In the course of this work, for the first time, to our knowledge, we studied motor rhythms via electroencephalography during video feedback therapy of the lower limb. During the task, in the alpha band, we found a comparable ERD in O and OI over precentral, central and parietal electrodes. In the OM, the ERD involved the same regions but was stronger over central electrodes. In the beta band, there was a gradation of ERD intensity in O, OI and OM over central electrodes. After the task, the ERS (beta rebound) changes were weak during the O task but were strong in the OI and OM (Cz) groups, with no differences between the OI and OM groups.

Recruitment of the mirror neuron system

In our study, in the O condition, we found topographies similar to those in the literature, with bilateral central and parietal desynchronization [37]. However, we observed no difference between the O and OI conditions, particularly in the parietal regions, while other works found greater alpha desynchronization in an observation task than during imagination alone [37]. Moreover, in the OM condition, the topography of desynchronization was bilateral and comparable to that in the O and OI conditions but with stronger desynchronization in the precentral and central regions; however, in a high-density EEG study, alpha desynchronization during a motor task without visual feedback led to centro-parietal recruitment only contralateral to movement [38]. These differences are probably due to the addition of an action observation task to the motor imagery and execution task. This phenomenon has already been shown in fMRI data of upper limb movements, where the addition of action observation to a motor task results in bilateral centro-parieto-occipital recruitment, whereas the motor task alone was much more lateralized [39]. We assumed that this phenomenon was linked to the recruitment of mirror neurons. Indeed, above the sensory-motor regions, alpha band desynchronization is linked to Mu desynchronization [40]. The Mu rhythm is a well-known EEG rhythm containing two independent components, one in the alpha band and one in the beta band, encoding different parameters related to motricity [23]. In the alpha band, Mu desynchronization is generally considered to indicate the activity of the mirror neuron system (MNS, 21); this activity is present not only during action observation, motor imagery, and motor execution but also in other more complex tasks recruiting mirror neurons, notably in social cognition [41]. Mu rhythm is used as a biomarker of MNS recruitment and is generated around the centro-parietal regions [42]. The observation of bilateral, centro-parietal Mu desynchronization in our O, OI and OM conditions suggested that the observation of action in video therapy in the lower limb results in recruitment of the mirror neuron system, with activation of a bilateral centro-parietal network, which we also observe in connectivity. This recruitment of the mirror neuron system has been described in mirror therapy, notably for the upper limb in healthy subjects but also for stroke patients [11, 43, 44]. For the lower limb VOT, MNS recruitment has not been documented to our knowledge.

In the OM condition, we observed significant desynchronization in the precentral and central regions compared with the O and OI conditions. It is difficult to conclude whether this greater desynchronization reflects an increase in the activity of mirror neurons or whether it is linked to other phenomena involved in motor planning or execution. Indeed, during a movement, the Mu rhythm is involved in the integration of the movement’s somatosensory parameters [23], which may enhance desynchronization in the OM task. Similarly, although there are many similarities between the observation, imagery and motor execution networks, we know that there are also some differences since, according to fMRI, only action execution systematically recruits the primary motor cortex [46]. Therefore, although it is logically expected that motor execution leads to desynchronizations of greater intensity than motor observation or imagery through greater cortical recruitment, we must remain cautious about the precise interpretation of the neurophysiological mechanism at the source of this greater alpha desynchronization in OM.

One of the pitfalls of interpreting Mu rhythms is contamination by alpha occipital activity during signal analysis, which is present in the same frequency band [41]. In this work, we did not observe desynchronization in the occipital regions, probably because the subjects were constantly focusing their visual attention on the screen. Additionally, desynchronizations seem to emerge strictly from the central and parietal regions, making the hypothesis of contamination of the observed centro-parietal desynchronization by the alpha-occipital less plausible.

Modulation of beta activity according to the motor task

In O, we observed a weak, left parietal beta ERD. In the OI, the beta ERD was bilateral and centroparietal, and in the OM, the beta ERD was intense in the bilateral centroparietal regions. In O and OI patients, statistical analysis revealed no difference in the intensity of desynchronization. Interestingly, in OM, there was an increase in desynchronization in the motor, premotor and parietal regions compared to O. A comparison of OM and OI showed only an increase in desynchronization with respect to the central and parietal regions in favour of OM.

Beta ERD corresponds to disinhibition of somato-motor neuronal populations [23]; for example, there is a correlation between motor response intensity and desynchronization strength in stroke patient populations [24]. For lower limb movements, the beta ERD is classically localized on the vertex opposite to the moving limb [38]. Thus, since the subjects are passive and do not perform any motor planning or motor execution tasks, we did not find any clear desynchronization in O. Additionally, in the OM condition, as the subjects performed a motor task, they recruited the premotor (motor planning) and parietal (integration of proprioceptive feedback) regions, where beta desynchronization was significantly increased compared to that in the O condition. Conversely, the OI and OM comparisons revealed differences only in centro-parietal regions and not in prefrontal regions. This is probably because OIs and OMs need to develop a motor pattern (premotor cortex), but they differ in the recruitment of the primary motor cortex and parietal regions, which are much stronger in OMs (execution of task and integration of proprioceptive feedback).

Analysis of connectivity during action also reflected this gradation between the O, OI and OM conditions, with a gradation in the strength of connectivity in relation to the Cz between conditions. When analyzing connectivity links, we observed in O a fronto-occipital network, probably linked to the activation of an action observation network, whereas in OI and OM, the recruitment is mainly centro-parietal contralateral to movement and may be linked to an action planning or execution network. However, one must remain cautious about these descriptive connectivity results since no statistical difference was found between conditions.

The need for a motor plan for post-movement validation

In our study, we observed a beta rebound emerging from the vertex, with a topography different from that of beta and alpha desynchronization; this rebound was absent in O but present in OI and OM. Beta rebounding corresponds to hypersynchrony in the beta band following movement [33, 48]. It originates in the precentral gyrus, more precisely, in the motor cortex [49, 50]. Initially, described as participating in the maintenance of an idling state in sensorimotor regions, its interpretation has been broadened [51]. It appears that beta rebound is modulated by motor validation phenomena and temporal integration of somatosensory and motor parameters [26]. For example, the observation of an erroneous movement can modulate beta rebound [25], as can the introduction of errors in a motor task [52]. It is possible that this modulation of beta rebound emerges following the detection of a mismatch between the forward model and the sensory afferents, allowing an update of the motor pattern [53]. Our results seem to confirm this hypothesis, as we observed one desynchronization in OI and OM but no desynchronization in O. This finding confirms that the vision of a movement, even from a 1st-person point of view, will trigger a significant beta rebound only if it is perceived as feedback for a motor pattern, which is either executed (OM) or simulated (OI).

No significant difference in rebound intensity was observed between the OI and OM groups, although there was a proprioceptive/visual mismatch in the OI group (the leg was motionless during the video in the OI group). Negative proprioceptive feedback is known to negatively modulate beta rebound [51]. Providing correct visual feedback in video therapy could therefore minimize the effects of incorrect proprioceptive feedback on rebound formation. This result is of particular interest in neurological rehabilitation, where beta rebound is known to be a marker for monitoring neurological recovery [24]. However, to test this hypothesis, it would be interesting to study beta rebound in patients with cerebellar stroke (alteration of the forward model [54, 55]) and patients with proprioceptive disorders.

Interestingly, in terms of connectivity after the task, we did not observe an increase in connectivity in the beta band over the Cz, suggesting a topographically localized phenomenon. However, we observed an increase in connectivity in the frontal, parietal and occipital regions in the alpha band, with a gradation between the O, OI and OM tasks. This increase in connectivity was associated with a decrease in prefrontal connectivity. To our knowledge, this change in connectivity has never been described previously. We know that there are differences in low-beta and high-beta band function during beta rebound [53], with involvement of the frontal and parietal cortexes in addition to the motor regions [56]. However, our observations were in the alpha band and may be indicative of another mechanism involved. Additionally, PLV connectivity is sensitive to volume conduction, making topographical analysis of such broad phenomena less robust. We must therefore remain cautious when interpreting these observations.

From healthy subjects to neurological patients

As this study was carried out on healthy subjects, we must remain cautious regarding the transposition of these results to a pathological population, with extremely different functional cortical dynamics and brain rhythms [57]. However, some general conclusions can be formulated.

Firstly, this work demonstrates a gradation of engagement between the tasks (O, OI, OM), that enables personalization of the therapy offered to patients according to their level of recovery. Indeed, we demonstrate that passive observation in lower limb video feedback therapy leads to sensori-motor cortical recruitment. Although weak, this activation could be of interest in the very early phases of neurological rehabilitation, for example to tackle maladaptive plasticity as demonstrated for upper limb video feedback therapy [58]. However, it seems necessary to work with motor imagery or motor execution tasks as soon as possible, to maximize cortical recruitment and to trigger motor validation phenomena such as beta rebound, which are absent in passive observation task.

Also, this work provides a physiological basis for understanding the specific effects of visual feedback. Indeed, many rehabilitation studies focus on mirror therapy, which combines two distinct tasks: (1) a systematic movement of the patient’s healthy hand or foot, and (2) an observation of the visual feedback on the mirror. Yet these two tasks have distinct neurological effects, whose interpretation often gets mixed up. Indeed, for the upper limb a bimanual movement may lead to bimanual facilitation [59], with an increase of motor cortex excitability [60] and a modulation of EEG rhythms and connectivity [61], which can bias the understanding of the specific effects of the visual feedback. Yet, the understanding of these specific effects is crucial to the personalization of the therapies. For example, we don’t know how the visual feedback is integrated in patients with neuro-visual disorders, or if proprioceptive disorders may conflict with correct the visual feedback, and thus decrease the sensori-motor cortex recruitment.

We also have few points of comparison between lower and upper limb video feedback therapy since most of the studies on visual feedback focus on upper limb rehabilitation. The main EEG rhythms dynamics (ERD, ERS) seem to be generally the same, with a different topography above for motor areas, with a gradation between O, OI, and OM cortical recruitment between tasks, and enhanced recruitment in action observation as compared to motor imagery alone [37]. Yet, many questions hypothesis demonstrated for upper limb rehabilitation remain to be tested for lower limb rehabilitation. For example, for upper limb rehabilitation, in an action observation and motor imagery condition (similar to our OI), it appears that the alpha band ERD is enhanced by the vision of own hand movement, as compared to a non-subjective movement [19], especially in a first person perspective [45]. Considering that action observation facilitates motor cortical activity after stroke [12], we believe that lower limb rehabilitation with action observation should always try to implement a first person subjective visual feedback, in order to maximize the cortical recruitment. This hypothesis remains however to be tested.

Our next step will be to study the specific effects of the visual feedback for stroke patients, regarding their lesion topography.

Limitations

There are several limitations to this work. First, we chose to add a two second interval between the end of the video and the beginning of the next epoch, in order to monitor the beta rebound. However, this interval appeared to be short since the beta rebound had not entirely returned to baseline at the start of the next epoch. For future work we will consider putting at least three seconds between the end of a video and the beginning of the next epoch.

A longer time at the beginning of the epochs should also be considered. Indeed, we observed in all the time–frequency maps an artifact at the beginning of the signal, that may have been caused by an event related potential due to a slight saccade in the video loop, maximal over occipital brain regions. Our baseline may also have been contaminated by some motor-preparation rhythms. Statistical comparison was performed between time–frequency maps with different baselines (500–1500 ms baseline versus 700–1200 ms baseline) and found no difference. Even if this did not change the overall significancy of our results, further studies should include at least three seconds of pause at the beginning of each epoch.

Finally, for connectivity we chose to perform a PLV analysis, which can be sensitive to volume conduction effects. We tried to mitigate these volume conduction effects by using a surface Laplacian filter [34]. However, PLI (Phase Lag Index) and wPLI (Weighted Phase Lag Index) connectivity analysis, which are insensitive to volume conduction artifacts, showed no interpretable results. Though this work presents original connectivity data during lower limb visual-feedback therapy tasks, interpretation of these connectivity changes between conditions must be cautious, especially considering the absence of statistically significant results. A specific study of connectivity changes using a high-density EEG headset and a source level analysis could prove interesting.