Fig 1.
The waveform represents the varying CO2 levels during the respiratory cycle. Typical segments and phases are named according to [18].
Fig 2.
The three different types of observed artifact.
(A) Type I, located in the plateau of the capnogram; type II, in the baseline, and type III, spanning from the plateau to the baseline. Each capnogram is depicted with the corresponding CD signal. (B) Power spectral density (PSD) of each capnogram (in solid blue line) and CD signal (in dotted red line). Capnograms present a significant peak at the fundamental frequency of the artifact, fcc, with highest amplitudes in type III samples.
Fig 3.
Annotation of ventilations and chest compressions.
Ventilations were annotated using the low frequency component of the TI signal (upper panel, in blue), obtained by low-pass filtering the raw TI signal (in gray). Each ventilation was annotated at the rise of a TI fluctuation (red vertical lines). In the capnogram (middle panel), these annotations corresponded to CO2 concentration’s rapid decay to zero. Chest compression instances were annotated in the CD signal (lower panel), and are depicted with red dots corresponding to the instants where the maximum compression depth was achieved.
Fig 4.
Scheme of the ventilation detector.
The algorithm locates upstrokes (tup) and downstrokes (tdw) in the capnogram signal (right) applying a fixed amplitude threshold Thamp. It extracts the duration of the intervals Dex and Din. Finally, fixed duration thresholds Thex and Thin are used to discriminate true ventilation from the potential candidates. Detected ventilations are depicted with vertical red dotted lines in the bottom panel.
Fig 5.
Examples of filtering performance.
Original capnogram with clean and distorted respiration cycles (top panel). Detected ventilations are depicted with vertical lines. Distorted ventilations could not be detected by the algorithm. Lower panels show the filtered capnogram (in blue) superimposed to the original capnogram (in gray), for the three filtering alternatives. Detected ventilations are depicted with vertical red dashed lines. In this example, all ventilations were correctly detected after filtering.
Table 1.
Characteristics of the episodes included in the study.
Values are expressed as mean (±standard deviation).
Table 2.
Performance of the ventilation detection algorithm before and after filtering for each type of artifact.
Fig 6.
(A) Distributions of Se/PPV values per episode in each artifact category, before and after filtering. (B) Distribution of the unsigned error in percentage in the estimation of ventilation rate. Results are provided for all categories: C: clean. D: distorted. I: type I artifact. II: type II; III: type III.
Table 3.
Detection of over-ventilation (ventilation rate >10 min-1).