1 Introduction

Brain-computer Interface (BCI) provides an alternative communication and control channel for healthy or disabled users to interact with the external environment through brain activity alone [1]. Scalp-recorded EEG is comprised of a wide variety of oscillatory activities, such as delta ([2 4] Hz), theta ([4 8] Hz) alpha ([8 13] Hz), beta ([13 26] Hz), gamma ([30 70] Hz), even near to direct current (DC) component (< 1 Hz), which we call slow cortical potential (SCP) [2]. Among those oscillatory activities, sensorimotor rhythm frequency band ([8 to 30] Hz) has been shown to be correlated with movement or imagined movement intention [3], and also somatosensory attention [4]. By detection of the changes of band power, subjects’ motor or sensation intention can be reliably recognized by the BCI system and transferred to control external devices [5, 6]. However, there is a latency, usually in the order of seconds, between the motor imagery (MI) task and the generation of SMR patterns [7,8,9], making it difficult to develop a highly interactive BCI. This is especially the case in stroke neurorehabilitation [10, 11], when the delay between motor intention and the corresponding detection is required to render the Hebbian principle effective [12, 13].

Another signal modality, called movement-related cortical potential (MRCP) has been shown able to reflect the subject’s motor intentions within a few hundreds of milliseconds, thus it is critical for the afferent feedback to be timed to arrive in synchrony with the movement intention [14,15,16]. The MRCP based closed-loop BCI would provide a novel neuromodulation system to enhance the neuroplasticity more effectively. The performance of the MRCP based BCI system is a key factor influencing stroke recovery, as it is necessary to accurate single-trial detection of the MRCP waveform in real-time. However, the waveform of MRCP can vary substantially and therefore affect the performance of the brain switch. Factors influencing the waveform of MRCP need to be quantified, in part, by quantitatively analyze its trial by trial variability. Understanding the MRCP variability in a single-trial basis would provide a new way to enhance the corresponding BCI detection performance.

The MRCP consists of a Bereitschafts potential (BP) [17], followed by a motor potential (MP) [18] and a movement monitoring potential (MMP) [19, 20]. The BP consists of a slow decrease in EEG amplitude starting approximately 1500 ms prior to the onset of the movement, and is considered as a cortical representation of motor preparation. MRCP is one kind of a slow cortical potentials, within the frequency range of 0.05 to 3 Hz, but the waveform is related to movement – either real movement or imagined movement. The SCP can be self-regulated through neurofeedback training [21], i.e. the voluntary production of negative and positive potential shifts. There is an apparent overlap in frequency band between MRCP and SCP, both near the DC frequency range, hence the spontaneous SCP would be one of the factors influencing the MRCP waveform. The variability of single-trial waveform would be explained by the background SCP activity.

In this study, the spontaneous SCP will be topographically presented to subjects in real-time, and subjects will be instructed to perform self-paced real movement in the following three conditions: (1) without neurofeedback; (2) with negative SCP potential feedback; and (3) with positive SCP potential feedback. The variability of the MRCP waveform will be systematically investigated and compared between conditions.

2 Methods

Subjects

Four healthy subjects participated in the experiments (two female, all right handed, average age 22 ± 3.5 years), all were BCI naïve subjects. All participants have normal or corrected to normal vision, and none reported to be diagnosed with any neurological disorder. This study was approved by the Ethics Committee of the University of Waterloo, Waterloo, Canada. All participants signed an informed consent form before participation.

EEG Recording and EMG Recording

EEG signals were recorded using a 44-channel g.USBamp EEG system (g.tec, Austria). The electrodes were placed according to the extended 10/20 system, as shown in Fig. 1. The reference electrode was located on the right earlobe, and the ground electrode on the forehead. A hardware notch filter at 60 Hz was applied to the raw signals. The signals were digitally sampled at 1200 Hz.

Fig. 1.
figure 1

Topographic localization of EEG electrodes.

Full size image

One channel surface electromyography (EMG) was also recorded with the g.USBamp amplifier. EMG was acquired in monopolar montage from the tibialis anterior (TA) muscle with disposable electrodes. The electrode was placed on the mid-belly of the right TA muscle, while the reference and ground electrodes were placed on the bony surface of the right knee and right ankle, respectively.

Real-Time Spontaneous SCP Neurofeedback Interface

The EEG signals were continually filtered between 0.05 to 3 Hz using a second order Butterworth filter. The filtered SCP signals was then averaged within 100 ms windows. The potential across electrodes was presented to subjects on a color scalp topographic map, which was updated every 100 ms, as shown in Fig. 2.

Fig. 2.
figure 2

Graphic Demonstration of Experiment Paradigm. (A) No neurofeedback was pre-sented to subjects. (B) Neurofeedback was presented to subjects, when negative potential was presented around Cz vertex subjects performed a real movement. (C) Neurofeedback was pre-sented to subjects, when positive potential was presented around Cz vertex subjects performed a real movement. Color bar indicates the voltage value. (Color figure online)

Full size image

Experimental Procedure

The experiment paradigm was shown in Fig. 2. Subjects were required to perform movement tasks according to different conditions: (1) No neurofeedback, (2) Neurofeedback with Negative Potential, (3) Neurofeedback with Positive Potential.

The subject was seated on a comfortable armchair, with both forearms and hands resting on the armrests. The subjects were instructed to limit their eye, facial and arm movements. Every subject performed three runs of foot dorsiflexion task, with 20 trials per run. Before the start of every run, subjects were required to stay still and not make any foot movement in the first 10 s, which was used as baseline for the subsequent online EMG detection. During the first run, subjects performed self-paced foot dorsiflexion, with 5–8 s between each task. No neurofeedback was provided to the participants in this run, as shown in Fig. 2(A). During the second run, subjects received the real-time slow cortical potential maps (the whole scalp), and subjects were instructed to perform foot dorsiflexion when the center of the scalp (Cz) turns into blue color, which corresponded to negative potential as shown in Fig. 2(B), and rest for 5–8 s before next task. During the third run, the participants received the same neurofeedback as in the second run, but were instructed to perform the foot dorsiflexion task when the center of the scalp turns to red color, as shown in Fig. 2(C), which corresponded to positive potential. Subjects rested 4 to 10 min between runs.

Data Analysis

The Teager–Kaiser energy operator was used to detect movement onset from the EMG, which has been shown to be more accurate than using the amplitude of the surface EMG [22]. The TK value of EMG in the first ten second of every run was used as baseline for the subsequent EMG onset detection. If the TK value of the EMG surpass that of three times of the baseline value, then the onset of EMG was detected, which corresponded to the start of the foot dorsiflexion task. A band-pass filter from 0.05 to 3 Hz and the large Laplacian spatial filter centered at Cz were used to enhance the signal-noise ratio of Cz. The spatial and spectrally filtered virtual Cz was used in subsequent processing steps. 2.5 s before and 1.5 s after each movement onset of the preprocessed EEG was extracted for the waveform analysis.

3 Results and Discussion

MRCP Waveform in Different Conditions

Figure 3 illustrates the averaged MRCP waveforms ([0.05 3] Hz) under the three different conditions. Evidently, the peak negativity peak in run 2 was more pronounced, reaching up to −15 uv, while the peak negativity was only approximately −5 uv in run 3. The peak negativity was approximately −10 uv from run 1. The averaged MRCP essentially overlapped with the waveform from run 1.

Fig. 3.
figure 3

MRCP waveform in different neurofeedback conditions. Black line indicates the averaged waveform across all trials (20 × 3 = 60 trial); red dashed line indicates the waveform in Run 1; green dash-dotted line indicates the waveform in Run 2; blue dotted line indicates the waveform in Run 3. 0 s indicates the start of the task. (Color figure online)

Full size image

SCP Topography in different Time Period

Figure 4 illustrates the SCP maps in different time period ([0.05 3] Hz) in no neurofeedback condition (run 1), 0 corresponds to the onset on movement as detected by EMG. The potential with the corresponding time window was averaged, including time window of [−2.0 −1.8] s, [−1.5 −1.3] s, [−1.0 −0.8] s, [−0.5 −0.3] s, [−0.2 0] s, [0 0.2] s, [0.4 0.6] s, and [0.8 1] s.

Fig. 4.
figure 4

SCP topography in different time period in no neurofeedback condition. 0 corresponds to the onset of the EMG. Color bar indicates the voltage. (Color figure online)

Full size image

Figure 5 illustrates the SCP maps in different time period in neurofeedback with negative potentials (run 2). Figure 6 illustrates the SCP maps in different time period in neurofeedback with positive potentials (run 3).

Fig. 5.
figure 5

SCP topography in different time period in neurofeedback with negative potentials. 0 corresponds to the onset of the EMG. Color bar indicates the voltage. (Color figure online)

Full size image
Fig. 6.
figure 6

SCP topography in different time period in neurofeedback with positive potentials. 0 corresponds to the onset of the EMG. Color bar indicates the voltage. (Color figure online)

Full size image

From the time period between −2.0 s to −1.8 s, the potential distribution around the scalp were similar among the three conditions. While in the time period between −0.5 s to −0.3 s, there was a clear difference between the three conditions; there was a more negative potential around the Cz channel in the negative potential neurofeedback condition, and a more positive potential around the Cz channel in the positive potential neurofeedback condition. These differences were explicitly induced by our experiment protocol, and it is clear that they are influencing factors for the subsequent MRCP amplitude changes. At around 0 s, the resulting negative potential was more pronounced in run 2 than that in both run 1 and run 3.

In this preliminary neurofeedback study, the background SCP changes were found to be correlated to the variation of the MRCP waveform. To the best of our knowledge, this was the first study to address the waveform variation due to the background SCP changes. Through the proposed neurofeedback strategy, the MRCP amplitude can be enhanced or reduced. This confirmed our hypothesis that there would be an additive effect between the background SCP and self-induced MRCP signal in the resulting waveform.

4 Conclusion

The variation of the MRCP was influenced by the background spontaneous SCP activity. Real-time neurofeedback provided a new approach to quantify and affect the self-paced MRCP waveform, which would have a direct influence on the MRCP BCI performance.