EMG-Centered Multisensory Based Technologies for Pattern Recognition in Rehabilitation: State of the Art and Challenges

Chaoming Fang, Bowei He, Yixuan Wang, Jin Cao, Shuo GaoBiosensors 2020 · DOI

TL;DR

This review paper explores the exciting field of EMG-centered multisensory fusion for rehabilitation, highlighting how combining muscle activity signals (EMG) with other sensor data (kinematics and kinetics) significantly improves the accuracy and robustness of recognizing human intentions and movements. It provides a comprehensive overview of current techniques, applications, and challenges, paving the way for more effective assistive technologies and diagnostic tools in rehabilitation.

What Problem Does This Paper Address?

The core challenge this paper addresses is the inherent limitation of using electromyography (EMG) signals alone for accurate and reliable pattern recognition in rehabilitation. While EMG is a direct window into muscle activity and user intention, it struggles with:

  • Recognizing subtle movements: Slight body movements, especially in fine motor tasks like finger control, produce weak EMG signals that are difficult to differentiate.
  • Environmental variability: EMG signals are highly susceptible to external factors, leading to inconsistent detection accuracy.
  • Limited information: EMG primarily reflects muscle activation, not the full picture of movement dynamics (e.g., joint angles, forces, or external interactions).

These limitations hinder the development of sophisticated rehabilitation technologies like advanced prosthetics, exoskeletons, and diagnostic tools that require precise understanding of human movement and intent. The paper argues that integrating EMG with other sensor modalities is crucial to overcome these challenges.

Methods

The paper provides a detailed review of the entire EMG pattern recognition pipeline, emphasizing multisensory fusion:

  • EMG Signal Generation & Physiology: Explains the basic biological process of how muscles generate electrical signals.
  • Signal Preprocessing: Covers essential steps like filtering (e.g., bandpass filters) and windowing to clean and segment raw EMG data.
  • Feature Extraction: Discusses various methods to extract meaningful information from EMG signals, including:
  • Time-domain features: Mean Absolute Value (MAV), Root Mean Square (RMS), Zero Crossings (ZC), etc.
  • Frequency-domain features: Mean Frequency (MNF), Median Frequency (MDF), etc.
  • Autoregressive (AR) features: Capturing temporal dependencies.
  • Handcrafted features: Leveraging biomechanical knowledge.
  • Dimensionality Reduction: Explores techniques to manage high-dimensional feature spaces, such as:
  • Feature Projection: Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF).
  • Feature Selection: Identifying the most informative features.
  • Classification Algorithms: Reviews common classifiers used for EMG data, including Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Convolutional Neural Networks (CNN).
  • Multisensory Fusion Strategies: This is the core of the paper, detailing how EMG is combined with other sensors:
  • Fusion with Kinematic Sensors: Integrating data from accelerometers, gyroscopes, and IMUs to capture motion information (joint angles, velocities, acceleration).
  • Fusion with Kinetic Sensors: Incorporating data from force sensors and ground reaction force (GRF) sensors to measure forces and interactions.
  • Fusion with Both Kinematics and Kinetics: Combining all three types of information for a comprehensive understanding.

Datasets

The paper does not introduce new datasets. Instead, it reviews existing research that utilizes various datasets for training and evaluating their proposed methods. The authors highlight a general lack of large, open-source, and standardized EMG datasets, which they identify as a significant challenge.

Key Insights and Ideas

  • EMG Alone is Insufficient for Complex Rehabilitation Tasks: The paper strongly emphasizes that while EMG is valuable, it's often not enough for nuanced tasks like recognizing fine finger movements or adapting to varied terrains.
  • Multisensory Fusion is the Key to Enhanced Accuracy: Combining EMG with kinematic (motion) and kinetic (force) data significantly boosts recognition accuracy and robustness, especially in challenging scenarios.
  • Different Sensor Combinations Suit Different Applications: Kinematic sensors are great for capturing overall limb movement, while kinetic sensors are crucial for understanding interaction with the environment. The optimal fusion strategy depends on the specific rehabilitation goal.
  • High-Density EMG (HD-EMG) Shows Promise for Fine Movements: HD-EMG, with its dense electrode arrays, can capture more detailed muscle activity, potentially enabling recognition of very subtle movements and reducing the impact of EMG crosstalk.
  • Data Quality is Paramount: The paper repeatedly stresses that low-quality EMG data, due to noise, crosstalk, or electrode shift, is a major bottleneck for accurate multisensory fusion.

Key Results

The paper summarizes results from various studies, indicating significant improvements with multisensory fusion:

  • Locomotion Mode Recognition: Fusion strategies achieved accuracies of 95% or higher in some studies, outperforming EMG-only approaches. For example, one study reported 99% accuracy for gait phase recognition using EMG and GRF.
  • Hand Gesture Recognition: Combining EMG with IMU data improved accuracy to 94% in one application.
  • Prosthetic Control: Fusion approaches have demonstrated accuracies over 95% for locomotion mode identification, significantly better than traditional methods.
  • Disease Diagnosis: Multisensory fusion has achieved diagnostic accuracy rates of over 90% for conditions like Parkinson's disease.

Limitations and Future Work

The authors identify several key challenges and propose future directions:

  • Low Data Quality: EMG signals are inherently noisy and susceptible to interference (crosstalk, electrode shift).
  • Future Work: Development of advanced sensing techniques like High-Density EMG (HD-EMG) to improve signal quality and spatial resolution.
  • Inadequate and Undisclosed Data: A lack of large, standardized, and publicly available datasets hinders research reproducibility and algorithm development.
  • Future Work: Establishment of "big data-based EMG" initiatives, creation of open-source databases, and promotion of data sharing through "data papers."
  • Discrete Interpretation of Continuous Movements: Current methods often struggle with real-time, continuous recognition of dynamic movements, leading to delays.
  • Future Work: Development of better transition logic between movement patterns, predictive models, and overlapping windowing techniques.
  • Need for Robust and Flexible Systems: Current systems are often application-specific and lack generality.
  • Future Work: Integration of biomechanical models with pattern recognition for more intuitive and adaptable control.

The paper concludes that multisensory EMG PR holds immense potential for rehabilitation, driven by advancements in sensing technology, data availability, and algorithm development.

Summary updated

Abstract

In the field of rehabilitation, the electromyography (EMG) signal plays an important role in interpreting patients' intentions and physical conditions. Nevertheless, utilizing merely the EMG signal suffers from difficulty in recognizing slight body movements, and the detection accuracy is strongly influenced by environmental factors. To address the above issues, multisensory integration-based EMG pattern recognition (PR) techniques have been developed in recent years, and fruitful results have been demonstrated in diverse rehabilitation scenarios, such as achieving high locomotion detection and prosthesis control accuracy. Owing to the importance and rapid development of the EMG centered multisensory fusion technologies in rehabilitation, this paper reviews both theories and applications in this emerging field. The principle of EMG signal generation and the current pattern recognition process are explained in detail, including signal preprocessing, feature extraction, classification algorithms, etc. Mechanisms of collaborations between two important multisensory fusion strategies (kinetic and kinematics) and EMG information are thoroughly explained; corresponding applications are studied, and the pros and cons are discussed. Finally, the main challenges in EMG centered multisensory pattern recognition are discussed, and a future research direction of this area is prospected.
BibTeX
@inproceedings{fang2020emg,
  author = {Chaoming Fang and Bowei He and Yixuan Wang and Jin Cao and Shuo Gao},
  title = {{EMG-Centered Multisensory Based Technologies for Pattern Recognition in Rehabilitation: State of the Art and Challenges}},
  booktitle = {Biosensors},
  year = {2020},
  doi = {10.3390/bios10080085},
}
Figure