# Cross-Modal Synergy Representation of EMG and Joint Angular Acceleration During Gait in Parkinson’s Disease Using NMF and Multimodal Matrix Factorization

## TL;DR

This study uses advanced data analysis to uncover how muscle activity and joint movements are linked during walking in Parkinson’s disease patients.

## Contribution

A novel joint framework combining EMG NMF and cross-modal MMF to characterize gait patterns in Parkinson’s disease.

## Key findings

- Four muscle synergies were identified, with ankle-related muscles playing a dominant role.
- Eight cross-modal synergies were found, linking muscle activity with joint angular acceleration dynamics.
- The framework provides a low-dimensional method to describe gait-related neuromuscular patterns in Parkinson’s disease.

## Abstract

Parkinson’s disease often affects walking, increasing the risk of falls and reducing independence and quality of life. To improve understanding of how walking is controlled in Parkinson’s disease, we studied 19 people while they walked at a natural pace. We recorded electrical activity from key leg muscles using skin sensors and measured how quickly the pelvis, hip, knee, and ankle joints sped up or slowed down during the walking cycle. We then used a computer method that identifies a small set of repeating patterns in complex data and another method that links muscle patterns and movement patterns using the same timing structure. We found that four main muscle patterns could describe most of the muscle activity, and these patterns were strongly influenced by ankle-related muscles. When muscle signals and joint speed-up and slow-down signals were analyzed together, eight combined patterns were identified, showing how changes in muscle activity co-varied with joint angular-acceleration dynamics across the whole body, from the pelvis to the ankle. This framework offers a clearer, measurable way to describe gait-related neuromuscular and kinematic patterns in this Parkinson’s disease cohort and may help future work validate these patterns against matched healthy references and clinically meaningful tasks.

The aims of this research were to characterize neuromuscular control features within the gait cycle in Parkinson’s disease (PD) from the perspectives of muscle synergies and cross-modal coupling and to propose a joint representation of the relationship between muscle activation patterns and kinematic dynamic outputs. PD participants (n = 19) were included. Lower-limb surface electromyography (EMG) and kinematic dynamic channels, including pelvic/hip, knee, and ankle angular acceleration, were collected during level-ground natural walking. EMG signals were first decomposed using non-negative matrix factorization (NMF) to extract muscle synergies, and the number of synergies was evaluated using reconstruction performance (R2). Multimodal matrix factorization (MMF) was then applied to jointly decompose the EMG and angular-acceleration channels, yielding a cross-modal synergy representation comprising a shared temporal structure (H) and modality-specific weight structures (W): non-negativity was imposed on EMG weights, whereas kinematic weights were allowed to take positive and negative values to encode directional contributions. Under the current task and muscle set, NMF achieved high EMG reconstruction performance with four synergies (R2 = 0.882). The synergy weights showed an ankle-dominant pattern: tibialis anterior (TA) consistently carried high weights across multiple synergies, while lateral gastrocnemius (LG) and soleus (SOL) contributed prominently to another synergy. The synergy activation profiles exhibited phase-dependent fluctuations with multiple rises and falls across the gait cycle, suggesting that synergy output was primarily characterized by continuous modulation rather than single-peak recruitment. MMF further identified eight cross-modal synergies, simultaneously capturing the shared contributions of key muscle groups (e.g., RF, TA, and SOL) and pelvic/hip and knee/ankle angular-acceleration channels within the same decomposition framework and summarizing their descriptive co-variation through the shared temporal structure (H). Overall, A low-dimensional synergy analysis combining EMG-only NMF with cross-modal MMF enables simultaneous characterization of cohort-level modular organization of muscle activity during gait and its descriptive association with pelvis-to-lower-limb dynamic output. This joint framework provides a methodological basis for quantitatively describing gait-related modular organization and temporal modulation patterns in this PD cohort under natural level-ground walking and lays the groundwork for subsequent testing of associations between synergy features and gait phenotypes, clinical severity, and rehabilitation responses.