Surface diaphragmatic EMG extraction from noisy recordings

• The diaphragmatic electromyogram (EMGdi) carries critical information about respiration and can monitor respiratory conditions.¹
• Noninvasive surface EMGdi (sEMGdi) recorded from chest electrodes is convenient but weak and difficult to extract from noise, limiting clinical use versus esophageal EMGdi.¹
• A modified progressive FastICA peel-off (PFP) framework has been proposed to extract weak sEMGdi from noisy surface recordings for physiological monitoring.^{1 1}

Gardner updates

• A modified progressive FastICA peel-off (PFP) framework extracts weak surface diaphragmatic EMG (sEMGdi) from noisy chest recordings to support noninvasive respiratory monitoring. ¹

Weekly enrichment (2026-07-20)

• The work appeared as “Extraction of Weak Surface Diaphragmatic Electromyogram Using Modified Progressive FastICA Peel-Off” in IEEE Transactions on Biomedical Engineering, 73(3):1191–1201, March 2026 (authors Yao Li, Dongsheng Zhao, Haowen Zhao, Min Shao, Xu Zhang; DOI 10.1109/TBME.2025.3599637).^{2 3}
• The core problem is that surface diaphragmatic EMG (sEMGdi) recorded from chest electrodes is very weak and buried under strong, repetitive electrocardiogram (ECG) interference whose spectrum overlaps the EMGdi band, historically limiting sEMGdi versus invasive esophageal EMGdi.^{2 3}
• The proposed pipeline is a two-step blind source separation—an initial FastICA followed by a constrained FastICA that extracts and refines the strong ECG and then the respiration-related sEMGdi source—combined with a peel-off strategy that iteratively removes dominant components to progressively recover weaker sEMGdi.^{2 3}
• The constrained FastICA used a delay factor K of 15, which drives convergence toward a component closely matching the target source signal.³
• On synthetic data the method outperformed state-of-the-art comparison methods in signal-to-interference ratio (SIR) and correlation coefficient (CORR) across all tested noise levels.^{2 3}
• On clinical recordings it achieved 95.06% accuracy and a 96.73% F2-score for breath identification, indicating reliable extraction with minimal distortion.^{2 3}
• The intended application is noninvasive respiratory monitoring and ventilator synchrony—for example triggering neurally adjusted ventilatory assist (NAVA)—with relevance to respiratory rehabilitation and human–machine synchronization.³
• The approach fits a broader line of respiratory sEMG signal processing that uses blind source separation (e.g., wavelet-domain cardiac removal plus nonnegative matrix factorization to separate inspiratory from expiratory activity, or ICA combined with wavelet thresholding to strip ECG from EMGdi).^{4 5}

Footnotes

1. http://ieeexplore.ieee.org/document/11134129 ↩ ↩² ↩³ ↩⁴ ↩⁵

2. https://doi.org/10.1109/tbme.2025.3599637 ↩ ↩² ↩³ ↩⁴ ↩⁵

3. https://arxiv.org/abs/2406.01284 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

4. https://doi.org/10.1088/1361-6579/ac799c ↩

5. https://www.sciencedirect.com/science/article/abs/pii/S156849461630103X ↩