A simple algorithm for ventilation detection in the capnography signal during cardiopulmonary resuscitation

During cardiopulmonary resuscitation, excessive ventilation rates reduce the chance of survival. We have developed a simple method to automatically detect ventilations based on the analysis of the capnography signal recorded with monitor-defibrillators. We used 60 out-of-hospital cardiac arrest episodes that contained both clean and chest compressions (CC) corrupted capnograms. The detection algorithm first identified ventilation candidates in the capnography signal. Then, it characterized every candidate by features related to inspiration and expiration durations, and finally a decision system based on static thresholds was applied in order to determine whether each candidate corresponded to a true ventilation. Sensitivity (Se) and positive predictive value (PPV) for the clean set (3905 ventilations) were 99.8% and 99.1%, respectively. With the corrupted set (6778 ventilations) Se and PPV decreased to 85.3% and 85.6%, respectively. For the whole test set (10683 ventilations) Se and PPV were 90.6% and 90.6%, respectively. Detector's performance clearly degraded when applied to corrupted episodes, this demonstrates the need for techniques to suppress CC artefact to improve ventilation detection.