Decoding motor imagery related to major mimetic muscles from electroencephalography

Decoding motor imagery related to major mimetic muscles from electroencephalography | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Decoding motor imagery related to major mimetic muscles from electroencephalography Haoran Sun, Mengkun Ding, Xiaofeng Shan, Shang Xie, Dongming Chang, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7908162/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Background Functional and aesthetic deficits in individuals with facial nerve paralysis (FNP) significantly impair their quality of life. By decoding motor intentions and controlling rehabilitation devices, motor imagery (MI)-based brain-computer interfaces can improve outcomes in people with peripheral paralysis. However, the electroencephalography (EEG) features underlying different facial MIs and their decodability remain unclear. This study aims to investigate the feasibility of achieving accurate decoding of facial MIs related to major mimetic muscles and the corresponding decoding strategies. Methods After comparing block and event-related designs to identify the appropriate paradigm for facial MIs, 20 healthy participants performed four types of facial MIs (eyebrow raising, eye closing, lip puckering and grinning) in two modalities: kinesthetic and visual, from which event-related desynchronization/synchronization (ERD/S) features were extracted using time-frequency analysis. A deep learning model integrating a temporal convolutional network with a spatial attention mechanism was then developed for both within-subject and cross-subject decoding, thereby identifying the contribution of each EEG channel. Finally, the model was further evaluated on EEG data from six individuals with FNP. Results Participants showed better performance in the block design, in which facial MIs induced significant ERD in the low-frequency band in the left prefrontal and right central-temporal regions, co-occurring with shorter and weaker ERS in higher frequencies. Regarding MI decoding in healthy participants, the model achieved the highest average accuracy of 85.17% in within-subject classification of kinesthetic MI, with EEG features from the left frontal and parietal regions contributing most to decoding. Combining these findings, the model obtained an average accuracy of 76.46% on patients’ data, with half the number of MI tasks and 25% fewer EEG channels. Conclusion This study demonstrated that major mimetic muscle-related MIs can be accurately recognized from EEG using deep learning, with a suitable decoding strategy involving within-subject decoding of kinesthetic MI collected through a block design. Brain-computer interfaces Motor imagery Electroencephalography Event-related desynchronization/synchronization Facial paralysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Background Facial nerve paralysis (FNP), with an annual incidence of about 20–40 per 100,000 [ 1 ], leads to motor disorders of facial muscles, resulting in a range of physiological and psychological sequelae in patients [ 2 ]. However, traditional treatment strategies, including pharmacotherapy, physical therapy, and surgery, focus mainly on the recovery of peripheral nerves and muscles [ 3 ]. Their effectiveness is limited by the slow and incomplete process of axonal regeneration [ 4 ], and they rarely address the maladaptive motor cortex reorganization that further hampers functional recovery [ 5 ]. One promising strategy to overcome these limitations is the application of motor imagery (MI)-based brain-computer interfaces (BCIs). As a communication system capable of decoding spontaneous cerebral activity to control external devices and provide feedback without involving peripheral nerves, BCIs have been proven to improve the prognosis of individuals with paralysis [ 6 – 8 ]. When combined with functional electrical stimulation (FES) devices, BCIs can establish a closed-loop system that helps improve motor control or restore motor function by inducing impaired facial movements [ 9 – 11 ]. However, the realization of such a BCI-FES system critically depends on the accurate decoding of facial MIs. Among available neural recording techniques, electroencephalography (EEG) remains the preferred choice for the establishment of BCIs due to its portability, affordability, and noninvasive nature, despite its limited spatial resolution and signal-to-noise ratio. Previous functional magnetic resonance imaging (fMRI) studies have shown that the somatotopy of different facial MIs overlaps within the small facial area of the primary motor cortex, suggesting inherent challenges for decoding [ 12 , 13 ]. Nevertheless, recent advances in machine learning have made it feasible to decode EEG signals associated with fine motor tasks, such as wrist rotation, fist clenching, and single-finger flexion/extension [ 14 – 16 ]. Therefore, it is imperative to investigate the potential of EEG in facial MI decoding to bridge the current gap in the development of BCIs for FNP rehabilitation. At the beginning of MI, motor execution (ME), or action observation (AO), a decrease predominantly in alpha (8–13 Hz) and beta (13–30 Hz) band power over the motor-related cortical areas can be observed, which is known as event-related desynchronization (ERD) and reflects cortical activation. Conversely, event-related synchronization (ERS) describes a rebound in beta band power following the termination of these activities [ 17 ]. Although no studies have yet reported features of facial MI-related ERD/ERS, several investigations have observed low-frequency ERD in the primary visual cortex and temporoparietal junction, often with a certain degree of right-sided dominance, during the observation of nonemotional facial expressions [ 18 – 19 ]. Regarding facial ME, one high-density EEG study reported that alpha and beta ERD associated with lip movement over the primary sensorimotor cortex and thalamus showed lower intensity but greater bilateral coordination compared with limb movements [ 20 ]. Another factor influencing decoding performance is the modality of MI. Kinesthetic motor imagery (KMI), the first-person mental rehearsal of motor sensations, and visual motor imagery (VMI), the visualization of movement from a first- or third-person perspective, activate overlapping but distinct neural networks that resemble ME and AO, respectively [ 21 – 23 ]. When asked to perform MI for different movements, individuals tend to use KMI and exhibit greater cerebral activity, especially for movements in which they have physical experience [ 24 ]. Therefore, most decoding algorithms are developed based on KMI datasets, making KMI-based BCIs generally exhibit better performance than those based on VMI. However, some users, especially those with motor disabilities, are unable to generate vivid MI to control BCIs even after training [ 25 ]. This phenomenon is also called "BCI illiteracy" and has a higher incidence in KMI than in VMI [ 26 ]. To the best of our knowledge, no study has systematically reported EEG features of different facial MIs or evaluated their decodability, leaving this area largely unexplored. The present study aims to address this gap by analyzing spatiotemporal EEG features of kinesthetic and visual facial MIs across major facial movements, and by assessing their decoding feasibility, thereby providing a foundation for facial MI-based BCIs in facial rehabilitation. Methods Participants 20 healthy participants (14 males and 6 females; aged 23.5 ± 1.9 years) and six individuals with FNP (5 males and 1 female; aged 44.5 ± 14.4 years) were recruited for this study. All participants had no history of neurological or psychiatric disorders and were novices in BCIs. The ability of motor imagery in all participants was evaluated using the Kinesthetic and Visual Imagery Questionnaire (KVIQ-20), with scores of 40.79 ± 4.33 for KMI and 41.42 ± 4.08 for VMI [27]. For the patients, we additionally collected their clinical characteristics and assessed their facial nerve function using the Facial Nerve Grading System 2.0 (FNGS 2.0) [28]. EEG recordings During the experiment, participants sat comfortably in a fixed chair without armrests and were asked to focus on the instructions shown on the screen (Apple Inc., CA, USA) approximately 80 cm in front of them. EEG signals were recorded using an amplifier (XLTEK EMU40EX, Natus, WI, USA) and an EEG cap (Tenocom, Shandong, China) with 19 recording electrodes positioned according to the international 10–20 system [29], a reference electrode placed in the parietal region, and a grounding electrode located in the frontal region. All electrode impedances were kept below 5 kΩ during recording and the sampling rate was set at 256 Hz. Experimental paradigms To determine the optimal decoding strategy for facial MIs, we divided the experiment into three steps. To enhance the performance of facial MIs in participants, we designed KMI tasks after ME and VMI tasks after AO. In all tasks, participants were specifically instructed to avoid emotional involvement as much as possible. (Fig. 1) First, to identify the more suitable paradigm for facial MIs, we compared block and event-related designs (illustrated in Fig. 1) using EEG data from seven of the 20 healthy participants. The total duration of both designs was controlled at approximately 90 minutes (including rest time) and the data were collected respectively with an interval of seven days. The event-related design consisted of 60 blocks, with rest intervals of five minutes between every 20 blocks, and each block comprised a single round of randomized and discrete four-type facial KMI trials: eyebrow raising, eye closing, lip puckering and grinning. At the beginning of each block, participants were asked to execute one of the facial movements in about three seconds according to the auditory prompts. To obtain the baseline for reference, participants were instructed to get relaxed as much as possible and focus on the white fixation cross presented on the screen without any movement for four seconds before and after the imagining phases. When performing the KMI, participants were asked to recall the sensation felt during the ME phase and to imagine performing the same facial movements without any muscle contraction. The start and end times of each KMI were annotated by participants by pressing the corresponding buttons. Before the next trial began, participants were provided a 10-second rest. The block design consisted of four blocks, with five-minute rest intervals in between, and each block comprised 24 trials, all of the same facial KMI type. Participants followed a similar sequence of "ME–relax–KMI–relax–rest" in each trial, while they were instructed to imagine facial movements continuously five times in a single KMI trial, namely, 120 times of KMI for each type of facial movement. The rest time between trials was 10 seconds. In the following experiment, data of KMI and VMI were collected following the better design, from a total of 20 healthy participants. In VMI tasks, the only difference was that participants were asked to perform VMI (imagining seeing facial movements without kinesthetic sensation) after observing a video of a facial movement displayed on the screen. Finally, we collected single-modality facial MI EEG data from patients and halved the number of tasks after determining the appropriate decoding strategy. EEG data preprocessing The entire process of EEG decoding was conducted with MATLAB (MathWorks, Natick, MA, USA) and the EEGLAB toolbox. Raw EEG data were first band-pass filtered between 0.5 and 50 Hz, with a 50 Hz notch filter applied to suppress power-line interference. Next, an infinite reference using the reference electrode standardization technique was employed to mitigate the impact of electrode positional deviations. EEG channels with variance or root mean square exceeding three standard deviations from the mean across all electrodes were automatically excluded. To minimize interference from adjacent physiological activities, we visually inspected the entire EEG data and used artifact subspace reconstruction to identify and remove components related to eye blink, muscle contractions, and other abnormal artifacts. Finally, the cleaned data were segmented into epochs according to trial markers. To account for slight variations in data length between events, all EEG segments were resampled to three seconds to ensure consistency across trials and prevent the model from capturing length-related artifacts. Feature extraction of ERD/ERS To systematically investigate brain activity during facial MIs, we extracted both temporal and spatial features of ERD/ERS and plotted the corresponding maps. According to the neurophysiological ERD/ERS characteristics described in relevant studies, the alpha and beta bands were selected as the bands of interest. For each epoch, a time-frequency analysis was performed on all EEG channels by means of short-time Fourier transform (STFT), with a 1-s Hanning window, and a frequency resolution of 1 Hz. For the channels contributing the most to MI classification, we additionally generated spectrograms for the frequency range 0.5–50 Hz with the same resolution. The time resolution was set to 20 ms. Finally, the ERD/ERS for each MI trial was calculated by: where represents the power spectral density (PSD) of frequency f and EEG channel c at time t , and is the average PSD during the baseline period (-2 to -1 s prior to cue onset). The resulting ERD/ERS of all trials from the same MI task were averaged per healthy participant. Negative values represent ERD, while positive values indicate ERS. The development of the deep learning model We designed a deep neural network consisting of three primary components, with the overall structure illustrated in Fig. 2: Temporal Convolutional Network (TCN): This module extracts time-domain features from EEG signals using parallel convolutional branches with kernels of different scales. One branch uses a large-scale convolution kernel (1×100, stride = 20) to capture long-term temporal dependencies, while the other uses a small-scale kernel (1×25, stride = 5) to extract fine-grained temporal patterns. The outputs from both branches are concatenated to form a rich multi-scale temporal representation. Spatial Attention Mechanism: To enhance the model’s sensitivity to task-relevant brain regions, we introduced a channel-wise attention mechanism. This module computes attention weights across EEG channels using global average pooling followed by two fully connected layers and a softmax activation. These weights are then used to reweight the channel features extracted by the convolutional layers, enabling the model to focus on the most discriminative features dynamically. After model training, we extracted the weight values of each channel from the attention layer parameters and visualized them on a brain topographic map. Domain-Adversarial Training Layer: Inter-subject variability poses a major challenge in EEG decoding. To improve cross-subject generalizability, we incorporated a domain-adversarial learning strategy using a gradient reversal layer (GRL). This component introduces an auxiliary domain classifier connected to the shared feature space. During training, the GRL encourages the shared features to be indistinguishable across source and target subjects by reversing gradients from the domain classifier. This promotes the learning of domain-invariant features, enabling better generalization across individuals. The complete model was trained using a combined loss function consisting of a classification loss (cross-entropy) and a domain adversarial loss, with a trade-off coefficient λ set to 0.1. Training was conducted using the Adam optimizer (learning rate = 0.001, batch size = 32, epochs = 100), with early stopping based on validation accuracy. (Fig. 2) Classification tasks The model’s decoding ability was tested in both within-subject and cross-subject scenarios, with classification accuracy as the primary evaluation metric. To ensure robustness, we performed eight-fold cross-validation for within-subject classification tasks and leave-one-participant-out cross-validation for cross-subject tasks. The data were randomly partitioned at the trial level, with careful attention paid to avoid data leakage between the training and testing sets. The results were visualized using confusion matrices, with row-wise normalization applied. Statistical analysis The differences in EEG amplitudes and model performance across conditions were analyzed using paired t-tests or Wilcoxon signed-rank tests, depending on the normality of the data distribution. All statistical analyses were conducted with SPSS 27.0 (IBM, Armonk, NY, USA), with a significance level set at 0.05. Results Comparison between block design and event-related design Figure 3 presents comparisons of alpha ERD/ERS amplitudes between the event-related design and the block design in different brain regions. Significant differences were observed for eye closing in the occipital region (-1.46 ± 2.88% vs. 10.35 ± 14.09%, p = 0.047) and lip puckering in the temporal region (− 10.60 ± 15.37% vs. 4.05 ± 10.05%, p = 0.047). Although other comparisons did not reach statistical significance, participants consistently exhibited a stronger ERD/ERS with less inter-subject variability in the block design. As shown by the confusion matrices in Fig. 4 , the model, when trained on data collected from the block design, achieved approximately 40% higher accuracy than when trained on the event-related design (84.12 ± 11.19% vs. 45.11 ± 8.38%, p = 0.016), and its accuracy did not significantly change even when the amount of training data was reduced (84.12 ± 11.19% vs. 85.29 ± 10.07%, p > 0.05). Because performing facial MI tasks continuously improved both participant performance and model accuracy, the block design was selected for the following experiment. The number of each task was kept at 120 to enhance the representativeness of ERD/ERS features, even though halving the training set size did not substantially affect model performance. (Fig. 3 ) (Fig. 4 ) ERD/ERS spatial features during different facial MIs Figure 5 illustrates the alpha and beta ERD/ERS features during four facial MIs in different modalities on topographical maps. These features were distributed in bilateral brain regions and demonstrated similarities across four facial MIs in both KMI and VMI modalities. In the alpha band, a strong KMI-related ERD was found in the left prefrontal, right central, and right frontal-temporal regions, while VMI exhibited a weaker and more localized ERD in the same regions. Similarly, the ERD, with lower amplitude, was mainly observed in the left prefrontal and right frontal-temporal regions for KMI and VMI in the beta band. In contrast, no significant ERS was observed across all conditions. (Fig. 5) Despite these overarching similarities, the amplitudes and extent of ERD vary slightly across facial MIs. In the KMI condition, lip puckering showed relatively broader distributions extending toward the right temporal region in the alpha band. Additionally, lip puckering and grinning exhibited increased beta ERD in the middle frontal and right temporal-parietal regions compared to eyebrow raising and eye closing. For the VMI, except for eyebrow raising, other facial MIs tended to exhibit increased brain activity in the middle frontal region. Table 1 shows the comparison of alpha and beta ERD amplitudes in their associated brain regions between the two MI modalities. In the right frontal-temporal region, the KMI of eye closing and lip puckering exhibited a stronger beta ERD compared to VMI, while no statistically significant differences were observed in other conditions. Table 1 Comparison of alpha and beta ERD/ERS amplitudes during different facial MIs in their associated brain regions between two MI modalities. Frequency band Brain region MI modality ERD/ERS amplitudes (%) during different facial MIs Eyebrow raising Eye closing Lip puckering Grinning Alpha Left prefrontal region KMI -11.89 ± 10.34 -9.95 ± 8.71 -10.85 ± 9.77 -12.33 ± 11.60 VMI -10.37 ± 10.74 -6.85 ± 6.54 -6.74 ± 5.56 -8.58 ± 7.29 p -value ns ns ns ns Right central region KMI -11.89 ± 11.53 -10.49 ± 10.01 -11.72 ± 10.98 -13.16 ± 12.11 VMI -10.27 ± 9.81 -7.45 ± 6.98 -8.34 ± 7.27 -9.52 ± 7.51 p -value ns ns ns ns Right frontal-temporal region KMI -10.82 ± 10.85 -9.54 ± 8.65 -10.94 ± 9.61 -11.47 ± 11.49 VMI -9.80 ± 9.49 -7.10 ± 6.26 -7.39 ± 6.13 -8.18 ± 6.48 p -value ns ns ns ns Beta Left prefrontal region KMI -5.44 ± 3.98 -5.87 ± 3.83 -5.27 ± 3.49 -5.50 ± 3.15 VMI -4.81 ± 3.91 -3.13 ± 4.56 -4.19 ± 3.60 -3.65 ± 3.92 p -value ns ns ns ns Right frontal-temporal region KMI -5.65 ± 4.48 -7.27 ± 4.47 -6.96 ± 5.05 -6.88 ± 5.15 VMI -5.37 ± 3.51 -3.71 ± 3.14 -4.29 ± 3.70 -4.16 ± 3.35 p -value ns 0.0209 0.0226 ns ERD/ERS amplitudes from channels within the same brain region were averaged. The left prefrontal region was represented by Fp1, F3, and F7, the right central region by Cz and C4 and the right frontal-temporal region by F8 and T4. (Table 1 ) Time-frequency analysis of EEG channels from active regions in topographical maps According to the topographical maps, we selected channels Fp1, Fp2, F3, F7, C4, and T4 for the time-frequency analysis, with the spectrograms from these channels are illustrated in Fig. 6 and Supplementary Fig. s1 . Across all conditions and channels, we observed an obvious power decrease in the lower frequency range (0–15 Hz), particularly within the delta, theta, and alpha band, with a lesser extent in the low beta band. This ERD started shortly after the onset of facial MIs (approximately 0.5–1 seconds), peaked at around 1.5–2 seconds, and extended to about 3 seconds. In parallel, the ERS was found in the higher beta and low gamma bands (15–35 Hz), often co-occurring or followed by the ERD in time. Notably, the duration of ERS was generally shorter than that of ERD, typically appearing as a transient increase in power that did not persist for the full duration of the MI period. We also found a slight ERS in the gamma band, specifically between 40 Hz and 50 Hz, which might be associated with residual muscular artifacts in the EEG data. In addition, we averaged the ERD amplitudes within 0–15 Hz and 0.5–3 seconds, and the ERS amplitudes within 15–35 Hz and 0.5–2 seconds for statistical comparisons shown in Fig. 7 , which are indicated as white dotted lines on the spectrograms. Except for the differences in ERD amplitude at channel F3 between eyebrow raising and eye closing, and in ERS amplitude at channel Fp2, no other statistically significant differences were observed across the remaining conditions. (Fig. 6 ) (Fig. 7 ) Within-subject and cross-subject decoding results of healthy participants Figure 8 shows the overall confusion matrix and classification accuracy of two classification scenarios for healthy participants, with statistical comparisons of recall and precision between the two MI modalities presented in Supplementary Table s1 . In both scenarios, our model demonstrated significantly superior recognition performance for KMI compared to VMI across all facial MIs. For the within-subject decoding, we achieved an average accuracy of 85.17 ± 8.54% for KMI and 70.80 ± 12.99% for VMI. Specifically, the model exhibited the best recognition capability for eyebrow raising, with an average recall of 91.03 ± 7.10% for KMI and 80.47 ± 10.54% for VMI. Performance on eye closing and grinning was also relatively high for KMI (82.92 ± 10.66% and 86.67 ± 10.90%, respectively), while lip puckering showed the lowest recall of 80.09 ± 11.82%. By comparison, the model's performance for VMI showed a noticeable drop, with recall decreasing by around 10–20% across facial MIs. The highest decoding capability was still observed for eyebrow raising (80.47 ± 10.54%), while grinning yielded the lowest performance (63.34 ± 12.58%). In the cross-subject scenario, we obtained an average accuracy of 66.25 ± 10.60% for KMI and 48.44 ± 8.89% for VMI, both showing a significant decline of approximately 30% compared to the within-subject classification. The model retained the relative tendency among KMI tasks, achieving higher accuracy for eyebrow raising and grinning than others. In contrast, the model showed consistently poor performance in all VMI tasks, with recall for none of the facial MIs exceeding 50%. (Fig. 8 ) Due to the limited generalizability of our model, we further analyzed the cross-subject classification results through the corresponding accuracy and loss curves demonstrated in Supplementary Fig. s2. In the KMI condition, the training accuracy steadily increased with low training loss and eventually exceeded 85%, suggesting effective learning within the training data. However, the validation accuracy plateaued around 60%, and the validation loss began to rise after approximately 75 epochs, indicating potential overfitting. For the VMI, the model exhibited slower convergence, with training accuracy peaking at only about 45% and no clear plateau observed. Validation accuracy fluctuated throughout the training process, remaining within the 30–35% range. Meanwhile, the training loss continued to decrease, while the validation loss began to rise after roughly 120 epochs, further indicating poor generalization in the VMI condition. Contributions of EEG channels to facial MI classification The classification weight values of EEG channels produced by the attention layer are presented in Supplementary Table s2 and visualized in Fig. 9 . From these topographical maps, it is evident that the EEG features from the prefrontal and left frontal regions contributed the most to classification in both conditions, while features from the right prefrontal region exhibited the least contribution. The EEG channels in other regions showed distinct classification weights for KMI and VMI. In KMI condition, channels located in the bilateral temporal and right parietal regions presented second-highest classification weights, whereas medial parietal and left occipital channels exhibited no significant contribution to classification. The top three contributing channels were Fp2, Fp1, and F3. By comparison, in VMI condition, the channels in the right central and right temporal region exhibited contributions comparable to those of the frontal channels, followed by channels in the left parietal region. In contrast, the medial frontal, medial central, left central, and left temporal channels showed low classification weights. The leading contributors were identified as F3, Fp1, and C4. (Fig. 9 ) Within-subject classification results and ERD/ERS topographic maps of patients Patients’ clinical characteristics are listed in Supplementary Table s3. All patients presented with unilateral FNP, with etiologies including parotidectomy, trauma, and parotid malignant tumor. The duration ranged from 2 to 80 weeks, with variations in the affected regions and severity of facial nerve injuries. As shown in Fig. 10 a, the average classification accuracy was 83.81 ± 8.97% among these patients, with no significant difference compared to healthy participants. The model maintained the best decoding capability for eyebrow raising, with an average recall of 94.57 ± 5.17%, outperforming other facial MIs by 10–15%. The patients' topographic map of channel classification weights is visualized in Fig. 10 b. EEG channels in the left frontal and parietal regions maintained higher classification weights, whereas those in the right frontal region continued to exhibit limited contributions to classification. Moreover, we selected representative topographic maps from patients during facial MI of the movements affected by their facial nerve injuries, which are shown in Fig. 11 , with the remaining maps provided in Supplementary Fig. s3. In contrast to the consistent ERD features observed in healthy participants, the patients exhibited significant variations in their EEG features. For severely affected facial movements, patients tended to show reduced activity in the left prefrontal region. Additionally, certain subjects demonstrated stronger ERS than healthy participants. (Fig. 10 ) (Fig. 11 ) Results of within-subject classification of KMI after model retraining Based on the decoding strategy and channel classification weights, we removed five EEG channels (F4, F7, F8, C3 and T3) whose weights were in the lower 50% for both healthy subjects and patients in within-subject classification of KMI, and retrained the model. The overall confusion matrix is shown in Fig. 11 . The average classification accuracy decreased to 79.16 ± 7.06% for healthy subjects and to 74.83 ± 8.25% for patients. (Fig. 12 ) Discussion This study presented the first exploration of the EEG spatiotemporal features during facial MIs related to major mimetic muscles and the corresponding decoding strategies, in both healthy participants and individuals with FNP. Our results demonstrated that the difference across various facial MIs could be effectively discriminated by deep learning, with the peak average classification accuracy reaching 85.17% for healthy participants and 83.81% for patients, achieved through the strategy of within-subject decoding of KMI collected using a block design. The classification weight analysis revealed that EEG features from the left frontal and parietal regions provided the most critical information for facial MI decoding. The model retraining with 25% fewer channels, further combining the analysis, resulted in only about 5–10% decrease in average accuracy. Based on these findings, we can accurately detect the facial movement intentions of patients and lay the foundation for simplifying EEG acquisition devices to reduce the latency of BCIs. Similarity in ERD/ERS spatiotemporal features between KMI and VMI As revealed by the topographic maps, facial MIs, similar to other types of MI, recruit the motor cortex for the programming and planning of movements, and inhibit the overt ME through the left prefrontal cortex. There is also a beta ERD observed at channel P4, suggesting that the parietal lobe, known for its role in sensory integration and motor attention processing [ 30 ], is involved in facial MIs. In addition, the ERD originating from the right temporal region may involve activity in the superior temporal gyrus and fusiform gyrus, which are considered signature cortical areas for facial MIs, as they are associated with the processing of facial tasks and expressions [ 31 , 32 ]. Another notable aspect is the similarity of facial MI-related ERD/ERS in spatial distribution between KMI and VMI. Although the two MI modalities recruit partially distinct neural pathways, several studies have shown that their brain activation patterns are similar, with only a temporal difference [ 33 ]. Moreover, when performing a MI task, the selection of MI modalities and the quality of MI appear to depend on a person’s prior motor experience, especially the task-specific experience [ 34 , 35 ]. For an unfamiliar and short-duration MI task—such as the facial MIs in this study—participants may struggle to fully follow the researcher's instructions and tend to use KMI rather than VMI. The pronounced activation of the left prefrontal cortex, along with the lack of activity in the occipital regions, supports this interpretation. Performance gaps in cross-subject decoding of facial MIs To identify the most suitable decoding strategy for facial MIs, we tested the model performance in different conditions. Compared to intra-subject classification, where the model achieved relatively high accuracies for both KMI (85.17%) and VMI (70.80%), performance in the cross-subject scenario declined markedly, with an average drop of approximately 30% for both modalities. The decrease was especially pronounced for VMI, where recall for all tasks fell below 50%. In contrast, KMI retained its relative performance pattern across tasks, particularly for eyebrow raising and grinning. This difference was further supported by the confusion matrices and training curves. In the VMI condition, the model exhibited unstable learning, with low validation accuracy and increasing validation loss over time, indicating poor generalization. For KMI, although the model showed more stable training dynamics, signs of overfitting still appeared at an early stage. This performance gap may result from greater variability in VMI-related EEG features across subjects. Although VMI often seems to involve imagining how an action appears visually from a third-person perspective [ 24 , 35 ], it can also involve first-person visualization [ 36 ]. This contrasts with KMI, which consistently employs the first-person perspective [ 37 ]. One possible improvement to the decoding method is to use EEG features in source space rather than in sensor space. By projecting EEG data onto anatomically aligned cortical regions and further integrating brain connectivity analysis, this method can provide more consistent features for decoding and improve the robustness of cross-subject classification [ 39 ]. Recent studies have shown that deep learning models trained on these features can achieve a significant improvement over classic MI-EEG datasets [ 39 , 40 ]. Considering the plug-and-play functionality of cross-subject classification models, it is worth investigating their potential in facial MIs decoding using source-space features. Optimizing electrode layout for low-latency facial MIs decoding In addition to classification accuracy, another crucial factor affecting BCIs performance in practical applications is response time. To promote cortical plasticity and restore movement symmetry, the response time thresholds for BCIs applied to individuals with FNP should be 400 ms and 50 ms, respectively [ 41 , 42 ]. Therefore, designing appropriate electrode layouts is essential to minimize the data volume to be processed and shorten the response time. Encouragingly, our findings suggest that an average accuracy of over 75% can be achieved using only 14 out of 19 electrodes of the 10–20 system, with the number and layout of electrodes further optimizable based on the results of classification weight distribution across EEG channels. The use of EEG features in source space might also help reduce delays. Although additional EEG source reconstruction is required, data with lower variability contributes to reduced complexity in deep learning models, thereby decreasing decoding time [ 43 , 44 ]. Due to the poor performance of single-electrode placement in key regions for EEG source decoding, it is necessary to appropriately increase electrode density [ 45 ]. Still, a careful balance must be struck between computational cost and classification accuracy. Additionally, although previous EEG decoding studies have indicated that low-frequency EEG features typically contribute significantly to distinguishing different tasks [ 15 , 17 ], several channels with high classification weights (e.g., Fp2, P3, and P4) did not exhibit strong ERD in the present study. This suggests that contributions of different channels may vary across frequency bands, with higher-frequency features playing a more important role in certain regions. Overall, a more fine-grained channel–frequency analysis, together with facial MI-specialized EEG acquisition devices and optimized decoding algorithms, will be essential for developing BCIs that achieve both high accuracy and low latency in patients. Limitations and future work There are some limitations in our study. The first one is the number of patients, especially those with long durations. Because peripheral paralysis affects the motor cortex in a way that correlates with motor disability and follows a progressive course, long-term follow-up is needed for patients with permanent FNP or complications such as facial synkinesis to assess potential impairment of BCI performance. In addition, the diversity of facial nerve injuries and the variability of EEG signals may limit the representativeness of ERD/ERS features in patients. Further investigation through neuroimaging methods with superior spatial resolution, such as fMRI and functional near-infrared spectroscopy, should be conducted to ensure the impact of FNP on motor cortex plasticity. These methods based on blood-oxygen-level-dependent signals can also assess brain activity during ME of facial movements without being affected by electromyographic interference, providing more insights into the impact of FNP on cortex plasticity. Conclusion The present study systematically investigated spatiotemporal ERD/ERS features during facial MIs related to major mimetic muscles for the first time. We found that facial MIs can induce marked low-frequency ERD in the left prefrontal and right central-frontal-temporal regions, while the concurrent ERS was predominantly observed in the higher beta and gamma bands. These features demonstrated similarities across four facial MIs in two MI modalities. Furthermore, facial MIs could be accurately decoded from EEG using deep learning, with signals from the left frontal and parietal regions showing notable contributions to decoding for both KMI and VMI. According to the model performance in different conditions, we determined the decoding strategy of within-subject classification of KMI-related EEG collected from the block design. The model finally achieved an average accuracy of 83.81% on the patient dataset, with training data that could be collected within an hour. Overall, our findings demonstrate the feasibility of achieving accurate decoding of facial MIs through their ERD/ERS spatiotemporal features, paving the way for the development of BCIs in individuals with FNP. Abbreviations AO Action observation BCI Brain-computer interface EEG Electroencephalography ERD Event-related desynchronization ERS Event-related synchronization FES Functional electrical stimulation fMRI Functional magnetic resonance imaging FNGS 2.0 Facial Nerve Grading System 2.0 FNP Facial nerve paralysis GRL Gradient reversal layer KMI Kinesthetic motor imagery KVIQ-20 Kinesthetic and Visual Imagery Questionnaire ME Motor execution MI Motor imagery TCN Temporal convolutional network VMI Visual motor imagery Declarations Ethics approval and consent to participate This study was approved by the Ethics Committee of Peking University School and Hospital of Stomatology (reference number PKUSSIRB-202387067) and all participants provided written informed consent prior to their participation. Consent for publication Consent for publication of individual data has been obtained in writing from all participants of this study. Competing interests The authors declare no competing interests. Supplementary Information Supplementary material 1: Fig. s1 . Spectrograms from partial EEG channels in the KMI and VMI modalities. Table s1 . Comparison of performance metrics between KMI and VMI for different facial MIs in two classification scenarios. Fig. s2. The accuracy and loss curves for KMI and VMI cross-subject classification. Table s2. EEG channels’ classification weights under different conditions. Table s3. Characteristics of the patients. Fig. s3. Patients’ ERD/ERS topographical maps during different facial MIs. Funding This study was supported by the Capital Health Research and Development of Special (2022–2-4102) and the Clinical Research Foundation of Peking University School and Hospital of Stomatology (PKUSS-2023CRF102). Author Contribution The study was was conceptualized and designed by H.S., M.D., N.Z., and Z.C.. Methodology was developed by H.S., M.D. and Z.C.. Data collection was performed by H.S.. The model was developed by D.C.. Data analysis and visualizations were carried out by H.S. and D.C., and N.Z. interpreted the results. The original draft was written by H.S., while M.D., X.S., S.X., D.C., N.Z., and Z.C. reviewed and edited the the manuscript. All authors have read and approved the final version of the manuscript. Acknowledgement We gratefully thank the assistance of Xin Liu and Meiyuan Sun from the Department of Neurology at Beijing Zhongguancun Hospital in the EEG recordings for this research. Data Availability The datasets of the present study are available from the corresponding author upon reasonable request. References Peitersen E. Bell's palsy: the spontaneous course of 2,500 peripheral facial nerve palsies of different etiologies. Acta Otolaryngol Suppl. 2002;(549):4–30. Saadi R, Shokri T, Schaefer E, Hollenbeak C, Lighthall JG. Depression Rates After Facial Paralysis. Ann Plast Surg. 2019;83(2):190–4. https://doi:10.1097/SAP.0000000000001908 . Nakano H, Fujiwara T, Tsujimoto Y, Morishima N, Kasahara T, Ameya M, et al. Physical therapy for peripheral facial palsy: A systematic review and meta-analysis. Auris Nasus Larynx. 2024;51(1):154–60. https://doi.org/10.1016/j.anl.2023.04.007 . Sunderland S. A classification of peripheral nerve injuries producing loss of function. Brain. 1951;74(4):491–516. Li C, Liu SY, Pi W, Zhang PX. Cortical plasticity and nerve regeneration after peripheral nerve injury. Neural Regen Res. 2021;16(8):1518–23. https://doi.org/10.4103/1673-5374.303008 . Baniqued PDE, Stanyer EC, Awais M, Alazmani A, Jackson AE, Mon-Williams MA, et al. Brain–computer interface robotics for hand rehabilitation after stroke: a systematic review. J Neuroeng Rehabil. 2021;18(1):15. https://doi.org/10.1186/s12984-021-00820-8 . Lorach H, Galvez A, Spagnolo V, Martel F, Karakas S, Intering N, et al. Walking naturally after spinal cord injury using a brain–spine interface. Nature. 2023;618(7963):126–33. https://doi.org/10.1038/s41586-023-06094-5 . Zhang M, Li C, Liu S-Y, Zhang F-S, Zhang P-X. An electroencephalography-based human-machine interface combined with contralateral C7 transfer in the treatment of brachial plexus injury. Neural Regeneration Res. 2022;17(12):2600–5. https://doi.org/10.4103/1673-5374.335838 . Canny E, Vansteensel MJ, van der Salm SMA, Müller-Putz GR, Berezutskaya J. Boosting brain-computer interfaces with functional electrical stimulation: potential applications in people with locked-in syndrome. J Neuroeng Rehabil. 2023;20(1):157. https://doi.org/10.1186/s12984-023-01272-y . Ilves M, Lylykangas J, Rantanen V, Mäkelä E, Vehkaoja A, Verho J, et al. Facial muscle activations by functional electrical stimulation. Biomed Signal Process Control. 2019;48:248–54. https://doi.org/10.1016/j.bspc.2018.10.015 . Mäkelä E, Venesvirta H, Ilves M, Lylykangas J, Rantanen V, Ylä-Kotola T, et al. Facial muscle reanimation by transcutaneous electrical stimulation for peripheral facial nerve palsy. J Med Eng Technol. 2019;43(3):155–64. https://doi.org/10.1080/03091902.2019.1637470 . Soliman RS, Lee S, Eun S, Mohamed AZ, Lee J, Lee E, et al. Brain correlates to facial motor imagery and its somatotopy in the primary motor cortex. NeuroReport. 2017;28(5):285–91. https://doi.org/10.1097/wnr.0000000000000758 . Makary MM, Eun S, Park K. Greater corticostriatal activation associated with facial motor imagery compared with motor execution: a functional MRI study. NeuroReport. 2017;28(10):610–7. https://doi.org/10.1097/wnr.0000000000000809 . Ofner P, Schwarz A, Pereira J, Müller-Putz GR. Upper limb movements can be decoded from the time-domain of low-frequency EEG. PLoS ONE. 2017;12(8):e0182578. https://doi.org/10.1371/journal.pone.0182578 . Xu B, Wang Y, Deng L, Wu C, Zhang W, Li H, et al. Decoding Hand Movement Types and Kinematic Information From Electroencephalogram. IEEE Trans Neural Syst Rehabil Eng. 2021;29:1744–55. https://doi.org/10.1109/tnsre.2021.3106897 . Sun Q, Merino EC, Yang L, Van Hulle MM. Unraveling EEG correlates of unimanual finger movements: insights from non-repetitive flexion and extension tasks. J Neuroeng Rehabil. 2024;21(1):228. https://doi.org/10.1186/s12984-024-01533-4 . Pfurtscheller G, Lopes da Silva FH. Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin Neurophysiol. 1999;110(11):1842–57. https://doi.org/10.1016/s1388-2457(99)00141-8 . Aktürk T, de Graaf TA, Abra Y, Şahoğlu-Göktaş S, Özkan D, Kula A, et al. Event-related EEG oscillatory responses elicited by dynamic facial expression. Biomed Eng Online. 2021;20(1):41. https://doi.org/10.1186/s12938-021-00882-8 . Rayson H, Bonaiuto JJ, Ferrari PF, Murray L. Mu desynchronization during observation and execution of facial expressions in 30-month-old children. Dev Cogn Neurosci. 2016;19:279–87. https://doi.org/10.1016/j.dcn.2016.05.003 . Zhao M, Marino M, Samogin J, Swinnen SP, Mantini D. Hand, foot and lip representations in primary sensorimotor cortex: a high-density electroencephalography study. Sci Rep. 2019;9(1):19464. https://doi.org/10.1038/s41598-019-55369-3 . Féry YA. Differentiating visual and kinesthetic imagery in mental practice. Can J Exp Psychol. 2003;57(1):1–10. https://doi.org/10.1037/h0087408 . Hardwick RM, Caspers S, Eickhoff SB, Swinnen SP. Neural correlates of action: Comparing meta-analyses of imagery, observation, and execution. Neurosci Biobehavioral Reviews. 2018;94:31–44. https://doi.org/10.1016/j.neubiorev.2018.08.003 . Farabbi A, Figueiredo P, Ghiringhelli F, Mainardi L, Sanches JM, Moreno P, et al. Investigating the impact of visual perspective in a motor imagery-based brain-robot interaction: A pilot study with healthy participants. Front Neuroergonomics. 2023;4–2023. https://doi.org/10.3389/fnrgo.2023.1080794 . Kraeutner SN, Eppler SN, Stratas A, Boe SG. Generate, maintain, manipulate? Exploring the multidimensional nature of motor imagery. Psychol Sport Exerc. 2020;48:101673. https://doi.org/10.1016/j.psychsport.2020.101673 . Elashmawi WH, Ayman A, Antoun M, Mohamed H, Mohamed SE, Amr H, et al. A Comprehensive Review on Brain–Computer Interface (BCI)-Based Machine and Deep Learning Algorithms for Stroke Rehabilitation. Appl Sci. 2024;14(14):6347. https://doi.org/10.3390/app14146347 . Yang C, Chen Z, Wang S, Zhang Z, Kong L, Chen X. The Impact of Visual and Kinesthetic Motor Imagery on Mental Fatigue and Classification Performance in Untrained Participants. Int J Human–Computer Interact. 2025;41(8):4594–608. https://doi.org/10.1080/10447318.2024.2352224 . Malouin F, Richards CL, Jackson PL, Lafleur MF, Durand A, Doyon J. The Kinesthetic and Visual Imagery Questionnaire (KVIQ) for assessing motor imagery in persons with physical disabilities: a reliability and construct validity study. J Neurol Phys Ther. 2007;31(1):20–9. https://doi.org/10.1097/01.npt.0000260567.24122.64 . Vrabec JT, Backous DD, Djalilian HR, Gidley PW, Leonetti JP, Marzo SJ, et al. Facial Nerve Grading System 2.0. Otolaryngol Head Neck Surg. 2009;140(4):445–50. https://doi.org/10.1016/j.otohns.2008.12.031 . Herwig U, Satrapi P, Schönfeldt-Lecuona C. Using the international 10–20 EEG system for positioning of transcranial magnetic stimulation. Brain Topogr. 2003;16(2):95–9. https://doi.org/10.1023/b:brat.0000006333.93597.9d . Souza-Couto D, Bretas R, Aversi-Ferreira TA. Neuropsychology of the parietal lobe: Luria’s and contemporary conceptions. Front NeuroSci. 2023;17–2023. https://doi.org/10.3389/fnins.2023.1226226 . Reisch LM, Wegrzyn M, Mielke M, Mehlmann A, Woermann FG, Bien CG, et al. Face processing and efficient recognition of facial expressions are impaired following right but not left anteromedial temporal lobe resections: Behavioral and fMRI evidence. Neuropsychologia. 2022;174:108335. https://doi.org/10.1016/j.neuropsychologia.2022.108335 . Herlin B, Navarro V, Dupont S. The temporal pole: From anatomy to function—A literature appraisal. J Chem Neuroanat. 2021;113:101925. https://doi.org/10.1016/j.jchemneu.2021.101925 . Guillot A, Collet C, Nguyen VA, Malouin F, Richards C, Doyon J. Brain activity during visual versus kinesthetic imagery: An fMRI study. Hum Brain Mapp. 2009;30(7):2157–72. https://doi.org/10.1002/hbm.20658 . Kraeutner SN, Stratas A, McArthur JL, Helmick CA, Westwood DA, Boe SG. Neural and Behavioral Outcomes Differ Following Equivalent Bouts of Motor Imagery or Physical Practice. J Cogn Neurosci. 2020;32(8):1590–606. https://doi.org/10.1162/jocn_a_01575 . Xu X, Fan X, Dong J, Zhang X, Song Z, Bai D, et al. Enhancing motor imagery in the third-person perspective by manipulating sense of body ownership with virtual reality. Eur J Neurosci. 2024;60(7):5750–63. https://doi.org/10.1111/ejn.16515 . Farabbi A, Figueiredo P, Ghiringhelli F, Mainardi L, Sanches JM, Moreno P, et al. Investigating the impact of visual perspective in a motor imagery-based brain-robot interaction: A pilot study with healthy participants. Front Neuroergonomics. 2023;4–2023. https://doi.org/10.3389/fnrgo.2023.1080794 . Guilbert J, Fernandez J, Molina M, Morin M-F, Alamargot D. Imagining handwriting movements in a usual or unusual position: effect of posture congruency on visual and kinesthetic motor imagery. Psychol Res. 2021;85(6):2237–47. https://doi.org/10.1007/s00426-020-01399-w . Brusini L, Stival F, Setti F, Menegatti E, Menegaz G, Storti SF. A Systematic Review on Motor-Imagery Brain-Connectivity-Based Computer Interfaces. IEEE Trans Human-Machine Syst. 2021;51(6):725–33. https://doi.org/10.1109/THMS.2021.3115094 . Kaviri SM, Vinjamuri R. Decoding motor execution and motor imagery from EEG with deep learning and source localization. Biomedical Eng Adv. 2025;9:100156. https://doi.org/10.1016/j.bea.2025.100156 . Ma S, Zhang DA, Cross-Attention. -Based Class Alignment Network for Cross-Subject EEG Classification in a Heterogeneous Space. Sensors. 2024;24(21):7080. https://doi.org/10.3390/s24217080 . Xu R, Jiang N, Mrachacz-Kersting N, Lin C, Asín Prieto G, Moreno JC, et al. A closed-loop brain-computer interface triggering an active ankle-foot orthosis for inducing cortical neural plasticity. IEEE Trans Biomed Eng. 2014;61(7):2092–101. https://doi.org/10.1109/tbme.2014.2313867 . Hohman MH, Kim SW, Heller ES, Frigerio A, Heaton JT, Hadlock TA. Determining the threshold for asymmetry detection in facial expressions. Laryngoscope. 2014;124(4):860–5. https://doi.org/10.1002/lary.24331 . Fang T, Wang J, Mu W, Song Z, Zhang X, Zhan G, et al. Noninvasive neuroimaging and spatial filter transform enable ultra low delay motor imagery EEG decoding. J Neural Eng. 2022;19(6). https://doi.org/10.1088/1741-2552/aca82d . Liu T, Li B, Zhang C, Chen P, Zhao W, Yan B. Real-Time Classification of Motor Imagery Using Dynamic Window-Level Granger Causality Analysis of fMRI Data. Brain Sci. 2023;13(10). https://doi.org/10.3390/brainsci13101406 . Delavari F, Santaniello S. Role of Scalp EEG Brain Connectivity in Motor Imagery Decoding for BCI Applications. Annu Int Conf IEEE Eng Med Biol Soc. 2024;2024:1–4. https://doi.org/10.1109/embc53108.2024.10781532 . Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial1.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 13 Dec, 2025 Reviews received at journal 08 Dec, 2025 Reviewers agreed at journal 17 Nov, 2025 Reviews received at journal 14 Nov, 2025 Reviewers agreed at journal 30 Oct, 2025 Reviewers invited by journal 29 Oct, 2025 Editor assigned by journal 22 Oct, 2025 Submission checks completed at journal 22 Oct, 2025 First submitted to journal 20 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7908162","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":541980181,"identity":"0d024f03-ea22-4c7e-b571-0595d71a6347","order_by":0,"name":"Haoran Sun","email":"","orcid":"","institution":"Peking University School and Hospital of Stomatology","correspondingAuthor":false,"prefix":"","firstName":"Haoran","middleName":"","lastName":"Sun","suffix":""},{"id":541980182,"identity":"4ddc51bd-9b5a-4592-8857-7c94ebe85dd5","order_by":1,"name":"Mengkun Ding","email":"","orcid":"","institution":"Peking University School and Hospital of Stomatology","correspondingAuthor":false,"prefix":"","firstName":"Mengkun","middleName":"","lastName":"Ding","suffix":""},{"id":541980185,"identity":"a1f619d8-5a61-49a4-bde8-5aca5e5034b9","order_by":2,"name":"Xiaofeng Shan","email":"","orcid":"","institution":"Peking University School and Hospital of Stomatology","correspondingAuthor":false,"prefix":"","firstName":"Xiaofeng","middleName":"","lastName":"Shan","suffix":""},{"id":541980187,"identity":"0adb377b-567b-4d11-82ca-4c5abab663e3","order_by":3,"name":"Shang Xie","email":"","orcid":"","institution":"Peking University School and Hospital of Stomatology","correspondingAuthor":false,"prefix":"","firstName":"Shang","middleName":"","lastName":"Xie","suffix":""},{"id":541980189,"identity":"085791be-b606-49f0-b313-6a4da96293eb","order_by":4,"name":"Dongming Chang","email":"","orcid":"","institution":"Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Dongming","middleName":"","lastName":"Chang","suffix":""},{"id":541980190,"identity":"16ce87cf-e251-487a-84d0-222fc9ef9ab2","order_by":5,"name":"Nianming Zuo","email":"","orcid":"","institution":"Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Nianming","middleName":"","lastName":"Zuo","suffix":""},{"id":541980191,"identity":"848ae41d-89e5-4587-9f5b-d7ba43c42b51","order_by":6,"name":"Zhigang Cai","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuklEQVRIie3QsQrCMBCA4SuFdAl2jSj2FU58AR8lItjFgmOm0iKkS59B38L5Qta8gYu+gd066uognptDvvl+jjuAKPpHWdu4B8BCZB0xE+mdVwCriQyamajdliTA5qTWyCvyZo80tXVpFWgYzYWxhALSIfjKzlpK+nD9nmDSIylDlZ2TThPLSVKJJLEuhdLITITQJE2q+YkKKXkV/NK+nuxYt+Tn+3EYbF0UXeduo2Ekb+jH+SiKouiTJ03JPhtGHAOIAAAAAElFTkSuQmCC","orcid":"","institution":"Peking University School and Hospital of Stomatology","correspondingAuthor":true,"prefix":"","firstName":"Zhigang","middleName":"","lastName":"Cai","suffix":""}],"badges":[],"createdAt":"2025-10-20 18:23:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7908162/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7908162/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95655721,"identity":"8d797938-1612-46f7-82ca-8b097be2946c","added_by":"auto","created_at":"2025-11-11 16:16:46","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":83199,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/720f5ff62f763b8ea7d4679e.docx"},{"id":95566650,"identity":"949df6ce-65f7-492a-851b-65e82190e328","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18412,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/4491e2969d04059d4012dc37.docx"},{"id":95566653,"identity":"869b1190-7942-48a9-9408-82363e7cf3f3","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"tiff","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":78692,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/f31726b8c0cec6f6b0e616de.tiff"},{"id":95656018,"identity":"c5df0279-66b4-48bc-8b67-c087118280d1","added_by":"auto","created_at":"2025-11-11 16:17:33","extension":"tiff","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1072002,"visible":true,"origin":"","legend":"","description":"","filename":"Figure10.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/f54a2cb47df78aecf4bd07f8.tiff"},{"id":95566665,"identity":"3eef8a8c-f341-469b-8efd-0baae8898fb4","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"tiff","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3371208,"visible":true,"origin":"","legend":"","description":"","filename":"Figure11.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/f65ad9340ffd12dd1d7ede71.tiff"},{"id":95655780,"identity":"f313bd1d-5a8d-403e-b6e4-3240d5b408d7","added_by":"auto","created_at":"2025-11-11 16:16:54","extension":"tiff","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1126576,"visible":true,"origin":"","legend":"","description":"","filename":"Figure12.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/14e8f5c9215f08d194e8f6c6.tiff"},{"id":95654882,"identity":"3ca36697-38d6-4d81-89a7-7e75fbe31161","added_by":"auto","created_at":"2025-11-11 16:13:37","extension":"tiff","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":912814,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/899352fe9287574ecad9570f.tiff"},{"id":95656007,"identity":"2b992d82-f9d5-4a00-979e-df9e32402ab0","added_by":"auto","created_at":"2025-11-11 16:17:32","extension":"tiff","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":761014,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/e74ff7095fd552aebadb6d56.tiff"},{"id":95655154,"identity":"9e346f5a-1068-4c2e-bec8-ac3ec708a432","added_by":"auto","created_at":"2025-11-11 16:14:24","extension":"tiff","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1137312,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/7549c570fcbea9ddeca95647.tiff"},{"id":95655329,"identity":"fddf92ff-851a-4b78-8bac-4449b94686af","added_by":"auto","created_at":"2025-11-11 16:15:27","extension":"tiff","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4772728,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/3917a33a8e6e661872d70001.tiff"},{"id":95656045,"identity":"a335fd99-643c-4bf6-bd85-44a25a14feb7","added_by":"auto","created_at":"2025-11-11 16:17:40","extension":"tiff","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2915848,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/cc19872c59b41d36922b467e.tiff"},{"id":95655573,"identity":"fdaefe1a-9b89-46c6-9f8f-d6151c204a1d","added_by":"auto","created_at":"2025-11-11 16:16:30","extension":"tiff","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":824312,"visible":true,"origin":"","legend":"","description":"","filename":"Figure7.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/3ed968320d832118a3f2064b.tiff"},{"id":95654800,"identity":"960d348b-6201-4d8a-8cd3-1fd19382a879","added_by":"auto","created_at":"2025-11-11 16:13:07","extension":"tiff","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2485294,"visible":true,"origin":"","legend":"","description":"","filename":"Figure8.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/798c16c34b0e762e55fd9c2e.tiff"},{"id":95566675,"identity":"946b6af1-65c7-4bbc-ab63-1c8e02a8bbc9","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"tiff","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074198,"visible":true,"origin":"","legend":"","description":"","filename":"Figure9.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/218a81033f41a71318c769a4.tiff"},{"id":95566672,"identity":"df635c2b-8fbc-4ab4-92db-5f9f07e696ae","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"json","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9091,"visible":true,"origin":"","legend":"","description":"","filename":"f869775fc9684cfcb444e9de6d26a5f5.json","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/6d7122bc04775efd449fab8a.json"},{"id":95566686,"identity":"69a09ddd-ce2f-4a2d-bdd7-15691b1feea1","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"docx","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11247382,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/4598f993c5395e3e54d00b81.docx"},{"id":95566674,"identity":"7c218b41-3644-4421-8f86-46f25d4180a2","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":147181,"visible":true,"origin":"","legend":"","description":"","filename":"f869775fc9684cfcb444e9de6d26a5f51enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/01d89d3ade15b979f9b0a24c.xml"},{"id":95566677,"identity":"625d5f3b-4b9d-404f-bdd7-b180a5820742","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"tiff","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":78692,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/fc2bf86f11ae702fccf6097b.tiff"},{"id":95654760,"identity":"ca406b07-8d99-4fba-b2b1-7ae78cf8ede5","added_by":"auto","created_at":"2025-11-11 16:12:57","extension":"tiff","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1072002,"visible":true,"origin":"","legend":"","description":"","filename":"Figure10.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/1b9f729dbf439ac1939409dd.tiff"},{"id":95566684,"identity":"62bb08b0-820f-4d49-9ee1-9ed02ae91a34","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"tiff","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3371208,"visible":true,"origin":"","legend":"","description":"","filename":"Figure11.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/76dead548c0163c51a0042a4.tiff"},{"id":95655619,"identity":"104de9dd-98f6-46fc-8d78-9963adb65bde","added_by":"auto","created_at":"2025-11-11 16:16:34","extension":"tiff","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1126576,"visible":true,"origin":"","legend":"","description":"","filename":"Figure12.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/af4f757b381dc6f7419b0064.tiff"},{"id":95655217,"identity":"e9ca78a4-ef04-484d-9568-3b8bcd10704c","added_by":"auto","created_at":"2025-11-11 16:14:43","extension":"tiff","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":912814,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/9cfe1463eb9cd71e330e2e49.tiff"},{"id":95654859,"identity":"af1a3bfd-cb62-4ec7-9cd0-6d21d0c9c29c","added_by":"auto","created_at":"2025-11-11 16:13:23","extension":"tiff","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":761014,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/6561f80e2d12d7e05b89928c.tiff"},{"id":95654625,"identity":"006235f2-28a5-40e2-9548-fe10b51efba8","added_by":"auto","created_at":"2025-11-11 16:12:37","extension":"tiff","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1137312,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/2046ab38ddf51f266b0e1903.tiff"},{"id":95655005,"identity":"d5db123d-0e2e-4ce7-8740-7a5e70ff3875","added_by":"auto","created_at":"2025-11-11 16:14:02","extension":"tiff","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4772728,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/910bb55c7a126cbc8834f8aa.tiff"},{"id":95655055,"identity":"f7de09e6-be36-4df2-a431-01443215e8bf","added_by":"auto","created_at":"2025-11-11 16:14:10","extension":"tiff","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2915848,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/cf13b717a2026d7c381b3dd6.tiff"},{"id":95566681,"identity":"8ac6fee9-abea-469b-991f-003d7c91768a","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"tiff","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":824312,"visible":true,"origin":"","legend":"","description":"","filename":"Figure7.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/cb0762f9fc3e7e6529d4c2a9.tiff"},{"id":95566691,"identity":"a0c8458f-dc6e-487f-8919-428ebaf739f6","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"tiff","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2485294,"visible":true,"origin":"","legend":"","description":"","filename":"Figure8.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/2fea121a5b047beb68656ca1.tiff"},{"id":95566682,"identity":"d4ca4e07-d66e-4fca-a6e9-05ee959b0307","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"tiff","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074198,"visible":true,"origin":"","legend":"","description":"","filename":"Figure9.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/a2f73e2d531e94cefca7d6ac.tiff"},{"id":95566687,"identity":"577b5258-78f2-4e4c-9ac1-4a7bb07992db","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"xml","order_by":41,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":143929,"visible":true,"origin":"","legend":"","description":"","filename":"f869775fc9684cfcb444e9de6d26a5f51structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/88e9aebfb3dd856fb59b2594.xml"},{"id":95654974,"identity":"91a45614-2ca6-4aab-a294-ef53705bcb4c","added_by":"auto","created_at":"2025-11-11 16:13:55","extension":"html","order_by":42,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":157752,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/f7d7b249e68005a83b6c5a6a.html"},{"id":95566657,"identity":"0edfd514-d9f7-4fc6-a99e-88ce97313a55","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":16036,"visible":true,"origin":"","legend":"\u003cp\u003eTwo experimental designs in the study. Throughout the experiment, participants performed KMI tasks after ME and VMI tasks after AO. The facial MI tasks included eyebrow raising, eye closing, lip puckering, and grinning. The paradigm for the following experiment was determined after comparing two paradigms of equal total duration. In all trials, participants self-recorded the start and end times of each task by pressing the corresponding buttons.\u003c/p\u003e","description":"","filename":"OnlineFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/f73f2c43cb58c00ef8936390.png"},{"id":95566655,"identity":"ac10d042-c215-4456-a76d-6ba9f432367e","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":145847,"visible":true,"origin":"","legend":"\u003cp\u003eOverall structure of the deep learning model. Following signal preprocessing and event segmentation, the EEG data were fed into the TCN for feature extraction. Then a spatial attention module dynamically computed and reallocated the classification weights of EEG channels, guiding the model focus on brain regions most relevant to facial MIs. The outputs were then passed into the MI classifier for classification. Finally, the raw classification results were normalized by the softmax activation function. For cross-subject classification, we additionally incorporated a domain-adversarial training layer to enable better generalization across individuals. The complete model was trained using a combined classification and domain adversarial loss with early stopping based on validation accuracy.\u003c/p\u003e","description":"","filename":"OnlineFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/5ef6e933ecf50834798e47f0.png"},{"id":95655744,"identity":"e38ee4fd-e2b0-4dc2-9406-963b0b7b6106","added_by":"auto","created_at":"2025-11-11 16:16:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":129976,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of alpha ERD/ERS amplitudes in different brain regions between the two experimental designs. Stronger and more consistent ERD/ERS were observed in the block design than in the event-related design(*: \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05).\u003c/p\u003e","description":"","filename":"OnlineFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/d8d83b01a9ace439a868c039.png"},{"id":95655727,"identity":"a9dbb21b-dbc9-4072-9cc2-dc995e5d1b7f","added_by":"auto","created_at":"2025-11-11 16:16:48","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":168250,"visible":true,"origin":"","legend":"\u003cp\u003eThe overall confusion matrix for event-related, block, and reduced-data block designs. The model trained on data from the block design did not show a significantly change even after the amount of training data was reduced (ER: eyebrow raising, EC: eye closing, LP: lip puckering G: grinning).\u003c/p\u003e","description":"","filename":"OnlineFigure4.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/e089c9382264058a8fd6ec8d.png"},{"id":95655486,"identity":"03c0c655-e988-413f-9cc8-f40842be9880","added_by":"auto","created_at":"2025-11-11 16:16:19","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":602348,"visible":true,"origin":"","legend":"\u003cp\u003eTopographical maps of ERD/ERS in \u003cstrong\u003ea\u003c/strong\u003e, \u003cstrong\u003ec\u003c/strong\u003e alpha and \u003cstrong\u003eb\u003c/strong\u003e, \u003cstrong\u003ed\u003c/strong\u003ebeta bands in \u003cstrong\u003ea\u003c/strong\u003e, \u003cstrong\u003eb\u003c/strong\u003e KMI and \u003cstrong\u003ec\u003c/strong\u003e, \u003cstrong\u003ed \u003c/strong\u003eVMI modalities. ERD features appeared bilaterally and shown similarities across the four facial MIs. In the alpha band, KMI induced obvious ERD in the left prefrontal, right central, and right frontal-temporal regions, while VMI shown weaker, more localized ERD in the same areas. In the beta band, both modalities exhibited lower-amplitude ERD mainly in the left prefrontal and right frontal-temporal regions.No significant ERS was observed.\u003c/p\u003e","description":"","filename":"OnlineFigure5.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/a22afb3fa1b266d95a96cace.png"},{"id":95655485,"identity":"2d08d303-5186-4c92-8194-0e86f060b6a1","added_by":"auto","created_at":"2025-11-11 16:16:19","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":936692,"visible":true,"origin":"","legend":"\u003cp\u003eSpectrograms from partial EEG channels in the KMI and VMI modalities. ERD was predominantly observed in the low frequency range of 0–15 Hz, initiating 0.5–1 seconds after the onset of facial MIs and persisting for about 3 seconds, whereas ERS primarily found in the 20–35 Hz range, commencing simultaneously with ERD but exhibiting a shorter duration. White dotted lines indicate the time–frequency regions averaged for the statistical analysis.\u003c/p\u003e","description":"","filename":"OnlineFigure6.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/ca84145fe656691863fe535f.png"},{"id":95655456,"identity":"c4c2a4a2-7363-4a09-8283-25db7dc693d2","added_by":"auto","created_at":"2025-11-11 16:16:13","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":151929,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of ERD and ERS amplitudes across different facial MIs in\u003cstrong\u003e \u003c/strong\u003eKMI and VMI modalities. Except for the differences in ERD amplitude at channel F3 between eyebrow raising and eye closing, and in ERS amplitude at channel Fp2, no other statistically significant differences were observed across the remaining conditions. (*: \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05).\u003c/p\u003e","description":"","filename":"OnlineFigure7.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/afcbd81ba64a4eb5db293fdc.png"},{"id":95566670,"identity":"f0a31fdd-2c83-43f5-b087-19ac122cffaa","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":255597,"visible":true,"origin":"","legend":"\u003cp\u003eClassification results of \u003cstrong\u003ea\u003c/strong\u003e,\u003cstrong\u003e b\u003c/strong\u003e KMI and \u003cstrong\u003ec\u003c/strong\u003e,\u003cstrong\u003e d\u003c/strong\u003e VMI in \u003cstrong\u003ea\u003c/strong\u003e,\u003cstrong\u003e c\u003c/strong\u003e within-subject and \u003cstrong\u003eb\u003c/strong\u003e,\u003cstrong\u003ed\u003c/strong\u003e cross-subject scenarios for healthy participants. The model achieved the highest within-subject classification accuracy for KMI, while other scenarios shown a significant decline of approximately 20–40%. For facial MI types, eyebrow raising and grinning were recognized more accurately than eye closing and lip puckering, except in the VMI cross-subject classification, where the accuracy was consistently low (ER: eyebrow raising, EC: eye closing, LP: lip puckering G: grinning).\u003c/p\u003e","description":"","filename":"OnlineFigure8.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/aa3e0c1871a87c3aaa497d9f.png"},{"id":95655895,"identity":"0e6d802b-0594-4c9e-abfc-1eea979c8e5c","added_by":"auto","created_at":"2025-11-11 16:17:11","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":157447,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of EEG channels’ classification weights under KMI and VMI conditions. EEG features from the left frontal and parietal regions show notable contributions for both MI modalities decoding.\u003c/p\u003e","description":"","filename":"OnlineFigure9.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/26fb7ac65613592329f88dfe.png"},{"id":95566668,"identity":"cfd5f07b-eb24-4a58-b2c5-674ed7c216aa","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":138592,"visible":true,"origin":"","legend":"\u003cp\u003eWithin-subject classification results of patients. \u003cstrong\u003ea\u003c/strong\u003eThe overall confusion matrix and \u003cstrong\u003eb\u003c/strong\u003e visualization of EEG channel classification weights. Compared to healthy participants, there was no significant decline in classification accuracy for patients, and channels from the frontal and parietal regions maintained similar classification weights (ER: eyebrow raising, EC: eye closing, LP: lip puckering G: grinning).\u003c/p\u003e","description":"","filename":"OnlineFigure10.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/7bf0c93720d5de9cf3a419b2.png"},{"id":95566659,"identity":"ed7a6744-4cf6-41bb-9724-5e1e7ace7af2","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":393727,"visible":true,"origin":"","legend":"\u003cp\u003ePartial patients’ ERD/ERS topographical maps during MIs of their impaired facial movements. Compared with healthy participants who showed consistent ERD patterns, patients exhibited notable variability in their EEG features. In cases of severely impaired facial movements, reduced activity was observed in the left prefrontal region, while some patients showed stronger ERS.\u003c/p\u003e","description":"","filename":"OnlineFigure11.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/bc169a4f7a77ab09c61ab6ab.png"},{"id":95566662,"identity":"27eb3f66-2da5-4205-ad33-bfe9d6dcf040","added_by":"auto","created_at":"2025-11-10 16:19:50","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":145894,"visible":true,"origin":"","legend":"\u003cp\u003eWithin-subject classification of KMI after removing low-weight EEG channels. Model retraining with 25% fewer channels resulted in a modest 5–10% decrease in average accuracy (ER: eyebrow raising, EC: eye closing, LP: lip puckering G: grinning).\u003c/p\u003e","description":"","filename":"OnlineFigure12.png","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/fef29a00a34bb1c28efa86d7.png"},{"id":95660176,"identity":"981c6a71-2f68-4e7d-bd90-0bcb96edc427","added_by":"auto","created_at":"2025-11-11 16:30:58","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5530169,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/c4452b58-f37f-466a-804f-19c3c23501b8.pdf"},{"id":95566673,"identity":"0e26b9ca-ee7f-43c7-aeed-dd7f745f66fc","added_by":"auto","created_at":"2025-11-10 16:19:51","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":11247382,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7908162/v1/2e03beb8ae6b20a11d027e9c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Decoding motor imagery related to major mimetic muscles from electroencephalography","fulltext":[{"header":"Background","content":"\u003cp\u003eFacial nerve paralysis (FNP), with an annual incidence of about 20\u0026ndash;40 per 100,000 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], leads to motor disorders of facial muscles, resulting in a range of physiological and psychological sequelae in patients [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, traditional treatment strategies, including pharmacotherapy, physical therapy, and surgery, focus mainly on the recovery of peripheral nerves and muscles [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Their effectiveness is limited by the slow and incomplete process of axonal regeneration [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], and they rarely address the maladaptive motor cortex reorganization that further hampers functional recovery [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. One promising strategy to overcome these limitations is the application of motor imagery (MI)-based brain-computer interfaces (BCIs). As a communication system capable of decoding spontaneous cerebral activity to control external devices and provide feedback without involving peripheral nerves, BCIs have been proven to improve the prognosis of individuals with paralysis [\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. When combined with functional electrical stimulation (FES) devices, BCIs can establish a closed-loop system that helps improve motor control or restore motor function by inducing impaired facial movements [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, the realization of such a BCI-FES system critically depends on the accurate decoding of facial MIs.\u003c/p\u003e\u003cp\u003eAmong available neural recording techniques, electroencephalography (EEG) remains the preferred choice for the establishment of BCIs due to its portability, affordability, and noninvasive nature, despite its limited spatial resolution and signal-to-noise ratio. Previous functional magnetic resonance imaging (fMRI) studies have shown that the somatotopy of different facial MIs overlaps within the small facial area of the primary motor cortex, suggesting inherent challenges for decoding [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Nevertheless, recent advances in machine learning have made it feasible to decode EEG signals associated with fine motor tasks, such as wrist rotation, fist clenching, and single-finger flexion/extension [\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Therefore, it is imperative to investigate the potential of EEG in facial MI decoding to bridge the current gap in the development of BCIs for FNP rehabilitation.\u003c/p\u003e\u003cp\u003eAt the beginning of MI, motor execution (ME), or action observation (AO), a decrease predominantly in alpha (8\u0026ndash;13 Hz) and beta (13\u0026ndash;30 Hz) band power over the motor-related cortical areas can be observed, which is known as event-related desynchronization (ERD) and reflects cortical activation. Conversely, event-related synchronization (ERS) describes a rebound in beta band power following the termination of these activities [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Although no studies have yet reported features of facial MI-related ERD/ERS, several investigations have observed low-frequency ERD in the primary visual cortex and temporoparietal junction, often with a certain degree of right-sided dominance, during the observation of nonemotional facial expressions [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Regarding facial ME, one high-density EEG study reported that alpha and beta ERD associated with lip movement over the primary sensorimotor cortex and thalamus showed lower intensity but greater bilateral coordination compared with limb movements [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAnother factor influencing decoding performance is the modality of MI. Kinesthetic motor imagery (KMI), the first-person mental rehearsal of motor sensations, and visual motor imagery (VMI), the visualization of movement from a first- or third-person perspective, activate overlapping but distinct neural networks that resemble ME and AO, respectively [\u003cspan additionalcitationids=\"CR22\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. When asked to perform MI for different movements, individuals tend to use KMI and exhibit greater cerebral activity, especially for movements in which they have physical experience [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Therefore, most decoding algorithms are developed based on KMI datasets, making KMI-based BCIs generally exhibit better performance than those based on VMI. However, some users, especially those with motor disabilities, are unable to generate vivid MI to control BCIs even after training [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. This phenomenon is also called \"BCI illiteracy\" and has a higher incidence in KMI than in VMI [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTo the best of our knowledge, no study has systematically reported EEG features of different facial MIs or evaluated their decodability, leaving this area largely unexplored. The present study aims to address this gap by analyzing spatiotemporal EEG features of kinesthetic and visual facial MIs across major facial movements, and by assessing their decoding feasibility, thereby providing a foundation for facial MI-based BCIs in facial rehabilitation.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eParticipants\u003c/p\u003e\n\u003cp\u003e20 healthy participants (14 males and 6 females; aged 23.5 ± 1.9 years) and six individuals with FNP (5 males and 1 female; aged 44.5 ± 14.4 years) were recruited for this study. All participants had no history of neurological or psychiatric disorders and were novices in BCIs. The ability of motor imagery in all participants was evaluated using the Kinesthetic and Visual Imagery Questionnaire (KVIQ-20), with scores of 40.79 ± 4.33 for KMI and 41.42 ± 4.08 for VMI [27]. For the patients, we additionally collected their clinical characteristics and assessed their facial nerve function using the Facial Nerve Grading System 2.0 (FNGS 2.0) [28].\u003c/p\u003e\n\u003cp\u003eEEG recordings\u003c/p\u003e\n\u003cp\u003eDuring the experiment, participants sat comfortably in a fixed chair without armrests and were asked to focus on the instructions shown on the screen (Apple Inc., CA, USA) approximately 80 cm in front of them. EEG signals were recorded using an amplifier (XLTEK EMU40EX, Natus, WI, USA) and an EEG cap (Tenocom, Shandong, China) with 19 recording electrodes positioned according to the international 10–20 system [29], a reference electrode placed in the parietal region, and a grounding electrode located in the frontal region. All electrode impedances were kept below 5 kΩ during recording and the sampling rate was set at 256 Hz.\u003c/p\u003e\n\u003cp\u003eExperimental paradigms\u003c/p\u003e\n\u003cp\u003eTo determine the optimal decoding strategy for facial MIs, we divided the experiment into three steps. To enhance the performance of facial MIs in participants, we designed KMI tasks after ME and VMI tasks after AO. In all tasks, participants were specifically instructed to avoid emotional involvement as much as possible.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(Fig. 1)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirst, to identify the more suitable paradigm for\u0026nbsp;facial MIs, we compared block and event-related designs (illustrated in Fig. 1)\u0026nbsp;using EEG data from seven of the 20 healthy participants.\u0026nbsp;The total duration of both designs was controlled at approximately 90 minutes (including rest time) and the data were collected respectively with an interval of seven days. The\u0026nbsp;event-related design consisted of 60 blocks, with rest intervals of five minutes between every 20 blocks, and each block comprised a single round of randomized and discrete four-type facial KMI trials: eyebrow raising, eye closing, lip puckering and grinning.\u0026nbsp;At the beginning of each block, participants were asked to execute one of the facial movements in about three seconds according to the auditory prompts. To obtain the baseline for reference, participants were instructed to get relaxed as much as possible and focus on the white fixation cross presented on the screen without any movement for four seconds before and after the imagining phases. When performing the KMI, participants were asked to recall the sensation felt during the ME phase and to imagine performing the same facial movements without any muscle contraction. The start and end times of each KMI were annotated by participants by pressing the corresponding buttons. Before the next trial began, participants were provided a 10-second rest. The block design consisted of four blocks, with five-minute rest intervals in between, and each block comprised 24 trials, all of the same facial KMI type. Participants followed a similar sequence of \"ME–relax–KMI–relax–rest\" in each trial, while they were instructed to imagine facial movements continuously five times in a single KMI trial, namely, 120 times of KMI for each type of facial movement. The rest time between trials was 10 seconds.\u003c/p\u003e\n\u003cp\u003eIn the following experiment, data of KMI and VMI were collected following the better design, from a total of 20 healthy participants. In VMI tasks, the only difference was that participants were asked to perform VMI (imagining seeing facial movements without kinesthetic sensation) after observing a video of a facial movement displayed on the screen. Finally, we collected single-modality facial MI EEG data from patients and halved the number of tasks after determining the appropriate decoding strategy.\u003c/p\u003e\n\u003cp\u003eEEG data preprocessing\u003c/p\u003e\n\u003cp\u003eThe entire process of EEG decoding was conducted with MATLAB (MathWorks, Natick, MA, USA) and the EEGLAB toolbox. Raw EEG data were first band-pass filtered between 0.5 and 50 Hz, with a 50 Hz notch filter applied to suppress power-line interference. Next, an infinite reference using the reference electrode standardization technique was employed to mitigate the impact of electrode positional deviations. EEG channels with variance or root mean square exceeding three standard deviations from the mean across all electrodes were automatically excluded. To minimize interference from adjacent physiological activities, we visually inspected the entire EEG data and used artifact subspace reconstruction to identify and remove components related to eye blink, muscle contractions, and other abnormal artifacts. Finally, the cleaned data were segmented into epochs according to trial markers. To account for slight variations in data length between events, all EEG segments were resampled to three seconds to ensure consistency across trials and prevent the model from capturing length-related artifacts.\u003c/p\u003e\n\u003cp\u003eFeature extraction of ERD/ERS\u003c/p\u003e\n\u003cp\u003eTo systematically investigate brain activity during facial MIs, we extracted both temporal and spatial features of ERD/ERS and plotted the corresponding maps.\u003c/p\u003e\n\u003cp\u003eAccording to the neurophysiological ERD/ERS characteristics described in relevant studies, the alpha and beta bands were selected as the bands of interest. For each epoch, a time-frequency analysis was performed on all EEG channels by means of short-time Fourier transform (STFT), with a 1-s Hanning window, and a frequency resolution of 1 Hz. For the channels contributing the most to MI classification, we additionally generated spectrograms for the frequency range 0.5–50 Hz with the same resolution. The time resolution was set to 20 ms. Finally, the ERD/ERS for each MI trial was calculated by:\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"https://myfiles.space/user_files/127393_c7e80a1c9bb65875/127393_custom_files/img1762779690.png\" style=\"width: 648px;\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere \u0026nbsp; represents the power spectral density (PSD) of frequency \u003cem\u003ef\u0026nbsp;\u003c/em\u003eand EEG channel \u003cem\u003ec\u0026nbsp;\u003c/em\u003eat time \u003cem\u003et\u003c/em\u003e, and \u0026nbsp; is the average PSD during the baseline period (-2 to -1 s prior to cue onset). The resulting ERD/ERS of all trials from the same MI task were averaged per healthy participant. Negative values represent ERD, while positive values indicate ERS.\u003c/p\u003e\n\u003cp\u003eThe development of the deep learning model\u003c/p\u003e\n\u003cp\u003eWe designed a deep neural network consisting of three primary components, with the overall structure illustrated in\u0026nbsp;Fig. 2:\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003eTemporal Convolutional Network (TCN):\u003cbr\u003e\u0026nbsp;This module extracts time-domain features from EEG signals using parallel convolutional branches with kernels of different scales. One branch uses a large-scale convolution kernel (1×100, stride = 20) to capture long-term temporal dependencies, while the other uses a small-scale kernel (1×25, stride = 5) to extract fine-grained temporal patterns. The outputs from both branches are concatenated to form a rich multi-scale temporal representation.\u003c/li\u003e\n \u003cli\u003eSpatial Attention Mechanism:\u003cbr\u003e\u0026nbsp;To enhance the model’s sensitivity to task-relevant brain regions, we introduced a channel-wise attention mechanism. This module computes attention weights across EEG channels using global average pooling followed by two fully connected layers and a softmax activation. These weights are then used to reweight the channel features extracted by the convolutional layers, enabling the model to focus on the most discriminative features dynamically. After model training, we extracted the weight values of each channel from the attention layer parameters and visualized them on a brain topographic map.\u003c/li\u003e\n \u003cli\u003eDomain-Adversarial Training Layer:\u003cbr\u003e\u0026nbsp;Inter-subject variability poses a major challenge in EEG decoding. To improve cross-subject generalizability, we incorporated a domain-adversarial learning strategy using a gradient reversal layer (GRL). This component introduces an auxiliary domain classifier connected to the shared feature space. During training, the GRL encourages the shared features to be indistinguishable across source and target subjects by reversing gradients from the domain classifier. This promotes the learning of domain-invariant features, enabling better generalization across individuals.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThe complete model was trained using a combined loss function consisting of a classification loss (cross-entropy) and a domain adversarial loss, with a trade-off coefficient λ set to 0.1. Training was conducted using the Adam optimizer (learning rate = 0.001, batch size = 32, epochs = 100), with early stopping based on validation accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(Fig. 2)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eClassification tasks\u003c/p\u003e\n\u003cp\u003eThe model’s decoding ability was tested in both within-subject and cross-subject scenarios, with classification accuracy as the primary evaluation metric. To ensure robustness, we performed eight-fold cross-validation for within-subject classification tasks and leave-one-participant-out cross-validation for cross-subject tasks. The data were randomly partitioned at the trial level, with careful attention paid to avoid data leakage between the training and testing sets. The results were visualized using confusion matrices, with row-wise normalization applied.\u003c/p\u003e\n\u003cp\u003eStatistical analysis\u003c/p\u003e\n\u003cp\u003eThe differences in EEG amplitudes and model performance across conditions were analyzed using paired t-tests or Wilcoxon signed-rank tests, depending on the normality of the data distribution. All statistical analyses were conducted with SPSS 27.0 (IBM, Armonk, NY, USA), with a significance level set at 0.05.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eComparison between block design and event-related design\u003c/p\u003e\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents comparisons of alpha ERD/ERS amplitudes between the event-related design and the block design in different brain regions. Significant differences were observed for eye closing in the occipital region (-1.46\u0026thinsp;\u0026plusmn;\u0026thinsp;2.88% vs. 10.35\u0026thinsp;\u0026plusmn;\u0026thinsp;14.09%, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.047) and lip puckering in the temporal region (\u0026minus;\u0026thinsp;10.60\u0026thinsp;\u0026plusmn;\u0026thinsp;15.37% vs. 4.05\u0026thinsp;\u0026plusmn;\u0026thinsp;10.05%, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.047). Although other comparisons did not reach statistical significance, participants consistently exhibited a stronger ERD/ERS with less inter-subject variability in the block design. As shown by the confusion matrices in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the model, when trained on data collected from the block design, achieved approximately 40% higher accuracy than when trained on the event-related design (84.12\u0026thinsp;\u0026plusmn;\u0026thinsp;11.19% vs. 45.11\u0026thinsp;\u0026plusmn;\u0026thinsp;8.38%, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.016), and its accuracy did not significantly change even when the amount of training data was reduced (84.12\u0026thinsp;\u0026plusmn;\u0026thinsp;11.19% vs. 85.29\u0026thinsp;\u0026plusmn;\u0026thinsp;10.07%, p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Because performing facial MI tasks continuously improved both participant performance and model accuracy, the block design was selected for the following experiment. The number of each task was kept at 120 to enhance the representativeness of ERD/ERS features, even though halving the training set size did not substantially affect model performance.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e)\u003c/h2\u003e\u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003eERD/ERS spatial features during different facial MIs\u003c/p\u003e\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e illustrates the alpha and beta ERD/ERS features during four facial MIs in different modalities on topographical maps. These features were distributed in bilateral brain regions and demonstrated similarities across four facial MIs in both KMI and VMI modalities. In the alpha band, a strong KMI-related ERD was found in the left prefrontal, right central, and right frontal-temporal regions, while VMI exhibited a weaker and more localized ERD in the same regions. Similarly, the ERD, with lower amplitude, was mainly observed in the left prefrontal and right frontal-temporal regions for KMI and VMI in the beta band. In contrast, no significant ERS was observed across all conditions.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\n\u003ch3\u003e(Fig. 5)\u003c/h3\u003e\n\u003cp\u003eDespite these overarching similarities, the amplitudes and extent of ERD vary slightly across facial MIs. In the KMI condition, lip puckering showed relatively broader distributions extending toward the right temporal region in the alpha band. Additionally, lip puckering and grinning exhibited increased beta ERD in the middle frontal and right temporal-parietal regions compared to eyebrow raising and eye closing. For the VMI, except for eyebrow raising, other facial MIs tended to exhibit increased brain activity in the middle frontal region. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the comparison of alpha and beta ERD amplitudes in their associated brain regions between the two MI modalities. In the right frontal-temporal region, the KMI of eye closing and lip puckering exhibited a stronger beta ERD compared to VMI, while no statistically significant differences were observed in other conditions.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eComparison of alpha and beta ERD/ERS amplitudes during different facial MIs in their associated brain regions between two MI modalities.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eFrequency band\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eBrain region\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eMI modality\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c7\" namest=\"c4\"\u003e\u003cp\u003eERD/ERS amplitudes (%) during different facial MIs\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEyebrow raising\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eEye closing\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLip puckering\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003eGrinning\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"8\" rowspan=\"9\"\u003e\u003cp\u003eAlpha\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eLeft prefrontal region\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-11.89\u0026thinsp;\u0026plusmn;\u0026thinsp;10.34\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-9.95\u0026thinsp;\u0026plusmn;\u0026thinsp;8.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-10.85\u0026thinsp;\u0026plusmn;\u0026thinsp;9.77\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-12.33\u0026thinsp;\u0026plusmn;\u0026thinsp;11.60\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-10.37\u0026thinsp;\u0026plusmn;\u0026thinsp;10.74\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-6.85\u0026thinsp;\u0026plusmn;\u0026thinsp;6.54\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-6.74\u0026thinsp;\u0026plusmn;\u0026thinsp;5.56\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-8.58\u0026thinsp;\u0026plusmn;\u0026thinsp;7.29\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eRight central region\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-11.89\u0026thinsp;\u0026plusmn;\u0026thinsp;11.53\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-10.49\u0026thinsp;\u0026plusmn;\u0026thinsp;10.01\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-11.72\u0026thinsp;\u0026plusmn;\u0026thinsp;10.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-13.16\u0026thinsp;\u0026plusmn;\u0026thinsp;12.11\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-10.27\u0026thinsp;\u0026plusmn;\u0026thinsp;9.81\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-7.45\u0026thinsp;\u0026plusmn;\u0026thinsp;6.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-8.34\u0026thinsp;\u0026plusmn;\u0026thinsp;7.27\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-9.52\u0026thinsp;\u0026plusmn;\u0026thinsp;7.51\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eRight frontal-temporal region\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-10.82\u0026thinsp;\u0026plusmn;\u0026thinsp;10.85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-9.54\u0026thinsp;\u0026plusmn;\u0026thinsp;8.65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-10.94\u0026thinsp;\u0026plusmn;\u0026thinsp;9.61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-11.47\u0026thinsp;\u0026plusmn;\u0026thinsp;11.49\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-9.80\u0026thinsp;\u0026plusmn;\u0026thinsp;9.49\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-7.10\u0026thinsp;\u0026plusmn;\u0026thinsp;6.26\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-7.39\u0026thinsp;\u0026plusmn;\u0026thinsp;6.13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-8.18\u0026thinsp;\u0026plusmn;\u0026thinsp;6.48\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eBeta\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eLeft prefrontal region\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-5.44\u0026thinsp;\u0026plusmn;\u0026thinsp;3.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-5.87\u0026thinsp;\u0026plusmn;\u0026thinsp;3.83\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-5.27\u0026thinsp;\u0026plusmn;\u0026thinsp;3.49\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-5.50\u0026thinsp;\u0026plusmn;\u0026thinsp;3.15\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-4.81\u0026thinsp;\u0026plusmn;\u0026thinsp;3.91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-3.13\u0026thinsp;\u0026plusmn;\u0026thinsp;4.56\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-4.19\u0026thinsp;\u0026plusmn;\u0026thinsp;3.60\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-3.65\u0026thinsp;\u0026plusmn;\u0026thinsp;3.92\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eRight frontal-temporal region\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-5.65\u0026thinsp;\u0026plusmn;\u0026thinsp;4.48\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-7.27\u0026thinsp;\u0026plusmn;\u0026thinsp;4.47\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-6.96\u0026thinsp;\u0026plusmn;\u0026thinsp;5.05\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-6.88\u0026thinsp;\u0026plusmn;\u0026thinsp;5.15\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-5.37\u0026thinsp;\u0026plusmn;\u0026thinsp;3.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-3.71\u0026thinsp;\u0026plusmn;\u0026thinsp;3.14\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-4.29\u0026thinsp;\u0026plusmn;\u0026thinsp;3.70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-4.16\u0026thinsp;\u0026plusmn;\u0026thinsp;3.35\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.0209\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.0226\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003ens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"7\"\u003eERD/ERS amplitudes from channels within the same brain region were averaged. The left prefrontal region was represented by Fp1, F3, and F7, the right central region by Cz and C4 and the right frontal-temporal region by F8 and T4.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e(Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003eTime-frequency analysis of EEG channels from active regions in topographical maps\u003c/p\u003e\u003cp\u003eAccording to the topographical maps, we selected channels Fp1, Fp2, F3, F7, C4, and T4 for the time-frequency analysis, with the spectrograms from these channels are illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and Supplementary Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003es1\u003c/span\u003e. Across all conditions and channels, we observed an obvious power decrease in the lower frequency range (0\u0026ndash;15 Hz), particularly within the delta, theta, and alpha band, with a lesser extent in the low beta band. This ERD started shortly after the onset of facial MIs (approximately 0.5\u0026ndash;1 seconds), peaked at around 1.5\u0026ndash;2 seconds, and extended to about 3 seconds. In parallel, the ERS was found in the higher beta and low gamma bands (15\u0026ndash;35 Hz), often co-occurring or followed by the ERD in time. Notably, the duration of ERS was generally shorter than that of ERD, typically appearing as a transient increase in power that did not persist for the full duration of the MI period. We also found a slight ERS in the gamma band, specifically between 40 Hz and 50 Hz, which might be associated with residual muscular artifacts in the EEG data. In addition, we averaged the ERD amplitudes within 0\u0026ndash;15 Hz and 0.5\u0026ndash;3 seconds, and the ERS amplitudes within 15\u0026ndash;35 Hz and 0.5\u0026ndash;2 seconds for statistical comparisons shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, which are indicated as white dotted lines on the spectrograms. Except for the differences in ERD amplitude at channel F3 between eyebrow raising and eye closing, and in ERS amplitude at channel Fp2, no other statistically significant differences were observed across the remaining conditions.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e)\u003c/h2\u003e\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003e Within-subject and cross-subject decoding results of healthy participants\u003c/p\u003e\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows the overall confusion matrix and classification accuracy of two classification scenarios for healthy participants, with statistical comparisons of recall and precision between the two MI modalities presented in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003es1\u003c/span\u003e. In both scenarios, our model demonstrated significantly superior recognition performance for KMI compared to VMI across all facial MIs. For the within-subject decoding, we achieved an average accuracy of 85.17\u0026thinsp;\u0026plusmn;\u0026thinsp;8.54% for KMI and 70.80\u0026thinsp;\u0026plusmn;\u0026thinsp;12.99% for VMI. Specifically, the model exhibited the best recognition capability for eyebrow raising, with an average recall of 91.03\u0026thinsp;\u0026plusmn;\u0026thinsp;7.10% for KMI and 80.47\u0026thinsp;\u0026plusmn;\u0026thinsp;10.54% for VMI. Performance on eye closing and grinning was also relatively high for KMI (82.92\u0026thinsp;\u0026plusmn;\u0026thinsp;10.66% and 86.67\u0026thinsp;\u0026plusmn;\u0026thinsp;10.90%, respectively), while lip puckering showed the lowest recall of 80.09\u0026thinsp;\u0026plusmn;\u0026thinsp;11.82%. By comparison, the model's performance for VMI showed a noticeable drop, with recall decreasing by around 10\u0026ndash;20% across facial MIs. The highest decoding capability was still observed for eyebrow raising (80.47\u0026thinsp;\u0026plusmn;\u0026thinsp;10.54%), while grinning yielded the lowest performance (63.34\u0026thinsp;\u0026plusmn;\u0026thinsp;12.58%). In the cross-subject scenario, we obtained an average accuracy of 66.25\u0026thinsp;\u0026plusmn;\u0026thinsp;10.60% for KMI and 48.44\u0026thinsp;\u0026plusmn;\u0026thinsp;8.89% for VMI, both showing a significant decline of approximately 30% compared to the within-subject classification. The model retained the relative tendency among KMI tasks, achieving higher accuracy for eyebrow raising and grinning than others. In contrast, the model showed consistently poor performance in all VMI tasks, with recall for none of the facial MIs exceeding 50%.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003eDue to the limited generalizability of our model, we further analyzed the cross-subject classification results through the corresponding accuracy and loss curves demonstrated in Supplementary Fig. s2. In the KMI condition, the training accuracy steadily increased with low training loss and eventually exceeded 85%, suggesting effective learning within the training data. However, the validation accuracy plateaued around 60%, and the validation loss began to rise after approximately 75 epochs, indicating potential overfitting. For the VMI, the model exhibited slower convergence, with training accuracy peaking at only about 45% and no clear plateau observed. Validation accuracy fluctuated throughout the training process, remaining within the 30\u0026ndash;35% range. Meanwhile, the training loss continued to decrease, while the validation loss began to rise after roughly 120 epochs, further indicating poor generalization in the VMI condition.\u003c/p\u003e\u003cp\u003eContributions of EEG channels to facial MI classification\u003c/p\u003e\u003cp\u003eThe classification weight values of EEG channels produced by the attention layer are presented in Supplementary Table s2 and visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e. From these topographical maps, it is evident that the EEG features from the prefrontal and left frontal regions contributed the most to classification in both conditions, while features from the right prefrontal region exhibited the least contribution. The EEG channels in other regions showed distinct classification weights for KMI and VMI. In KMI condition, channels located in the bilateral temporal and right parietal regions presented second-highest classification weights, whereas medial parietal and left occipital channels exhibited no significant contribution to classification. The top three contributing channels were Fp2, Fp1, and F3. By comparison, in VMI condition, the channels in the right central and right temporal region exhibited contributions comparable to those of the frontal channels, followed by channels in the left parietal region. In contrast, the medial frontal, medial central, left central, and left temporal channels showed low classification weights. The leading contributors were identified as F3, Fp1, and C4.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003eWithin-subject classification results and ERD/ERS topographic maps of patients\u003c/p\u003e\u003cp\u003ePatients\u0026rsquo; clinical characteristics are listed in Supplementary Table s3. All patients presented with unilateral FNP, with etiologies including parotidectomy, trauma, and parotid malignant tumor. The duration ranged from 2 to 80 weeks, with variations in the affected regions and severity of facial nerve injuries. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003ea, the average classification accuracy was 83.81\u0026thinsp;\u0026plusmn;\u0026thinsp;8.97% among these patients, with no significant difference compared to healthy participants. The model maintained the best decoding capability for eyebrow raising, with an average recall of 94.57\u0026thinsp;\u0026plusmn;\u0026thinsp;5.17%, outperforming other facial MIs by 10\u0026ndash;15%. The patients' topographic map of channel classification weights is visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eb. EEG channels in the left frontal and parietal regions maintained higher classification weights, whereas those in the right frontal region continued to exhibit limited contributions to classification. Moreover, we selected representative topographic maps from patients during facial MI of the movements affected by their facial nerve injuries, which are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e, with the remaining maps provided in Supplementary Fig. s3. In contrast to the consistent ERD features observed in healthy participants, the patients exhibited significant variations in their EEG features. For severely affected facial movements, patients tended to show reduced activity in the left prefrontal region. Additionally, certain subjects demonstrated stronger ERS than healthy participants.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e)\u003c/h2\u003e\u003cdiv id=\"Sec16\" class=\"Section3\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003eResults of within-subject classification of KMI after model retraining\u003c/p\u003e\u003cp\u003eBased on the decoding strategy and channel classification weights, we removed five EEG channels (F4, F7, F8, C3 and T3) whose weights were in the lower 50% for both healthy subjects and patients in within-subject classification of KMI, and retrained the model. The overall confusion matrix is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e. The average classification accuracy decreased to 79.16\u0026thinsp;\u0026plusmn;\u0026thinsp;7.06% for healthy subjects and to 74.83\u0026thinsp;\u0026plusmn;\u0026thinsp;8.25% for patients.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e(Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e)\u003c/h2\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study presented the first exploration of the EEG spatiotemporal features during facial MIs related to major mimetic muscles and the corresponding decoding strategies, in both healthy participants and individuals with FNP. Our results demonstrated that the difference across various facial MIs could be effectively discriminated by deep learning, with the peak average classification accuracy reaching 85.17% for healthy participants and 83.81% for patients, achieved through the strategy of within-subject decoding of KMI collected using a block design. The classification weight analysis revealed that EEG features from the left frontal and parietal regions provided the most critical information for facial MI decoding. The model retraining with 25% fewer channels, further combining the analysis, resulted in only about 5\u0026ndash;10% decrease in average accuracy. Based on these findings, we can accurately detect the facial movement intentions of patients and lay the foundation for simplifying EEG acquisition devices to reduce the latency of BCIs.\u003c/p\u003e\u003cp\u003eSimilarity in ERD/ERS spatiotemporal features between KMI and VMI\u003c/p\u003e\u003cp\u003eAs revealed by the topographic maps, facial MIs, similar to other types of MI, recruit the motor cortex for the programming and planning of movements, and inhibit the overt ME through the left prefrontal cortex. There is also a beta ERD observed at channel P4, suggesting that the parietal lobe, known for its role in sensory integration and motor attention processing [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], is involved in facial MIs. In addition, the ERD originating from the right temporal region may involve activity in the superior temporal gyrus and fusiform gyrus, which are considered signature cortical areas for facial MIs, as they are associated with the processing of facial tasks and expressions [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Another notable aspect is the similarity of facial MI-related ERD/ERS in spatial distribution between KMI and VMI. Although the two MI modalities recruit partially distinct neural pathways, several studies have shown that their brain activation patterns are similar, with only a temporal difference [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Moreover, when performing a MI task, the selection of MI modalities and the quality of MI appear to depend on a person\u0026rsquo;s prior motor experience, especially the task-specific experience [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. For an unfamiliar and short-duration MI task\u0026mdash;such as the facial MIs in this study\u0026mdash;participants may struggle to fully follow the researcher's instructions and tend to use KMI rather than VMI. The pronounced activation of the left prefrontal cortex, along with the lack of activity in the occipital regions, supports this interpretation.\u003c/p\u003e\u003cp\u003ePerformance gaps in cross-subject decoding of facial MIs\u003c/p\u003e\u003cp\u003eTo identify the most suitable decoding strategy for facial MIs, we tested the model performance in different conditions. Compared to intra-subject classification, where the model achieved relatively high accuracies for both KMI (85.17%) and VMI (70.80%), performance in the cross-subject scenario declined markedly, with an average drop of approximately 30% for both modalities. The decrease was especially pronounced for VMI, where recall for all tasks fell below 50%. In contrast, KMI retained its relative performance pattern across tasks, particularly for eyebrow raising and grinning. This difference was further supported by the confusion matrices and training curves. In the VMI condition, the model exhibited unstable learning, with low validation accuracy and increasing validation loss over time, indicating poor generalization. For KMI, although the model showed more stable training dynamics, signs of overfitting still appeared at an early stage. This performance gap may result from greater variability in VMI-related EEG features across subjects. Although VMI often seems to involve imagining how an action appears visually from a third-person perspective [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], it can also involve first-person visualization [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. This contrasts with KMI, which consistently employs the first-person perspective [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. One possible improvement to the decoding method is to use EEG features in source space rather than in sensor space. By projecting EEG data onto anatomically aligned cortical regions and further integrating brain connectivity analysis, this method can provide more consistent features for decoding and improve the robustness of cross-subject classification [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Recent studies have shown that deep learning models trained on these features can achieve a significant improvement over classic MI-EEG datasets [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Considering the plug-and-play functionality of cross-subject classification models, it is worth investigating their potential in facial MIs decoding using source-space features.\u003c/p\u003e\u003cp\u003eOptimizing electrode layout for low-latency facial MIs decoding\u003c/p\u003e\u003cp\u003eIn addition to classification accuracy, another crucial factor affecting BCIs performance in practical applications is response time. To promote cortical plasticity and restore movement symmetry, the response time thresholds for BCIs applied to individuals with FNP should be 400 ms and 50 ms, respectively [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Therefore, designing appropriate electrode layouts is essential to minimize the data volume to be processed and shorten the response time. Encouragingly, our findings suggest that an average accuracy of over 75% can be achieved using only 14 out of 19 electrodes of the 10\u0026ndash;20 system, with the number and layout of electrodes further optimizable based on the results of classification weight distribution across EEG channels. The use of EEG features in source space might also help reduce delays. Although additional EEG source reconstruction is required, data with lower variability contributes to reduced complexity in deep learning models, thereby decreasing decoding time [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Due to the poor performance of single-electrode placement in key regions for EEG source decoding, it is necessary to appropriately increase electrode density [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. Still, a careful balance must be struck between computational cost and classification accuracy.\u003c/p\u003e\u003cp\u003eAdditionally, although previous EEG decoding studies have indicated that low-frequency EEG features typically contribute significantly to distinguishing different tasks [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], several channels with high classification weights (e.g., Fp2, P3, and P4) did not exhibit strong ERD in the present study. This suggests that contributions of different channels may vary across frequency bands, with higher-frequency features playing a more important role in certain regions. Overall, a more fine-grained channel\u0026ndash;frequency analysis, together with facial MI-specialized EEG acquisition devices and optimized decoding algorithms, will be essential for developing BCIs that achieve both high accuracy and low latency in patients.\u003c/p\u003e\u003cp\u003eLimitations and future work\u003c/p\u003e\u003cp\u003eThere are some limitations in our study. The first one is the number of patients, especially those with long durations. Because peripheral paralysis affects the motor cortex in a way that correlates with motor disability and follows a progressive course, long-term follow-up is needed for patients with permanent FNP or complications such as facial synkinesis to assess potential impairment of BCI performance. In addition, the diversity of facial nerve injuries and the variability of EEG signals may limit the representativeness of ERD/ERS features in patients. Further investigation through neuroimaging methods with superior spatial resolution, such as fMRI and functional near-infrared spectroscopy, should be conducted to ensure the impact of FNP on motor cortex plasticity. These methods based on blood-oxygen-level-dependent signals can also assess brain activity during ME of facial movements without being affected by electromyographic interference, providing more insights into the impact of FNP on cortex plasticity.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe present study systematically investigated spatiotemporal ERD/ERS features during facial MIs related to major mimetic muscles for the first time. We found that facial MIs can induce marked low-frequency ERD in the left prefrontal and right central-frontal-temporal regions, while the concurrent ERS was predominantly observed in the higher beta and gamma bands. These features demonstrated similarities across four facial MIs in two MI modalities. Furthermore, facial MIs could be accurately decoded from EEG using deep learning, with signals from the left frontal and parietal regions showing notable contributions to decoding for both KMI and VMI. According to the model performance in different conditions, we determined the decoding strategy of within-subject classification of KMI-related EEG collected from the block design. The model finally achieved an average accuracy of 83.81% on the patient dataset, with training data that could be collected within an hour. Overall, our findings demonstrate the feasibility of achieving accurate decoding of facial MIs through their ERD/ERS spatiotemporal features, paving the way for the development of BCIs in individuals with FNP.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eAO Action observation\u003c/p\u003e\u003cp\u003eBCI Brain-computer interface\u003c/p\u003e\u003cp\u003eEEG Electroencephalography\u003c/p\u003e\u003cp\u003eERD Event-related desynchronization\u003c/p\u003e\u003cp\u003eERS Event-related synchronization\u003c/p\u003e\u003cp\u003eFES Functional electrical stimulation\u003c/p\u003e\u003cp\u003efMRI Functional magnetic resonance imaging\u003c/p\u003e\u003cp\u003eFNGS 2.0 Facial Nerve Grading System 2.0\u003c/p\u003e\u003cp\u003eFNP Facial nerve paralysis\u003c/p\u003e\u003cp\u003eGRL Gradient reversal layer\u003c/p\u003e\u003cp\u003eKMI Kinesthetic motor imagery\u003c/p\u003e\u003cp\u003eKVIQ-20 Kinesthetic and Visual Imagery Questionnaire\u003c/p\u003e\u003cp\u003eME Motor execution\u003c/p\u003e\u003cp\u003eMI Motor imagery\u003c/p\u003e\u003cp\u003eTCN Temporal convolutional network\u003c/p\u003e\u003cp\u003eVMI Visual motor imagery\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Ethics Committee of Peking University School and Hospital of Stomatology (reference number PKUSSIRB-202387067) and all participants provided written informed consent prior to their participation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConsent for publication of individual data has been obtained in writing from all participants of this study.\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003eSupplementary Information\u003c/h2\u003e\n\u003cp\u003eSupplementary material 1: Fig. \u003cspan class=\"InternalRef\"\u003es1\u003c/span\u003e. Spectrograms from partial EEG channels in the KMI and VMI modalities. Table \u003cspan class=\"InternalRef\"\u003es1\u003c/span\u003e. Comparison of performance metrics between KMI and VMI for different facial MIs in two classification scenarios. Fig. s2. The accuracy and loss curves for KMI and VMI cross-subject classification. Table s2. EEG channels\u0026rsquo; classification weights under different conditions. Table s3. Characteristics of the patients. Fig. s3. Patients\u0026rsquo; ERD/ERS topographical maps during different facial MIs.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis study was supported by the Capital Health Research and Development of Special (2022\u0026ndash;2-4102) and the Clinical Research Foundation of Peking University School and Hospital of Stomatology (PKUSS-2023CRF102).\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eThe study was was conceptualized and designed by H.S., M.D., N.Z., and Z.C.. Methodology was developed by H.S., M.D. and Z.C.. Data collection was performed by H.S.. The model was developed by D.C.. Data analysis and visualizations were carried out by H.S. and D.C., and N.Z. interpreted the results. The original draft was written by H.S., while M.D., X.S., S.X., D.C., N.Z., and Z.C. reviewed and edited the the manuscript. All authors have read and approved the final version of the manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgement\u003c/h2\u003e\n\u003cp\u003eWe gratefully thank the assistance of Xin Liu and Meiyuan Sun from the Department of Neurology at Beijing Zhongguancun Hospital in the EEG recordings for this research.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets of the present study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003ePeitersen E. Bell's palsy: the spontaneous course of 2,500 peripheral facial nerve palsies of different etiologies. Acta Otolaryngol Suppl. 2002;(549):4\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaadi R, Shokri T, Schaefer E, Hollenbeak C, Lighthall JG. Depression Rates After Facial Paralysis. Ann Plast Surg. 2019;83(2):190\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi:10.1097/SAP.0000000000001908\u003c/span\u003e\u003cspan address=\"https://doi:10.1097/SAP.0000000000001908\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNakano H, Fujiwara T, Tsujimoto Y, Morishima N, Kasahara T, Ameya M, et al. Physical therapy for peripheral facial palsy: A systematic review and meta-analysis. Auris Nasus Larynx. 2024;51(1):154\u0026ndash;60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.anl.2023.04.007\u003c/span\u003e\u003cspan address=\"10.1016/j.anl.2023.04.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSunderland S. A classification of peripheral nerve injuries producing loss of function. Brain. 1951;74(4):491\u0026ndash;516.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi C, Liu SY, Pi W, Zhang PX. Cortical plasticity and nerve regeneration after peripheral nerve injury. Neural Regen Res. 2021;16(8):1518\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/1673-5374.303008\u003c/span\u003e\u003cspan address=\"10.4103/1673-5374.303008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBaniqued PDE, Stanyer EC, Awais M, Alazmani A, Jackson AE, Mon-Williams MA, et al. Brain\u0026ndash;computer interface robotics for hand rehabilitation after stroke: a systematic review. J Neuroeng Rehabil. 2021;18(1):15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12984-021-00820-8\u003c/span\u003e\u003cspan address=\"10.1186/s12984-021-00820-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLorach H, Galvez A, Spagnolo V, Martel F, Karakas S, Intering N, et al. Walking naturally after spinal cord injury using a brain\u0026ndash;spine interface. Nature. 2023;618(7963):126\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-023-06094-5\u003c/span\u003e\u003cspan address=\"10.1038/s41586-023-06094-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang M, Li C, Liu S-Y, Zhang F-S, Zhang P-X. An electroencephalography-based human-machine interface combined with contralateral C7 transfer in the treatment of brachial plexus injury. Neural Regeneration Res. 2022;17(12):2600\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/1673-5374.335838\u003c/span\u003e\u003cspan address=\"10.4103/1673-5374.335838\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCanny E, Vansteensel MJ, van der Salm SMA, M\u0026uuml;ller-Putz GR, Berezutskaya J. Boosting brain-computer interfaces with functional electrical stimulation: potential applications in people with locked-in syndrome. J Neuroeng Rehabil. 2023;20(1):157. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12984-023-01272-y\u003c/span\u003e\u003cspan address=\"10.1186/s12984-023-01272-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIlves M, Lylykangas J, Rantanen V, M\u0026auml;kel\u0026auml; E, Vehkaoja A, Verho J, et al. Facial muscle activations by functional electrical stimulation. Biomed Signal Process Control. 2019;48:248\u0026ndash;54. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.bspc.2018.10.015\u003c/span\u003e\u003cspan address=\"10.1016/j.bspc.2018.10.015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM\u0026auml;kel\u0026auml; E, Venesvirta H, Ilves M, Lylykangas J, Rantanen V, Yl\u0026auml;-Kotola T, et al. Facial muscle reanimation by transcutaneous electrical stimulation for peripheral facial nerve palsy. J Med Eng Technol. 2019;43(3):155\u0026ndash;64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/03091902.2019.1637470\u003c/span\u003e\u003cspan address=\"10.1080/03091902.2019.1637470\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSoliman RS, Lee S, Eun S, Mohamed AZ, Lee J, Lee E, et al. Brain correlates to facial motor imagery and its somatotopy in the primary motor cortex. NeuroReport. 2017;28(5):285\u0026ndash;91. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/wnr.0000000000000758\u003c/span\u003e\u003cspan address=\"10.1097/wnr.0000000000000758\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMakary MM, Eun S, Park K. Greater corticostriatal activation associated with facial motor imagery compared with motor execution: a functional MRI study. NeuroReport. 2017;28(10):610\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/wnr.0000000000000809\u003c/span\u003e\u003cspan address=\"10.1097/wnr.0000000000000809\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOfner P, Schwarz A, Pereira J, M\u0026uuml;ller-Putz GR. Upper limb movements can be decoded from the time-domain of low-frequency EEG. PLoS ONE. 2017;12(8):e0182578. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0182578\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0182578\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu B, Wang Y, Deng L, Wu C, Zhang W, Li H, et al. Decoding Hand Movement Types and Kinematic Information From Electroencephalogram. IEEE Trans Neural Syst Rehabil Eng. 2021;29:1744\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/tnsre.2021.3106897\u003c/span\u003e\u003cspan address=\"10.1109/tnsre.2021.3106897\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSun Q, Merino EC, Yang L, Van Hulle MM. Unraveling EEG correlates of unimanual finger movements: insights from non-repetitive flexion and extension tasks. J Neuroeng Rehabil. 2024;21(1):228. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12984-024-01533-4\u003c/span\u003e\u003cspan address=\"10.1186/s12984-024-01533-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePfurtscheller G, Lopes da Silva FH. Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin Neurophysiol. 1999;110(11):1842\u0026ndash;57. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/s1388-2457(99)00141-8\u003c/span\u003e\u003cspan address=\"10.1016/s1388-2457(99)00141-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAkt\u0026uuml;rk T, de Graaf TA, Abra Y, Şahoğlu-G\u0026ouml;ktaş S, \u0026Ouml;zkan D, Kula A, et al. Event-related EEG oscillatory responses elicited by dynamic facial expression. Biomed Eng Online. 2021;20(1):41. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12938-021-00882-8\u003c/span\u003e\u003cspan address=\"10.1186/s12938-021-00882-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRayson H, Bonaiuto JJ, Ferrari PF, Murray L. Mu desynchronization during observation and execution of facial expressions in 30-month-old children. Dev Cogn Neurosci. 2016;19:279\u0026ndash;87. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.dcn.2016.05.003\u003c/span\u003e\u003cspan address=\"10.1016/j.dcn.2016.05.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhao M, Marino M, Samogin J, Swinnen SP, Mantini D. Hand, foot and lip representations in primary sensorimotor cortex: a high-density electroencephalography study. Sci Rep. 2019;9(1):19464. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-019-55369-3\u003c/span\u003e\u003cspan address=\"10.1038/s41598-019-55369-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eF\u0026eacute;ry YA. Differentiating visual and kinesthetic imagery in mental practice. Can J Exp Psychol. 2003;57(1):1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/h0087408\u003c/span\u003e\u003cspan address=\"10.1037/h0087408\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHardwick RM, Caspers S, Eickhoff SB, Swinnen SP. Neural correlates of action: Comparing meta-analyses of imagery, observation, and execution. Neurosci Biobehavioral Reviews. 2018;94:31\u0026ndash;44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neubiorev.2018.08.003\u003c/span\u003e\u003cspan address=\"10.1016/j.neubiorev.2018.08.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFarabbi A, Figueiredo P, Ghiringhelli F, Mainardi L, Sanches JM, Moreno P, et al. Investigating the impact of visual perspective in a motor imagery-based brain-robot interaction: A pilot study with healthy participants. Front Neuroergonomics. 2023;4\u0026ndash;2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnrgo.2023.1080794\u003c/span\u003e\u003cspan address=\"10.3389/fnrgo.2023.1080794\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKraeutner SN, Eppler SN, Stratas A, Boe SG. Generate, maintain, manipulate? Exploring the multidimensional nature of motor imagery. Psychol Sport Exerc. 2020;48:101673. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.psychsport.2020.101673\u003c/span\u003e\u003cspan address=\"10.1016/j.psychsport.2020.101673\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eElashmawi WH, Ayman A, Antoun M, Mohamed H, Mohamed SE, Amr H, et al. A Comprehensive Review on Brain\u0026ndash;Computer Interface (BCI)-Based Machine and Deep Learning Algorithms for Stroke Rehabilitation. Appl Sci. 2024;14(14):6347. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/app14146347\u003c/span\u003e\u003cspan address=\"10.3390/app14146347\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang C, Chen Z, Wang S, Zhang Z, Kong L, Chen X. The Impact of Visual and Kinesthetic Motor Imagery on Mental Fatigue and Classification Performance in Untrained Participants. Int J Human\u0026ndash;Computer Interact. 2025;41(8):4594\u0026ndash;608. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10447318.2024.2352224\u003c/span\u003e\u003cspan address=\"10.1080/10447318.2024.2352224\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMalouin F, Richards CL, Jackson PL, Lafleur MF, Durand A, Doyon J. The Kinesthetic and Visual Imagery Questionnaire (KVIQ) for assessing motor imagery in persons with physical disabilities: a reliability and construct validity study. J Neurol Phys Ther. 2007;31(1):20\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/01.npt.0000260567.24122.64\u003c/span\u003e\u003cspan address=\"10.1097/01.npt.0000260567.24122.64\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVrabec JT, Backous DD, Djalilian HR, Gidley PW, Leonetti JP, Marzo SJ, et al. Facial Nerve Grading System 2.0. Otolaryngol Head Neck Surg. 2009;140(4):445\u0026ndash;50. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.otohns.2008.12.031\u003c/span\u003e\u003cspan address=\"10.1016/j.otohns.2008.12.031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHerwig U, Satrapi P, Sch\u0026ouml;nfeldt-Lecuona C. Using the international 10\u0026ndash;20 EEG system for positioning of transcranial magnetic stimulation. Brain Topogr. 2003;16(2):95\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/b:brat.0000006333.93597.9d\u003c/span\u003e\u003cspan address=\"10.1023/b:brat.0000006333.93597.9d\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSouza-Couto D, Bretas R, Aversi-Ferreira TA. Neuropsychology of the parietal lobe: Luria\u0026rsquo;s and contemporary conceptions. Front NeuroSci. 2023;17\u0026ndash;2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnins.2023.1226226\u003c/span\u003e\u003cspan address=\"10.3389/fnins.2023.1226226\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eReisch LM, Wegrzyn M, Mielke M, Mehlmann A, Woermann FG, Bien CG, et al. Face processing and efficient recognition of facial expressions are impaired following right but not left anteromedial temporal lobe resections: Behavioral and fMRI evidence. Neuropsychologia. 2022;174:108335. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neuropsychologia.2022.108335\u003c/span\u003e\u003cspan address=\"10.1016/j.neuropsychologia.2022.108335\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHerlin B, Navarro V, Dupont S. The temporal pole: From anatomy to function\u0026mdash;A literature appraisal. J Chem Neuroanat. 2021;113:101925. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jchemneu.2021.101925\u003c/span\u003e\u003cspan address=\"10.1016/j.jchemneu.2021.101925\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuillot A, Collet C, Nguyen VA, Malouin F, Richards C, Doyon J. Brain activity during visual versus kinesthetic imagery: An fMRI study. Hum Brain Mapp. 2009;30(7):2157\u0026ndash;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/hbm.20658\u003c/span\u003e\u003cspan address=\"10.1002/hbm.20658\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKraeutner SN, Stratas A, McArthur JL, Helmick CA, Westwood DA, Boe SG. Neural and Behavioral Outcomes Differ Following Equivalent Bouts of Motor Imagery or Physical Practice. J Cogn Neurosci. 2020;32(8):1590\u0026ndash;606. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1162/jocn_a_01575\u003c/span\u003e\u003cspan address=\"10.1162/jocn_a_01575\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu X, Fan X, Dong J, Zhang X, Song Z, Bai D, et al. Enhancing motor imagery in the third-person perspective by manipulating sense of body ownership with virtual reality. Eur J Neurosci. 2024;60(7):5750\u0026ndash;63. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/ejn.16515\u003c/span\u003e\u003cspan address=\"10.1111/ejn.16515\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFarabbi A, Figueiredo P, Ghiringhelli F, Mainardi L, Sanches JM, Moreno P, et al. Investigating the impact of visual perspective in a motor imagery-based brain-robot interaction: A pilot study with healthy participants. Front Neuroergonomics. 2023;4\u0026ndash;2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnrgo.2023.1080794\u003c/span\u003e\u003cspan address=\"10.3389/fnrgo.2023.1080794\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuilbert J, Fernandez J, Molina M, Morin M-F, Alamargot D. Imagining handwriting movements in a usual or unusual position: effect of posture congruency on visual and kinesthetic motor imagery. Psychol Res. 2021;85(6):2237\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00426-020-01399-w\u003c/span\u003e\u003cspan address=\"10.1007/s00426-020-01399-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrusini L, Stival F, Setti F, Menegatti E, Menegaz G, Storti SF. A Systematic Review on Motor-Imagery Brain-Connectivity-Based Computer Interfaces. IEEE Trans Human-Machine Syst. 2021;51(6):725\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/THMS.2021.3115094\u003c/span\u003e\u003cspan address=\"10.1109/THMS.2021.3115094\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaviri SM, Vinjamuri R. Decoding motor execution and motor imagery from EEG with deep learning and source localization. Biomedical Eng Adv. 2025;9:100156. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.bea.2025.100156\u003c/span\u003e\u003cspan address=\"10.1016/j.bea.2025.100156\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa S, Zhang DA, Cross-Attention. -Based Class Alignment Network for Cross-Subject EEG Classification in a Heterogeneous Space. Sensors. 2024;24(21):7080. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s24217080\u003c/span\u003e\u003cspan address=\"10.3390/s24217080\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu R, Jiang N, Mrachacz-Kersting N, Lin C, As\u0026iacute;n Prieto G, Moreno JC, et al. A closed-loop brain-computer interface triggering an active ankle-foot orthosis for inducing cortical neural plasticity. IEEE Trans Biomed Eng. 2014;61(7):2092\u0026ndash;101. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/tbme.2014.2313867\u003c/span\u003e\u003cspan address=\"10.1109/tbme.2014.2313867\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHohman MH, Kim SW, Heller ES, Frigerio A, Heaton JT, Hadlock TA. Determining the threshold for asymmetry detection in facial expressions. Laryngoscope. 2014;124(4):860\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/lary.24331\u003c/span\u003e\u003cspan address=\"10.1002/lary.24331\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFang T, Wang J, Mu W, Song Z, Zhang X, Zhan G, et al. Noninvasive neuroimaging and spatial filter transform enable ultra low delay motor imagery EEG decoding. J Neural Eng. 2022;19(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1088/1741-2552/aca82d\u003c/span\u003e\u003cspan address=\"10.1088/1741-2552/aca82d\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu T, Li B, Zhang C, Chen P, Zhao W, Yan B. Real-Time Classification of Motor Imagery Using Dynamic Window-Level Granger Causality Analysis of fMRI Data. Brain Sci. 2023;13(10). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/brainsci13101406\u003c/span\u003e\u003cspan address=\"10.3390/brainsci13101406\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDelavari F, Santaniello S. Role of Scalp EEG Brain Connectivity in Motor Imagery Decoding for BCI Applications. Annu Int Conf IEEE Eng Med Biol Soc. 2024;2024:1\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/embc53108.2024.10781532\u003c/span\u003e\u003cspan address=\"10.1109/embc53108.2024.10781532\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuroengineering-and-rehabilitation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jner","sideBox":"Learn more about [Journal of NeuroEngineering and Rehabilitation](http://jneuroengrehab.biomedcentral.com/)","snPcode":"12984","submissionUrl":"https://submission.nature.com/new-submission/12984/3","title":"Journal of NeuroEngineering and Rehabilitation","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Brain-computer interfaces, Motor imagery, Electroencephalography, Event-related desynchronization/synchronization, Facial paralysis","lastPublishedDoi":"10.21203/rs.3.rs-7908162/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7908162/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eFunctional and aesthetic deficits in individuals with facial nerve paralysis (FNP) significantly impair their quality of life. By decoding motor intentions and controlling rehabilitation devices, motor imagery (MI)-based brain-computer interfaces can improve outcomes in people with peripheral paralysis. However, the electroencephalography (EEG) features underlying different facial MIs and their decodability remain unclear. This study aims to investigate the feasibility of achieving accurate decoding of facial MIs related to major mimetic muscles and the corresponding decoding strategies.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eAfter comparing block and event-related designs to identify the appropriate paradigm for facial MIs, 20 healthy participants performed four types of facial MIs (eyebrow raising, eye closing, lip puckering and grinning) in two modalities: kinesthetic and visual, from which event-related desynchronization/synchronization (ERD/S) features were extracted using time-frequency analysis. A deep learning model integrating a temporal convolutional network with a spatial attention mechanism was then developed for both within-subject and cross-subject decoding, thereby identifying the contribution of each EEG channel. Finally, the model was further evaluated on EEG data from six individuals with FNP.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eParticipants showed better performance in the block design, in which facial MIs induced significant ERD in the low-frequency band in the left prefrontal and right central-temporal regions, co-occurring with shorter and weaker ERS in higher frequencies. Regarding MI decoding in healthy participants, the model achieved the highest average accuracy of 85.17% in within-subject classification of kinesthetic MI, with EEG features from the left frontal and parietal regions contributing most to decoding. Combining these findings, the model obtained an average accuracy of 76.46% on patients\u0026rsquo; data, with half the number of MI tasks and 25% fewer EEG channels.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eThis study demonstrated that major mimetic muscle-related MIs can be accurately recognized from EEG using deep learning, with a suitable decoding strategy involving within-subject decoding of kinesthetic MI collected through a block design.\u003c/p\u003e","manuscriptTitle":"Decoding motor imagery related to major mimetic muscles from electroencephalography","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-10 16:19:45","doi":"10.21203/rs.3.rs-7908162/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-12-13T05:22:42+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-08T12:02:43+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"297694717343204518564217589940241621993","date":"2025-11-17T14:14:57+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-14T09:30:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"82646307899594285845475474523853633763","date":"2025-10-30T11:50:32+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-29T15:22:16+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-22T08:42:13+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-22T08:38:58+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of NeuroEngineering and Rehabilitation","date":"2025-10-20T18:15:19+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuroengineering-and-rehabilitation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jner","sideBox":"Learn more about [Journal of NeuroEngineering and Rehabilitation](http://jneuroengrehab.biomedcentral.com/)","snPcode":"12984","submissionUrl":"https://submission.nature.com/new-submission/12984/3","title":"Journal of NeuroEngineering and Rehabilitation","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ef5bbf0a-f5c9-4b39-8650-b33667da74ac","owner":[],"postedDate":"November 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-22T16:08:16+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-10 16:19:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7908162","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7908162","identity":"rs-7908162","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}