Stroke disables or kills several million people each year (Beal, 2010; Katan and Luft, 2018). The disability, especially the lower limb hemiplegia, highly impacts the lives of individuals (Rea et al., 2014; Katan and Luft, 2018). Rehabilitation therapy is vital for helping the survivor regain as much use of his/her lower limbs as possible. Traditional therapy acts on the distal physical level to indirectly influence the brain’s neural system, such as physical therapy and occupational therapy (Belda-Lois et al., 2011). However, these indirect therapies usually have poor efficiency. BCI directly detects and modulates brain activity (Abiri et al., 2019; Saha et al., 2021). There are two main BCI strategies to improve the lives of individuals among stroke patients, i.e., assistive BCI and rehabilitative BCI (Mane et al., 2020). In the last decade, rehabilitative BCI has emerged as one of the promising tools for lower-limb motor function restoration by adjusting neuronal plasticity in affected neural circuits (Mane et al., 2020; Romero-Laiseca et al., 2020; Bobrova et al., 2021). In the field of BCI, a lower limb exoskeleton control system based on steady state visual evoked potentials (SSVEP) is an efficient BCI system, such as achieved accuracies of 91.3 ± 5.73% and an information transfer rate (ITR) of 32.9 ± 9.13 bits/min (Kwak et al., 2015). However, recent study reviewed that the flickers used to encode the BCI command must be sufficiently intense to obtain a high-quality SSVEP, which will engage a relatively large portion of the visual resources. Moreover, such irritating visual stimuli are not only irrelevant to users’ subjective intent, but even disrupt users and make them feel uncomfortable. Therefore, even though this BCI system can work well to assist subjects with walking, its unnatural way of interaction is unacceptable to some users, which reduces its usefulness in practice (Xu et al., 2021).

LMI-BCI is a rehabilitative BCI and is natural. It detects the decrease/increase of power in the sensorimotor cortex to control external equipment and then reversely modulates brain activity by external equipment (Pfurtscheller and da Silva, 1999; Abiri et al., 2019; Zhang et al., 2021). Two types of MI can be distinguished: Kinesthetic Motor Imageries (KMI) and Visual Motor Imageries (VMI). A KMI can be described as the ability to imagine performing a movement without executing it, by imagining haptic sensations felt during the real movement (i.e., tactile, proprioceptive, and kinesthetic). In comparison, VMI represents a visualization of the corresponding movement incorporating the visual network (Rimbert et al., 2018; Yang et al., 2021). Nevertheless, there are three challenges for MI-BCI system: (1) EEG features is unstable in general; (2) It is difficult to detect lower-limb EEG features because the anatomical location of the lower-limb motor cortical area deep within the contralateral mesial cortex (Neuper and Pfurtscheller, 1996; Pfurtscheller et al., 1997; Pfurtscheller and da Silva, 1999; Boord et al., 2010); (3) Brain injury often inhibits the LMI ability after stroke, making the detection of EEG features harder (Takeda et al., 2007; Meyer et al., 2016; Park et al., 2016).To address the above challenges, several studies focused on enhancing LMI ability by Mirror Neuron system (MNS). MNS can transform visual sensory input of related behavior (e.g., Action Observation, AO) into one’s own brain impulse or motor output of behavior (Kohler et al., 2002; Rizzolatti and Sinigaglia, 2016; Tanaka, 2021). Li et al. (2015) enhanced LMI ability in an imagining playing football task by Action Observation and demonstrated more distinctive features in LMI with AO than without. As a kind of action observation, virtual reality (VR) plays an essential role in enhancing neural activity during motor imagery. These studies suggest that the use of immersive virtual reality headsets, with the illusion and embodiment they provide, can effectively improve motor imagery training and BCI performance (Choi et al., 2020; Ferrero et al., 2021). In the field of lower-limb rehabilitation, standing and sitting are two regular movements. The experimental results of Triana-Guzman et al. indicated that the classification of motor imagery and idle state provided a mean accuracy of 88.51 ± 1.43% and 85.29 ± 1.83% for the sit-to-stand and stand-to-sit transitions, respectively (Triana-Guzman et al., 2022). Additionally, Chaisaen et al. (2020) classified the AO/MI of sit-to-stand/stand-to-sit task, and the highest mean accuracy is 82.73 ± 2.54%.

In 2021, a latest study revealed that mimicking known biological control principles results in BCI performance that is closer to healthy human abilities (Flesher et al., 2021). Brain-muscle-kinesthesis loop is the known biological control principle (Daly and Wolpaw, 2008). Hemiplegia inhibits kinesthesis after stroke(De Vries and Mulder, 2007; Mane et al., 2020). kinesthetic illusion will complement the relevant loop. Neuroimaging studies have revealed that a relationship exists between the movement that has been imagined and the activation patterns of somatotopically organized motor and kinesthetic areas (Giannopulu and Mizutani, 2021). In the upper-limb MI-BCI study, KI induced by rubber hand can significantly amplify EEG features and provide better guidance to enhance upper-limb MI (Song and Kim, 2019). No previous study has investigated enhancing LMI-BCI performance by KI, although KI has important effects on BCI. This paper proposed an enhanced LMI paradigm by KI. Specifically, earlier studies proved that kinesthetic illusion where one feels muscle stretch could be induced by artificially vibrating the muscle spindle and tendon (Goodwin et al., 1972; Naito and Ehrsson, 2001; Naito, 2004; Tapin et al., 2021). Therefore, we induced KI by vibratory stimulation on Achilles tendon to enhance LMI ability during imagining kicking a football. We designed both research contents. The first research verified the feasibility of induced KI by vibratory stimulation and compared the brain activity during V-rest and rest task. The second research explored the effect of KI on LMI by comparing the LMI with and without KI (KI-LMI vs. no-LMI).

The rest of this paper is organized as follows. The experimental description and analysis methods are introduced in Section 2. Then, results are illustrated in Section 3 in terms of EEG analysis and BCI performance. Finally, discussion and conclusion are presented in Sections 4 and 5, respectively.

In the previous research, it was well known that brain injury often inhibits the LMI-ability after stroke, making the detection of EEG features harder (Takeda et al., 2007; Meyer et al., 2016; Park et al., 2016). The Higher BCI performance generally accelerates the patients’ recovery process (Guan, 2016). Therefore, enhancing LMI ability of stroke patients is critical to achieve rehabilitation. The latest study revealed that mimicking known biological control principles could improve BCI performance for subjects (Flesher et al., 2021). A natural way of interaction is acceptable to users, which increases its usefulness in practice (Xu et al., 2021). Previous studies used visual induction consistent with motor tasks to enhance LMI ability (Boord et al., 2010, Kitahara et al., 2017, Yu et al., 2018). Additionally, kinesthetic illusion (KI) is a type of proprioception that can complement the biological control loop. Two studies verified that a combination of upper-limb MI and KI feedback improved MI-BCI performance (Barsotti et al., 2018; Song and Kim, 2019). However, no previous study has investigated enhancing LMI ability by KI to design LMI-BCI paradigm. Therefore, based on the above researches, we designed two research contents to verify the feasibility of inducing KI via vibrating Achilles tendon, analyze the EEG features produced by KI, and explore whether KI could enhance LMI ability.

Long-term research found that KI, where one feels muscle stretch, could be induced by artificially vibrating the muscle spindle and tendon of the limbs (Naito, 2004; Tapin et al., 2021). Neuroimaging studies have revealed that KI and MI have similar activation patterns in brain regions (Naito and Ehrsson, 2001; Naito, 2004; Giannopulu and Mizutani, 2021). In addition, vibration is better transmitted to the muscle spindle if applied over the tendon, where it results in longitudinal stretch of the muscle fibers (Taylor et al., 2017). Vibration stimulation of tendon is an effective way to induce KI (Barsotti et al., 2018; Tapin et al., 2021). In this study, the most obvious tendon location (i.e., Achilles tendon) was selected as the vibration stimulation point. In medical contexts, using a MI questionnaire as an ability predictor tool could be one possible way to estimate BCI performance (Vasilyev et al., 2017; Rimbert et al., 2018). In the field of BCI, two classical work studied two different MI questionnaires. The first study concludes that the KMI scores obtained from the Kinesthetic and Visual Imagery Questionnaire could predict the performance of a MI-based BCI for able-bodied subjects (Vuckovic and Osuagwu, 2013). The second study found that the representation of subjective behaviors, calculated using the Motor Imagery Questionnaire Revised-Second Edition, and the control of the BCI seem to be strongly linked (Marchesotti et al., 2016). These studies have shown that MI questionnaires are probably the most accepted and validated methods to measure the subjective feelings of a subject. Therefore, our study conducted an oral questionnaire for all subjects to explore the results of the experiment. In this study, firstly, an oral questionnaire (Q1) of Table 2 verified that KI was felt subjectively by most subjects (13 out of 16 subjects) during vibrating Achilles tendon. Secondly, KI induced by vibratory stimulation enhanced the activation of the area around the Cz and CPz electrodes (i.e., around the sensorimotor cortex) during the V-rest task (see Figure 5). This result is consistent with previous research that activated cortex area (i.e., contralateral primary motor cortex) by KI was similar to the activated cortex area by MI (Rosenkranz and Rothwell, 2003; De Moraes Silva et al., 2015; Lopez et al., 2017; Giannopulu and Mizutani, 2021; Zhang et al., 2021). Similarly, a recent study showed that vibrotactile neurofeedback training on upper limbs can increase motor cortical excitability in hand muscle representation corresponding to a muscle engaged by the MI (Grigorev et al., 2021). Lastly, the classification performance of rest and V-rest shows a significant difference (see Figure 6), which suggests that activated the independent effects on the cerebral cortex by KI can be classified clearly. This experimental study verified that induced KI via vibrating Achilles tendon may be feasible, and provided a theoretical basis for applying KI to LMI-BCI paradigm. In order to better explore the difference and correlation between LMI and KI, we need to ensure the same duration of LMI and KI, so this study chose the same induction duration as LMI (i.e., 3.5 s). The above results verified the effectiveness of KI induction with a duration of 3.5 s. However, there are still some defects in this study. 3.5 s may not be the best KI induction duration. Future studies will investigate the optimal duration of KI induction.

Both studies demonstrated that a combination of upper-limb MI and KI feedback improved MI-BCI performance (Barsotti et al., 2018; Song and Kim, 2019). In our study, we used many analytical methods (e.g., ERD power, topographical distribution, oral questionnaire, and functional connectivity analysis) to explore whether KI enhances LMI ability. Previous studies explored functional connectivity of different brain regions based on Granger causality analysis. These studies found that the significant causal connection from the visual area to the motor area under the “visual–auditory context” and the “visual context” may indicate the information transmission process of the dorsal pathway evoked by the visual stimulus. Therefore, our study used iCOH algorithm to analyze the brain functional connectivity between different channels. The results in Figure 3 indicate that KI can enhance the information transmission process between brain regions (especially the sensorimotor area) during performing the LMI task, thus improving LMI ability. In previous mechanism studies, the anatomical location of the lower limbs motor cortex deep within the central cortex of the interhemispheric fissure, which corresponds to the Cz electrode of the 10/10 standard electrode distribution system (Neuper and Pfurtscheller, 1996; Jurcak et al., 2007; Boord et al., 2010). Therefore, in our study about LMI, the Cz electrode was selected to analyze the ERD trend in the 8–30 Hz band during LMI tasks. When adding KI to the LMI task (i.e., KI-LMI), the cortex activation around the Cz electrode is more significant (see Figure 5), and the Cz-ERD tendency was more obvious for most subjects (see Figure 7A). To evaluate the MI enhancements of the proposed paradigm, we compared the observed ERD for different LMI tasks. The peak ERD amplitudes (i.e., E _ min no / KI i ) were used to quantitatively evaluate the enhancement due to more immersive LMI from the paradigms (Song and Kim, 2019). Moreover, the ERD arrival time (i.e., T no / KI i ) was adopted as an indicator of temporal characteristics of ERD because it was used to evaluate the appropriateness of ERD for BCI, which is also an MI enhancement target for BCI system (Duann and Chiou, 2016; Song and Kim, 2019). In our study, T KI is significantly shorter during the KI-LMI task than the no-LMI task ( T KI < T no , see Figure 7C). Therefore, the ERD feature of KI-LMI was detected more quickly (i.e., better real-time; Duann and Chiou, 2016; Song and Kim, 2019), then the KI may speed up the online detection of LMI-BCI. The above significant ERD indexes are beneficial to EEG feature detection, thus improving the classification performance of BCI system. Referring to previous research methods (Storzer et al., 2016), this study plotted the grand-averaged relative power of Cz electrode (see Figure 8), in order to more clearly compare the effect of KI on different tasks. The results of Figure 8 and Figure 5 are consistent, and both show that relative power in the α band presents a more significant decline trend than that in the β band. This result indicated that the optimal frequency band of performing LMI is 8-13 Hz for some subjects in this study. And, above results suggest that KI enhances the relative power decline for all frequency bands during performing LMI and improves LMI ability of subjects.

In terms of BCI performance, KI significantly improved the average accuracy of classifying LMI and rest task (see Figure 9), achieving a 6.88% improvement. Especially, it is noteworthy that the average accuracy of all BCI-illiterate increased by 9.17%. Therefore, KI could enhance LMI, and enhanced LMI by KI may be more suitable for the BCI-illiterate (e.g., the partial stroke patients; Shu et al., 2017, 2019; Ren et al., 2020). Considering that if KI is applied to online LMI-BCI, vibration stimulation need to be performed simultaneously during LMI and rest tasks. Therefore, KI-LMI and V-rest are classified in our study to simulated online BCI performance. The results of this study show that KI is more conducive to the online classification of LMI and rest tasks for most subjects(see Figure 9, KI-LMI vs. V-rest, p < 0.05). Grand-averaged topographical distribution indicates that EEG features of KI-LMI and V-rest tasks have significant differences (i.e., More significant ERD feature were observed under the KI-LMI task). Therefore, better results can be obtained by classifying KI-LMI and V-rest tasks. Obviously, the high classification performance of KI-LMI/V-rest tasks is more suitable for the practical application of LMI-BCI paradigm. In addition, the results of Figure 7 and Figure 8 show that KI can enhance ERD feature and relative energy changes for most subjects. This result is more helpful for CSP to extract the two types of features with significant differences, so as to improve the classification performance (see Figure 9). In contrast, one subject (H3) thought that KI distracted his attention during the LMI task. Our research analyzed that higher vibration frequency might make this subject nervous because the Cz-ERD power of this subject was relatively unstable before the LMI task (see Figure 7A). The offline accuracy of the H3 is decreased by 1.25% (see Figure 9) due to the above reasons. Our research paradigm presented significant performance advantages for most subjects. Consequently, it may be more conducive to the training and rehabilitation of most stroke patients. Previous studies have found that activated areas of the cerebral cortex are biased toward the contralateral sensorimotor cortex during performing the LMI and visual observation task (Li et al., 2015; Yu et al., 2018). In this research (Tariq et al., 2020), LDA, SVM, and KNN model were used to classify bilateral foot LMI task. And, the single trial analysis and classification models resulted in high discrimination accuracies, i.e., maximum 83.4% for beta-ERS, 79.1% for beta-ERD, and 74.0% for mu-ERD. The above results are consistent with our study in Figure 5. Additionally, our statistical results in Table 4 proved that the KI significantly enhanced the cortex activation at the left and middle region of sensorimotor cortex (i.e., channels C3 and Cz) during the LMI task. This result is consistent with the topographical distribution (see Figure 5). Therefore, the KI may be conducive to distinguishing the left/right LMI task. This study provides a new idea to design the left/right LMI-BCI paradigm.

Previous studies have shown that 70 to 80 Hz is an effective way to induce KI, and high vibration frequencies cannot induce KI (Naito et al., 1999; Naito, 2004; Taylor et al., 2017). The results of our study indicate that 180 Hz can induce lower-limb KI, which is somewhat different from previous results. Different vibration stimulation locations will induce different KI effects. Previous studies explored the optimal stimulation frequency range through vibration of the upper limb tendon (Naito et al., 1999; Naito, 2004; Taylor et al., 2017; Barsotti et al., 2018), while this study explored the effect of inducing lower-limb KI through vibration of Achilles tendon. The cerebral cortex region corresponding to the upper-limb and lower-limb has a great difference in physiological structure (i.e., the anatomical location of the lower-limb motor cortical area is smaller and deep within the contralateral mesial cortex; the lower limb is located at the distal physical level) (Michael, 2016), which makes the optimal vibration frequency for inducing the upper-limb and lower-limb KI may have great difference. The experimental results of our study verified the feasibility of lower-limb KI induced by vibration of Achilles tendon at 180 Hz, which provides some reference for the research on inducing lower-limb KI with vibration stimulation. However, the vibration parameters (e.g., vibration frequency and vibration duration) were not optimized in this study. In the future, the influence of different vibration parameters on inducing lower-limb KI will be explored to produce the optimal induction effect. Studies have shown that MI-BCI is beneficial to improve the rehabilitation of stroke patients (Mane et al., 2020). Therefore, our research paradigm further improves the rehabilitation effect and lower-limb motor function of stroke patients by improving the performance of LMI-BCI. However, our study removed data with more artifacts before classification, which may affect the results of online classification in LMI-BCI. Future research will design online experiments to explore the effect of removing data with more artifacts on online application. Although the obtained results are statistically significant, it is acknowledged that the sample size of the current study is limited to 16 healthy subjects. Therefore, more healthy subjects and stroke patients should be recruited to further study the features in sensorimotor cortex caused by KI in different groups.

In this study, we integrated the induced KI by vibrating Achilles tendon with LMI tasks to enhance LMI ability. Our research 1 has verified that the KI was induced by vibratory stimulation on Achilles tendon, and found that the activated cortex area by the KI was similar to the activated cortex area by LMI. Research 2 demonstrated that the KI could enhance the activation of the sensorimotor cortex and improve the classification preformance of offline and simulated online LMI-BCI. Additionally, this study found that KI improved the ipsilateral difference of cerebral cortex. The LMI-BCI paradigm of this study is conducive to enhance LMI ability and provides a new idea to design a novel left/right LMI-BCI paradigm. This proposed approach enriched the content of LMI-BCI technology and accelerated its research progress.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the Ethics Committee of Xi’an Jiaotong University (Approval No. 2021–1577). The participants provided their written informed consent to participate in this study.

WW, BS, and JW contributed to conception and design of this study. WW and DW carried out collecting EEG data and statistical analysis. BS, GL, and JW provided many suggestions for this study. WW and GL worked together to complete the manuscript. All authors contributed to manuscript revision, read and approved the submitted version.

This work was supported by the open project of Henan Key Laboratory of Brain Science and Brain-Computer Interface Technology under Grant HNBBL230101 and HNBBL230207, and the Shenzhen Science and Technology Plan projects under Grant JSGG20201102145602006, and 2021 Basic Research Project of Shenzhen Science, Technology and Innovation Commission under Grant JCYJ20210324123414039.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.