons Dis. 2020;10(4):1675–93.

    Article  PubMed  PubMed Central  Google Scholar 

  55. Miyai I, Tanabe HC, Sase I, Eda H, Oda I, Konishi I, Tsunazawa Y, Suzuki T, Yanagida T, Kubota K. Cortical mapping of gait in humans: a near-infrared spectroscopic topography study. NeuroImage. 2001;14:1186–92.

    Article  CAS  PubMed  Google Scholar 

  56. Strangman G, Franceschini MA, Boas DA. Factors affecting the accuracy of near-infrared spectroscopy concentration calculations for focal changes in oxygenation parameters. NeuroImage. 2003;18:865–79.

    Article  PubMed  Google Scholar 

  57. Ye JC, Tak S, Jang KE, Jung J, Jang J. NIRS-SPM: statistical parametric mapping for near-infrared spectroscopy. NeuroImage. 2009;44(2):428–47.

    Article  PubMed  Google Scholar 

  58. Friston K, Josephs O, Zarahn E, Holmes A, Rouquette S, Poline J-B. To smooth or not to smooth? Bias and efficiency in Fmri time-series analysis. NeuroImage. 2000;12(2):196–208.

    Article  CAS  PubMed  Google Scholar 

  59. Tak S, Yoon SJ, Jang J, Yoo K, Jeong Y, Ye JC. Quantitative analysis of hemodynamic and metabolic changes in subcortical vascular dementia using simultaneous near-infrared spectroscopy and fMRI measurements. NeuroImage. 2011;55(1):176–84.

    Article  PubMed  Google Scholar 

  60. Schroeter ML, Zysset S, von Cramon DY. Shortening intertrial intervals in event-related cognitive studies with near-infrared spectroscopy. NeuroImage. 2004;22(1):341–6.

    Article  PubMed  Google Scholar 

  61. Snyder DB, Schmit BD, Hyngstrom AS, Beardsley SA. Electroencephalography resting-state networks in people with stroke. Brain Behav. 2021;11(5).

  62. Sheldon MR, Fillyaw MJ, Thompson WD. The use and interpretation of the Friedman test in the analysis of ordinal-scale data in repeated measures designs. Physiotherapy Res Int. 1996;1(4):221–8.

    Article  CAS  Google Scholar 

  63. Woolson RF. Wilcoxon signed-rank test. Wiley Encyclopedia Clin Trials. 2007:1–3.

  64. IBM Corp. IBM SPSS statistics for windows. Armonk, NY: IBM Corp; 2017.

    Google Scholar 

  65. Storey JD. A direct approach to false discovery rates. J Roy Stat Soc B. 2002;64:479–98.

    Article  Google Scholar 

  66. Walfish S. A review of statistical outlier methods. Pharm Technol. 2006;30(11):82.

    Google Scholar 

  67. Moore DS, McCabe GP. Introduction to the practice of statistics. WH Freeman/Times Books/Henry Holt & Co; 1989.

  68. Shibuya K, Sadamoto T, Sato K, Moriyama M, Iwadate M. Quantification of delayed oxygenation in ipsilateral primary motor cortex compared with contralateral side during a unimanual dominant-hand motor task using near-infrared spectroscopy. Brain Res. 2008;1210:142–7.

    Article  CAS  PubMed  Google Scholar 

  69. Andrade J, Cecílio J, Simões M, et al. Separability of motor imagery of the self from interpretation of motor intentions of others at the single trial level: an EEG study. J Neuroeng Rehabil. 2017;14:63. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12984-017-0276-4.

    Article  PubMed  PubMed Central  Google Scholar 

  70. Schnitzler A, Salenius S, Salmelin R, Jousmaki V, Hari R. Involvement of primary motor cortex in motor imagery: a neuromagnetic study. NeuroImage. 1997;6(3):201–8.

    Article  CAS  PubMed  Google Scholar 

  71. Sutton RS, Barto AG. Reinforcement learning: an introduction. 2nd ed. Cambridge (MA): MIT Press; 2018.

    Google Scholar 

  72. Schultz W, Dayan P, Montague PR. A neural substrate of prediction and reward. Science. 1997;275(5306):1593–9.

    Article  CAS  PubMed  Google Scholar 

  73. Li R, Potter T, Wang J, et al. Cortical hemodynamic response and connectivity modulated by sub-threshold high-frequency repetitive transcranial magnetic stimulation. Front Hum Neurosci. 2019;13:90.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  74. Yin X, Xu B, Jiang C, et al. A hybrid BCI based on EEG and fNIRS signals improves the performance of decoding motor imagery of both force and speed of hand clenching. J Neural Eng. 2015;12(3):036004.

    Article  PubMed  Google Scholar 

  75. Wang L, Zhang Y, Zhang J, Sang L, Li P, Yan R, et al. Aging changes effective connectivity of motor networks during motor execution and motor imagery. Front Aging Neurosci. 2019;11:312.

    Article  PubMed  PubMed Central  Google Scholar 

  76. Hammer EM, Halder S, Blankertz B, Sannelli C, Dickhaus T, Kleih S, et al. Psychological predictors of SMR-BCI performance. Biol Psychol. 2012;89(1):80–6.

    Article  PubMed  Google Scholar 

  77. Song M, Kim J. A paradigm to enhance motor imagery using rubber hand illusion induced by visuo-tactile stimulus. IEEE Trans Neural Syst Rehabil Eng. 2019;27(3):477–86.

    Article  PubMed  Google Scholar 

  78. Jeong H, Kim J. Development of a guidance system for motor imagery enhancement using the virtual hand illusion. Sensors. 2021;21(6):2197.

    Article  PubMed  PubMed Central  Google Scholar 

  79. Chen L, Tang C, Wang Z, Zhang L, Gu B, Liu X, et al. Enhancing motor sequence learning via transcutaneous auricular vagus nerve stimulation (taVNS): an EEG study. IEEE J Biomedical Health Inf. 2023;28(3):1285–96.

    Article  Google Scholar 

  80. Harris JE, Eng JJ. Individuals with the dominant hand affected following stroke demonstrate less impairment than those with the nondominant hand affected. Neurorehabil Neural Repair. 2006;20(3):380–9.

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

The authors would like to thank all volunteers who volunteered for the study.

Funding

This study was supported by the Translational Research Program for Rehabilitation Robots (#NRCTR-EX23004), National Rehabilitation Center, Ministry of Health and Welfare, Korea, in part by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2022-NR069426), and by the Technology Innovation Program and Industrial Strategic Technology Development Program (20018256, Development of service robot technologies for cleaning a table).

Author information

Authors and Affiliations

  1. School of Mechanical Engineering, Sungkyunkwan University, Suwon, Gyeonggi-do, 16419, Republic of Korea

    Hojun Jeong & Jonghyun Kim

  2. Department of Psychology, University of British Columbia, Vancouver, V6T 1Z4, Canada

    Minsu Song

  3. Department of Physical Medicine and Rehabilitation, College of Medicine, Yeungnam University, Daegu, 42415, Republic of Korea

    Sung-Ho Jang

Authors
  1. Hojun Jeong
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Minsu Song
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Sung-Ho Jang
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Jonghyun Kim
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

JK supervised, conceptualized, and designed the study. JK also acquired the funding and provided the resources for the study. HJ and MS implemented the proposed system. HJ designed the experiments. MS, HJ, and SJ recruited volunteers and prepared IRB for the experiments. HJ and MS acquired the data. HJ processed and analyzed the data from the experiments, interpreted results from the data, and also drafted the first draft of the manuscript. JK finalized the manuscript. All authors read and revised the manuscript, and approved the final manuscript for publication.

Corresponding author

Correspondence to Jonghyun Kim.

Ethics declarations

Ethics approval and consent to participate

All volunteers signed an informed consent approved by the Daegu Gyeongbuk Institute of Science and Technology IRB prior to the experiment (DGIST-170721-HR-025-08).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Jeong, H., Song, M., Jang, SH. et al. Investigating the cortical effect of false positive feedback on motor learning in motor imagery based rehabilitative BCI training. J NeuroEngineering Rehabil 22, 61 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12984-025-01597-w

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12984-025-01597-w

Keywords

Journal of NeuroEngineering and Rehabilitation

ISSN: 1743-0003

Contact us