Muscle Network Analysis for Adaptive Prosthetic and Wearable Systems

EMG-driven Ankle Prosthesis Control

Study Design

Participants:

• Below-knee amputees (TB group): 5 participants

• Healthy Able-body Participants (AB group): 5 participants.
  

Conditions:

• Pre-training: Baseline EMG, IMU, motion capture, and force plate data were collected during postural sway tasks using EMG controlled ankle prosthesis to assess initial coordination.

• Post-training: The same measurements were repeated after 4 weeks of training with an EMG-controlled robotic ankle prosthesis to evaluate adaptation.

• Reference condition: Able-bodied (AB) participants performing identical movement tasks for network comparison.
  

Data Analysis

The dataset included 11-channel surface EMG recordings from lower-limb muscles, synchronized with motion capture, IMU, and force plate data. The analysis pipeline combined signal processing, intermuscular coherence analysis, non-negative matrix factorization (NNMF), and graph-theory network analysis to characterize neuromuscular connectivity and coordination patterns. These methods quantified how muscle networks reorganized with EMG-controlled prosthesis training. The workflow integrated MATLAB-based custom algorithms, computational modeling, and data visualization tools.

Results

Muscle Connectivity Patterns

I analyzed how muscle coordination changed before and after participants trained with the EMG-controlled ankle prosthesis. The data revealed four main patterns of connectivity, each linked to a different type of neuromuscular control process:

• Component A – Slow movement-related oscillations:

  After training, participants showed more connections between muscles on both legs, suggesting improved overall balance and new muscle coordination pathways.

  
  

• Component B – Cortical involvement in coordination:

  This pattern showed stronger and more widespread connections, especially between muscles on opposite sides of the body. It indicates that participants relied more on brain-driven coordination after training, reflecting improved motor planning.

  
  

• Component C – Cortical control:

  Connections in this component became more focused after training, showing more precise, task-dependent control. This suggests that users learned to fine-tune muscle activation during the balance task.

  
  

• Component D – Fast corticospinal corrections:

  After training, local (within-limb) muscle connections became stronger. This points to faster, automatic spinal-level corrections that stabilize movement.

Together, these results show that training with the EMG-controlled prosthesis led to a shift from broad, compensatory coordination toward more efficient and specialized muscle interactions.

Network-Level Changes

Graph theory analysis quantified how these muscle networks reorganized:

• Mean Clustering:

  Reflects how tightly muscle groups work together. Clustering decreased slightly after training, suggesting that muscle coordination became more distributed rather than redundant each muscle contributed more uniquely to control.

  
  

• Global Efficiency:

  Measures how easily signals travel across the network. Efficiency increased after training, meaning communication between muscle groups became faster and more effective.

  
  

• Mean Betweenness:

  Indicates how much certain muscles act as “communication hubs.” Betweenness dropped in some frequency components after training, implying a more balanced distribution of control rather than overreliance on a few key muscles. Comparison to Healthy Participants

When compared to able-bodied individuals, the post-training networks of amputee participants became more similar to healthy muscle connectivity patterns. This indicates that EMG-controlled prosthesis training may promote neural and muscular adaptation that restores more natural coordination strategies.