Two-source Validation of Online Surface EMG Decomposition Using Progressive FastICA Peel-off

Neurology Electronic Information Engineering Information Science Life Sciences Bioengineering Medicine
surface electromyogram online decomposition progressive FastICA peel-off two-source validation motor unit

Two-Source Validation Study of Online Surface Electromyogram Decomposition

Academic Background

Surface electromyogram (SEMG) signals are crucial representations of muscle activity and are widely used in fields such as sports rehabilitation, robotic control, and human-machine interaction. However, SEMG signals are challenging to decompose due to their low signal-to-noise ratio, high similarity, and severe waveform superposition. In recent years, advancements in electronic and sensing technologies have enabled the collection of high-density surface electromyogram (HD SEMG) signals. Blind source separation (BSS) techniques such as convolution kernel compensation (CKC) and progressive FastICA peel-off (PFP) have shown significant progress in SEMG decomposition. Nevertheless, most current studies on online SEMG decomposition primarily validate using simulated signals, lacking comprehensive evaluation on experimental data. Therefore, this study aims to conduct a two-source validation by simultaneously collecting intramuscular EMG (IEMG) and HD SEMG signals to comprehensively assess the performance of online SEMG decomposition.

Paper Source

This paper was co-authored by Haowen Zhao, Maoqi Chen, Yunfei Liu, Xiang Chen, Ping Zhou, and Xu Zhang. The authors are affiliated with the School of Microelectronics at the University of Science and Technology of China and the Faculty of Biomedical and Rehabilitation Engineering at Qingdao University. The paper has been published in the journal IEEE Transactions on Biomedical Engineering, with a publication date of 2025, and the DOI number is 10.1109/TBME.2025.3538338.

Research Workflow

1. Experimental Design and Data Collection

The study recruited five healthy subjects (aged 36 ± 8 years), all without known muscle injuries or neuromuscular disorders. The experimental protocol adhered to the Declaration of Helsinki and was approved by the Committee for the Protection of Human Subjects (CPHS) at the University of Texas Health Science Center at Houston. The experiment involved simultaneous collection of SEMG and IEMG signals from the first dorsal interosseous (FDI) muscle of the dominant hand of each subject.

  • SEMG Acquisition: An 8 × 8 flexible two-dimensional electrode array (TMS International BV, Netherlands) was used to collect SEMG signals, with the reference electrode located near the elbow. Signals were recorded using a REFA 128-channel amplification system (TMS International BV, Netherlands) at a sampling frequency of 2 kHz, with a band-pass filter set at 10 Hz–500 Hz.
  • IEMG Acquisition: IEMG signals were collected using the Natus UltraPro S100 EMG system (Natus Neurology Inc., USA) and a conventional concentric needle electrode (diameter 0.58 mm, recording area 0.07 mm²). The reference electrode was attached to the back of the hand. Signal sampling frequency was set at 44.1 kHz, with a band-pass filter set at 10 Hz–10 kHz.

2. Signal Synchronization and Experimental Procedure

During the experiment, subjects were seated comfortably with the tested arm pronated and placed on a height-adjustable table. A resistance force was provided to the muscle, and subjects were instructed to perform an isometric contraction at a low-force level. Once obvious motor unit (MU) activity appeared on the screen, subjects were asked to maintain the contraction for at least 30 seconds. Each subject performed approximately 15 trials, with sufficient rest provided between consecutive trials. To ensure synchronization of SEMG and IEMG signals, an electrical stimulation protocol was applied in each trial. After completing a single trial, the ulnar nerve of the subject was stimulated about 2 cm proximal to the wrist crease to produce a visible M-wave response, allowing alignment of the recorded stimulus artifacts.

3. Data Analysis

The study employed a two-source validation method to independently decompose SEMG and IEMG signals.

  • SEMG Decomposition: Real-time processing was conducted using the online PFP method. A series of MU separation vectors were initialized offline using the Automatic PFP (APFP) method. During the online stage, these vectors were directly utilized to estimate source signals, and successive multi-threshold Otsu algorithms were applied to precisely extract MU spike trains (MUSTs).
  • IEMG Decomposition: A simplified version of the PFP method was used, combining the peel-off strategy with valley-seeking clustering. First, MUs with distinct amplitudes were extracted, followed by distinguishing superimposed MUs using valley-seeking clustering. The reliability of the decomposition results was assessed through various metrics (e.g., coefficient of variation of spike amplitudes and inter-spike intervals) to ensure accuracy in two-source validation.

4. Performance Evaluation

The study evaluated the performance of online SEMG decomposition using matching rate (MR), false negative rate (FNR), and false discovery rate (FDR). The MR was calculated as:

[ MR = \frac{2N_{com}}{N_g + N_s} ]

where (N_g) represents the number of firing events from the ground-truth reference, (Ns) represents the number of firing events from SEMG decomposition results, and (N{com}) indicates the number of common discharge events. FNR and FDR represent the proportions of “missing” and “wrong” discharge events, respectively.

Key Results

The study analyzed 50 trials of SEMG and IEMG signals, identifying 549 MUs from SEMG signals and 92 MUs from IEMG signals (used as the ground-truth reference). All MUs decomposed from IEMG signals matched those identified in online SEMG decomposition. The average MR during the online stage was (96 ± 1)%, with an FDR of 0.05 ± 0.02 and an FNR of 0.03 ± 0.01. The offline stage showed slightly better performance, with an MR of (99 ± 0.6)%, an FDR of 0.02 ± 0.01, and an FNR of 0.02 ± 0.01.

Conclusion and Significance

This study comprehensively evaluated the performance of online SEMG decomposition through two-source validation, demonstrating that the two-stage approach based on the PFP method can continuously and accurately track the same MUs in experimental SEMG signals. The findings provide strong evidence of the reliability of online SEMG decomposition, offering significant scientific and practical value, especially in fields such as sports rehabilitation, robotic control, and human-machine interaction.

Highlights of the Study

  1. Two-Source Validation: Comprehensive evaluation of online SEMG decomposition performance using simultaneously collected SEMG and IEMG signals, with IEMG decomposition results serving as the ground-truth reference.
  2. High Matching Rate: The matching rate between online decomposition results and ground-truth reference reached 96%, demonstrating the ability of separation vectors to continuously track MUs.
  3. Innovative Methodology: The use of a simplified PFP method combined with valley-seeking clustering for IEMG decomposition improved the accuracy and reliability of the decomposition process.

Other Valuable Information

The study also noted that future research could further optimize online decomposition algorithms, such as improving performance through local batch optimization of separation vectors. Additionally, increasing the number of IEMG recording channels could capture more common MUs, enabling more comprehensive validation.

Through this report, readers can gain a deeper understanding of the latest advances in online SEMG decomposition and its application value in experimental data.