An integrated framework for evaluation on typical ECG-derived respiration waveform extraction and respiration

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• ECG-derived respiration (EDR) methods aim to extract respiratory information from electrocardiogram (ECG) signals.
• Standardized performance evaluation of EDR techniques is crucial for reliable clinical application.

Purpose of the Study:

• To comprehensively evaluate the performance of 10 feature-based EDR methods.
• To assess EDR method performance under varying sampling rates, noise levels, and window lengths.
• To provide references for selecting appropriate EDR algorithms based on specific requirements.

Main Methods:

• Evaluation of 10 feature-based EDR methods using the Fantasia database, which includes simultaneous ECG and respiration recordings.
• Performance quantification based on waveform correlation and breathing rate (BR) errors.
• Analysis of the impact of sampling rate (≥150 Hz vs. <150 Hz), noise, and window length (32s, 16s, 8s) on EDR performance.

Main Results:

• Amplitude-based EDR methods (AMarea, AMQR, AMR) showed BR errors below 2 bpm at sampling rates >150 Hz, with a sharp ~60% decrease below this rate.
• Frequency-modulated respiration (FM_RR) demonstrated stable performance with errors <2 bpm across different sampling rates.
• Noise significantly impacted amplitude-based methods, reducing waveform correlation by up to 40%.
• Frequency-based EDR methods exhibited increased BR errors with shorter window lengths (16s vs. 32s).
• Time-based EDR techniques struggled to obtain BR errors within an 8s window for 30%-40% of cases.

Conclusions:

• The study provides an integrated evaluation of EDR waveform extraction and respiration rate calculation.
• Findings offer valuable references for algorithm selection in EDR applications, considering factors like sampling rate, noise tolerance, and window length constraints.
• Optimizing EDR method performance requires careful consideration of signal acquisition parameters and algorithm choice.