Estimation of Biomarkers for Respiratory Assessment: A Signal Processing Framework Using Capnostream Respiratory Monitors

The interdisciplinary field of Respiratory Signal Analysis examines various respiratory signals to extract valuable health-related information. The signals, generated during breathing are captured as electrical or mechanical data. The Capnostream™35 Portable Respiratory Monitor, which measures end-tidal carbon dioxide (etCO ₂ ), respiratory rate (RR), oxygen saturation (SpO ₂ ), and pulse rate (PR), is a widely used device in hospitals for continuous respiratory monitoring. However, it provides limited biomarker data, such as respiratory rate, neglecting crucial details, such as inhalation/exhalation time, tidal volume, and minute ventilation. This study presents a signal-processing framework using an open-source platform that can be integrated into a capnostream monitor to estimate respiratory biomarkers. The study utilized a sample of n = 1250, comprising 1250h recordings (1h per subject). To ensure signal stationarity, a one-minute segmentation window was employed for biomarker extraction. A Savitzky–Golay filter was applied to clean the raw signals for the analysis. The results indicated that the open-source signal-processing framework offers essential information for clinical respiratory diagnosis. The effectiveness of the framework was validated through statistical methods, including the Pearson correlation coefficient, bar chart, box plot, and Bland-Altman plot, which demonstrated strong agreement between the algorithm and Capnostream device outputs. Pearson-correlation analysis revealed a significant medium positive relationship (p < .001) between the algorithm outcomes and Capnostream output. Bland-Altman analysis showed mean differences of -0.88 and − 0.43 (bpm) between algorithm results (with and without segmentation) and Capnostream display results, with lower limits of agreement from − 5.33 to -4.8 and upper limits from 3.58 to 3.53, indicating excellent agreement for breathing rate. This was further supported by the bar graph and boxplot analyses, which showed similar mean and median values. The Bland–Altman analysis also demonstrated acceptable agreement for tidal volume and minute ventilation calculations using the algorithm compared to formula-based methods.