Surface EMG Signals

After having created the MUAPs, we can finally simulate the surface EMG by creating a surface EMG model.

Note

The surface EMG signals are the summation of the MUAPs at the surface of the skin.

In Myogen, we can simulate the surface EMG by convolving the MUAPs with the spike trains of the motor units.

import shutil

Import Libraries

from pathlib import Path

import joblib
import matplotlib
import numpy as np
import seaborn as sns
from matplotlib import pyplot as plt

Load Muscle Model

save_path = Path("./results")
muscle = joblib.load(save_path / "muscle_model.pkl")

Load Motor Neuron Pool

motor_neuron_pool = joblib.load(save_path / "motor_neuron_pool.pkl")

Load Surface EMG Model

Note

Because of computation time, we pre-computed the surface EMG model for 100 motor units.

surface_emg = joblib.load(save_path / "surface_emg_100.pkl")

Generate Surface EMG

To simulate the surface EMG, we need to run the simulate_surface_emg method of the SurfaceEMG object.

surface_emg_signals = surface_emg.simulate_surface_emg(
    motor_neuron_pool=motor_neuron_pool
)

print(f"Surface EMG simulation completed!")
print(f"Generated EMG shape: {surface_emg_signals.shape}")
print(f"  - {surface_emg_signals.shape[0]} pools")
print(f"  - {surface_emg_signals.shape[1]} electrode rows")
print(f"  - {surface_emg_signals.shape[2]} electrode columns")
print(f"  - {surface_emg_signals.shape[3]} time samples")

# Save the surface EMG results
joblib.dump(surface_emg_signals, save_path / "surface_emg_signals.pkl")

Simulating surface EMG (CPU):   0%|          | 0/2 [00:00<?, ?it/s]
Simulating surface EMG (CPU):  50%|█████     | 1/2 [00:22<00:22, 22.18s/it]
Simulating surface EMG (CPU): 100%|██████████| 2/2 [00:39<00:00, 19.39s/it]
Simulating surface EMG (CPU): 100%|██████████| 2/2 [00:39<00:00, 19.81s/it]
Surface EMG simulation completed!
Generated EMG shape: (1, 2, 5, 5, 40000)
  - 1 pools
  - 2 electrode rows
  - 5 electrode columns
  - 5 time samples

['results/surface_emg_signals.pkl']

Visualize Surface EMG Results

Note

Since MyoGen is a simulator, the results will have no real-world noise.

We can add noise to the surface EMG signals to make them more realistic.

For this the method add_noise is used.

surface_emg.add_noise(snr_db=10.0)

# Load input current matrix
input_current_matrix = joblib.load(save_path / "input_current_matrix.pkl")
shutil.rmtree(matplotlib.get_cachedir())
with plt.xkcd():
    plt.rcParams.update({"font.size": 24})
    # Create single plot with normalized signals
    fig, ax = plt.subplots(figsize=(12, 6))
    # Get the signals
    emg_signal = surface_emg.noisy_surface_emg[0, 0, 3, 3]
    current_signal = input_current_matrix[0]  # Only first current

    # Normalize EMG by dividing by maximum
    emg_normalized = emg_signal / np.max(emg_signal)

    # Normalize current between 0 and 1
    current_normalized = (current_signal - np.min(current_signal)) / (
        np.max(current_signal) - np.min(current_signal)
    )

    # Plot both normalized signals on same axis
ax.plot(
    motor_neuron_pool.times / 1000.0,
    emg_normalized,
    linewidth=2,
    label="Surface EMG",
)

with plt.xkcd():
    ax.plot(
        motor_neuron_pool.times / 1000.0,
        current_normalized,
        linewidth=2,
        label="Input Current",
        alpha=0.7,
    )

    ax.set_xlabel("Time (s)")
    ax.set_ylabel("Normalized Amplitude")
    ax.grid(True, alpha=0.3)
    ax.legend()

    sns.despine(trim=True, left=False, bottom=False, right=True, top=True, offset=5)

    plt.title("Normalized Surface EMG and Input Current")

plt.tight_layout()
plt.show()

Target SNR: 10.00 dB
Achieved SNR: 10.00 dB