TY - JOUR
T1 - Electrocardiogram Monitoring via Fiber Bragg Grating-Based Laryngeal Sensing and the Laryngeal-Vibration-to-ECG Network
AU - Tang, Jiaxuan
AU - Gao, Lei
AU - Yang, Hang
AU - Chen, Yang
N1 - Publisher Copyright:
© 2026 IEEE. All rights reserved.
PY - 2026/9/1
Y1 - 2026/9/1
N2 - Continuous electrocardiogram (ECG) monitoring is important for cardiovascular management, but conventional electrode-based systems often cause skin irritation and discomfort during long-term use. Since cardiac electrical activity is coupled with body surface mechanical vibrations, this study explores noninvasive ECG reconstruction from laryngeal vibration signals. Specifically, a fiber Bragg grating (FBG) sensing system was developed to capture subtle laryngeal vibrations, and a deep neural network, termed the laryngeal-vibration-to-ECG network (LV2E-Net), was proposed to reconstruct standard ECG waveforms in an end-to-end manner. Both subject-dependent and leave-one-subject-out (LOSO) evaluations were conducted. The reconstructed ECG waveforms preserved the major morphological structures and rhythm information of the reference ECG. Under the subject-dependent setting, the Pearson correlation coefficient (PCC) ranges from 0.967 to 0.994, with the median absolute R-peak time errors below 2 ms and mean absolute heart rate errors within 0.1-0.4 beat per minute (bpm) across all eight subjects. Under the LOSO setting, the PCC ranges from 0.721 to 0.907, while the median absolute R-peak time errors stay below 13 ms, and the mean absolute heart rate error is controlled within 0.15-1.20 bpm across subjects. These results demonstrate the feasibility of noninvasive ECG reconstruction from laryngeal FBG vibration signals for long-term cardiovascular monitoring.
AB - Continuous electrocardiogram (ECG) monitoring is important for cardiovascular management, but conventional electrode-based systems often cause skin irritation and discomfort during long-term use. Since cardiac electrical activity is coupled with body surface mechanical vibrations, this study explores noninvasive ECG reconstruction from laryngeal vibration signals. Specifically, a fiber Bragg grating (FBG) sensing system was developed to capture subtle laryngeal vibrations, and a deep neural network, termed the laryngeal-vibration-to-ECG network (LV2E-Net), was proposed to reconstruct standard ECG waveforms in an end-to-end manner. Both subject-dependent and leave-one-subject-out (LOSO) evaluations were conducted. The reconstructed ECG waveforms preserved the major morphological structures and rhythm information of the reference ECG. Under the subject-dependent setting, the Pearson correlation coefficient (PCC) ranges from 0.967 to 0.994, with the median absolute R-peak time errors below 2 ms and mean absolute heart rate errors within 0.1-0.4 beat per minute (bpm) across all eight subjects. Under the LOSO setting, the PCC ranges from 0.721 to 0.907, while the median absolute R-peak time errors stay below 13 ms, and the mean absolute heart rate error is controlled within 0.15-1.20 bpm across subjects. These results demonstrate the feasibility of noninvasive ECG reconstruction from laryngeal FBG vibration signals for long-term cardiovascular monitoring.
KW - Cardiac motions
KW - cross-modal mapping
KW - deep neural network
KW - electrocardiogram (ECG)
KW - fiber Bragg grating (FBG)
UR - https://www.scopus.com/pages/publications/105046114708
U2 - 10.1109/JSEN.2026.3715750
DO - 10.1109/JSEN.2026.3715750
M3 - 文章
AN - SCOPUS:105046114708
SN - 1530-437X
VL - 26
SP - 25812
EP - 25822
JO - IEEE Sensors Journal
JF - IEEE Sensors Journal
IS - 17
ER -