[home] [Personal Program] [Help]

---

Session: Poster session 2 (Odd numbers)
Session starts: Friday 27 January, 10:00
Presentation starts: 10:00

---

Gabriela Grońska (Eindhoven University of Technology)
Elisabetta Peri (Eindhoven University of Technology)
Xi Long (Eindhoven University of Technology)
Hans van Dijk (Center for Sleep Medicine, Kempenhaeghe)
Massimo Mischi (Eindhoven University of Technology)

Abstract:
Introduction: Diaphragmatic electromyography (dEMG) is a non-invasive method to monitor respiratory activity. However, the quality of dEMG is often compromised by the presence of multiple disturbances, such as cardiac electrical activity, noise, and motion artifacts. This impacts the clinical uptake of dEMG to monitor respiration. A recent paper from our group proposed an algorithm based on singular value decomposition (SVD) to remove cardiac interference, showing its superiority with respect to other strategies proposed in the literature [1]. Nevertheless, the method is not satisfactory in presence of motion artifacts. In this work, we aim at improving the performance of the SVD-based algorithm to remove both cardiac interference and motion artifacts. Methods: The original algorithm used signal-to-noise ratio as a criterium to select singular values corresponding with ECG contamination. In the current work, we investigate a new index to reject singular values which map components corresponding not only with ECG contamination but also with motion artifacts. The index was created based on the frequency domain range of ECG and motion artifacts (mainly 5-30Hz). Power spectral density (PSD) was obtained for each singular value, and the ratio between PSD5-30Hz and PSD30-250Hz was calculated as an index. Components with the highest indexes (outside 95% of the distribution) were rejected. A synthetic dataset was used to compare the performance of the algorithm with the new index to the original algorithm. Muscular activity of the biceps from nine healthy volunteers was combined with ECG only (dataset A) and with ECG mixed with motion artifacts (dataset B). The error in the EMG reconstruction was assessed as the root mean squared error (RMSE) in the time domain and mean-frequency difference (MFD) in the frequency domain, and it is reported as median (interquartile values). Results and Discussion: The proposed method achieved an RMSE of 9.9(1.8)%, and an MFD of 0.31(0.02)Hz for dataset A. The performance on dataset B showed an RMSE of 28.2(2.3)% and an MDF was 0.59(0.56)Hz. Results are significantly better than what was achieved with the original algorithm in time and frequency domains for both datasets (p-value<0.03). In future work, the proposed algorithm will be tested on real data. References [1] Peri E., et al. Sensors (2021); 21, 573