Brain-Muscle Atlas: A novel framework for Motor Brain-Computer Interfaces

Motor brain-computer interfaces (BCIs) enable the control of external devices by decoding neural signals. However, most existing systems rely on a direct “brain-machine” mapping, overlooking the hierarchical physiological pathway of natural movement, namely the “brain-muscle-joint” cascade. Due to the lack of explicit modeling and enhancement of this pathway, current systems are often constrained by the low amplitude and high noise of EEG signals, resulting in motor outputs that are unstable, discontinuous, and insufficiently natural.To address these limitations, this study introduces the concept of a brain-muscle atlas, designed to systematically characterize the mapping between motor cortical activity and corresponding muscle activation, thereby establishing a movement decoding framework that better aligns with neuromuscular physiology. Using synchronously recorded EEG-EMG data, we constructed the first brain-muscle atlas for elbow flexion-extension, achieving a structured mapping from cortical activity to muscle activation.Offline experiments demonstrate that the proposed atlas accurately reconstructs the temporal activation patterns of primary elbow agonists, achieving a maximum correlation coefficient of 0.8314, thereby validating its ability to capture cortical-muscular mapping. Furthermore, by leveraging atlas-derived muscle activation representations, we enabled continuous real-time control of a virtual elbow joint. All ten participants successfully completed the online flexion-extension task, indicating that the system robustly extracts motor intent even under low-SNR EEG conditions.

💡 Research Summary

The paper addresses a fundamental limitation of non‑invasive brain‑computer interfaces (BCIs): the direct “brain‑to‑machine” mapping that ignores the natural hierarchical motor pathway consisting of cortex, muscles, and joints. Because EEG signals are low‑amplitude and noisy, direct decoding often yields unstable, discontinuous, and unnatural control. To overcome this, the authors propose a “brain‑muscle atlas” (BMA) that explicitly models the transformation from motor cortical activity to muscle activation, thereby re‑introducing the physiological relay that the nervous system normally provides.

Data acquisition and preprocessing
Sixteen‑channel EEG (g.Nautilus, 500 Hz) was recorded over sensorimotor areas using the 10‑10 system, while twelve EMG channels (1000 Hz) covered the major elbow flexor and extensor muscles on both arms. A low‑rate force sensor captured elbow torque. EEG was re‑referenced with a common average reference (CAR) and band‑pass filtered (15‑35 Hz). EMG was filtered (20‑450 Hz) and notch‑filtered (48‑52 Hz) to remove power‑line interference. All signals were temporally aligned and segmented into 60 trials per participant, with only task‑related portions (identified via a force threshold) retained for analysis.

Construction of the brain‑muscle atlas
The core of the BMA is a sliding‑window Transformer model. Each window (e.g., 200 ms) overlaps with its neighbors, preserving temporal continuity while allowing the self‑attention mechanism to capture long‑range dependencies within the EEG stream. The Transformer encoder processes the EEG segment; the decoder predicts the corresponding EMG time series. Supervision comes from the simultaneously recorded EMG, enabling the model to learn a cross‑modal mapping that translates EEG patterns into “virtual muscle activation” signals. Training minimizes a combination of mean‑square error and a temporal‑alignment loss (e.g., dynamic time warping) to encourage realistic dynamics.

Offline validation
After training, the BMA was used to reconstruct EMG from EEG alone. Correlation between reconstructed and true EMG was quantified with Spearman’s rank correlation coefficient (SCC) and Pearson’s correlation coefficient (PCC). Across all participants and channels, the mean SCC was 0.28, with the strongest channels (biceps and triceps) reaching SCC ≈ 0.34‑0.39. The maximum PCC observed for any channel was 0.8314, indicating that the atlas can faithfully reproduce the temporal profile of primary elbow agonists. A zero‑masking perturbation analysis showed that frontal‑parietal electrodes contributed most to the reconstruction, confirming that the model leverages physiologically relevant cortical regions rather than exploiting spurious statistical regularities.

Brain‑muscle‑elbow interface (BMEI) and online control
To demonstrate real‑time utility, the authors built a BMEI that operates solely on EEG during the control phase. They introduced a hybrid Direction‑Proportional EMG (DP‑EMG) strategy: a convolutional neural network (CNN) first classifies movement direction (flexion vs. extension) from the virtual EMG output, while the amplitude of the most reliable muscle channel (typically the one representing the dominant direction) is used as a proportional control signal. This combination yields discrete directional decisions with smooth, continuous force modulation.

Ten participants performed a virtual elbow flexion‑extension task using only EEG‑driven control. All completed the task, and performance metrics (trajectory error, smoothness, latency) were markedly better than those obtained with conventional end‑to‑end EEG‑to‑kinematic decoders. The system produced continuous, stable joint commands despite the low SNR of the EEG input, confirming that the muscle‑level representation acts as an effective signal‑to‑noise enhancer.

Contributions and significance

1. First brain‑muscle atlas for a single joint – a computable, verifiable mapping from cortical activity to peripheral muscle activation.
2. Novel BCI architecture – the brain‑muscle‑elbow interface replaces the direct brain‑to‑machine pipeline with a physiologically grounded relay, improving naturalness, continuity, and robustness of control.
3. Demonstration of Transformer‑based cross‑modal decoding – shows that attention mechanisms can handle noisy, non‑stationary biosignals and learn meaningful neuro‑muscular relationships.
4. Proof‑of‑concept for future fully paralyzed users – although EMG was used only during training, the trained atlas enables EEG‑only control, suggesting a pathway toward assistive devices for patients lacking any residual muscle activity.

Limitations and future directions
The study is limited to healthy right‑handed participants and a single elbow joint. Extending the atlas to multi‑joint or whole‑body movements will require larger datasets and possibly hierarchical atlases. For completely paralyzed patients, alternative supervisory signals (e.g., imagined EMG, proprioceptive feedback) must replace real EMG during training. Finally, the Transformer model’s computational load may hinder deployment on low‑power embedded hardware; model compression or lightweight alternatives will be needed for practical assistive systems.

In summary, by re‑introducing the muscle as an explicit computational layer, the brain‑muscle atlas bridges the gap between low‑quality EEG and high‑fidelity motor intent, delivering continuous, stable control of a virtual joint. This work establishes a new paradigm for biologically informed BCI design and opens avenues for more natural, robust neuro‑prosthetic control.