Automatic segmentation and extraction of features from human respired carbon dioxide waveform

Automated selection and segmentation of human respired carbon dioxide (CO2) waveform is highly in need as measurable indices from the CO2 waveform, could allow an indirect assessment of asthma. Previous studies employed manual and time-setting as criteria for the selection and partitioned of the CO2 waveform, which may be a source of bias. Thus, despite many studies, which show a good correlation between CO2 signal’s indices and a spirometer parameter, monitoring of asthma has not yet become part of clinical practice. Herein, we propose an algorithm for automated selection and segmentation of the CO2 waveform. CO2 waveforms were recorded from 30 asthma and 20 non-asthma. We computationally extracted four physiologically based CO2 signal indices from each segmented phase. Further, the usefulness of indices and analysis of segmented phases of the CO2 signal was assessed by measuring the area (Az) under receiver operating characteristics (ROC) curve. Here, each breath cycle was considered valid based on power spectral density, frequency resolution, and end-tidal CO2, which was estimated by the max-min algorithm. In addition, we found that features extracted from all the segmented part were statistically significant except the combination of upper expiratory and alveolar. However, the strongest were found with the part of the upward expiratory phase (11-15mmHg) for the discrimination of asthma and non-asthma with an Az, ranges from 0.96 (95% CI: 0.92–1) to 0.97 (95% CI: 0.92-1). Thus, the presented algorithm has the potential to implement in real time for the automatic differentiation of non-asthma and asthma.