Abstract
Introduction
Impedance Cardiography (ICG) is a non-invasive technique allowing for the continuous monitoring of patients haemodynamic parameters, such as Cardiac Output and Stroke Volume, which can be analysed in a clinical setting. Traditionally, ICG’s are recorded on the thoracic region. However, long-term ambulatory ICG from the brachial artery along the left upper-arm offers highly-attractive clinical applications with user-compliant arm-ICG wearable device, providing reliable trend-based indicators of cardiac contractility in novel methods for cardiovascular disease diagnosis.
Methods
Seven healthy cases were considered in this study, with arm-ICG and conventional thorax-ICG data recorded simultaneously with ECG Lead-I. ICG real-time denoising performance for an optimised band-pass (0.5-8Hz) Butterworth IIR-filter, for a proposed Savitzky-Golay FIR-filter design and for the combined processing of both filters was assessed. For each subject-case and mode, a 700ms “noiseless” ICG-vector was extracted by signal-averaged of 600 ICG beats, against which Pearson correlation (p), residual-noise level RMS and S/N-ratio (SNR; signal being the noiseless-ICG-vector), were computed as denoising metrics on a beat-by-beat basis. Further denoising performance assessment was by their impact (error-rate%) on the accuracy of ICG waveform parameters: ICG peak-amplitude B (Ω/s) and Ventricular-Ejection-Time VET (ms), for estimating Stroke-Volume.
Results
Mean denoising performance metric p for Butter-IIR-filtering, SG-FIR-filter and combined (Butter-IIR)+(SG-FIR), on arm-ICG, were: p=0.87, p=0.86 and p=0.88 respectively, and on thorax-ICG: p=0.97, p=0.97 and p=0.98. Mean residual-noise RMS metric on arm-ICG, were: 0.087mV, 0.087mV and 0.077mV respectively, and mean SNR: 9.75, 7.85 and 12.46. Average (N=7) error-rate% for B: 21.7±10.6(SD), and VET: 13.88±4.59(SD).
Conclusions
Comparative assessment results by the three denoising performance metrics (p, RMS and SNR) consistently indicated highest performance for the combined (Butter-IIR)+(SG-FIR) denoising filter process, which also enabled the highest accuracy performance on arm-ICG waveform Stroke-Volume related parameters determination (B and VET), with error-rate statistics about 20% for B and 10% for VET, on average.
Impedance Cardiography (ICG) is a non-invasive technique allowing for the continuous monitoring of patients haemodynamic parameters, such as Cardiac Output and Stroke Volume, which can be analysed in a clinical setting. Traditionally, ICG’s are recorded on the thoracic region. However, long-term ambulatory ICG from the brachial artery along the left upper-arm offers highly-attractive clinical applications with user-compliant arm-ICG wearable device, providing reliable trend-based indicators of cardiac contractility in novel methods for cardiovascular disease diagnosis.
Methods
Seven healthy cases were considered in this study, with arm-ICG and conventional thorax-ICG data recorded simultaneously with ECG Lead-I. ICG real-time denoising performance for an optimised band-pass (0.5-8Hz) Butterworth IIR-filter, for a proposed Savitzky-Golay FIR-filter design and for the combined processing of both filters was assessed. For each subject-case and mode, a 700ms “noiseless” ICG-vector was extracted by signal-averaged of 600 ICG beats, against which Pearson correlation (p), residual-noise level RMS and S/N-ratio (SNR; signal being the noiseless-ICG-vector), were computed as denoising metrics on a beat-by-beat basis. Further denoising performance assessment was by their impact (error-rate%) on the accuracy of ICG waveform parameters: ICG peak-amplitude B (Ω/s) and Ventricular-Ejection-Time VET (ms), for estimating Stroke-Volume.
Results
Mean denoising performance metric p for Butter-IIR-filtering, SG-FIR-filter and combined (Butter-IIR)+(SG-FIR), on arm-ICG, were: p=0.87, p=0.86 and p=0.88 respectively, and on thorax-ICG: p=0.97, p=0.97 and p=0.98. Mean residual-noise RMS metric on arm-ICG, were: 0.087mV, 0.087mV and 0.077mV respectively, and mean SNR: 9.75, 7.85 and 12.46. Average (N=7) error-rate% for B: 21.7±10.6(SD), and VET: 13.88±4.59(SD).
Conclusions
Comparative assessment results by the three denoising performance metrics (p, RMS and SNR) consistently indicated highest performance for the combined (Butter-IIR)+(SG-FIR) denoising filter process, which also enabled the highest accuracy performance on arm-ICG waveform Stroke-Volume related parameters determination (B and VET), with error-rate statistics about 20% for B and 10% for VET, on average.
| Original language | English |
|---|---|
| Publication status | Accepted/In press - 25 Jun 2023 |
| Event | Computing in Cardiology 2023, Atlanta, USA - Atlanta, United States Duration: 1 Oct 2023 → 4 Oct 2023 |
Conference
| Conference | Computing in Cardiology 2023, Atlanta, USA |
|---|---|
| Abbreviated title | CinC 2023 |
| Country/Territory | United States |
| City | Atlanta |
| Period | 1/10/23 → 4/10/23 |
