Closed-loop adaptive filtering for supressing chest compression oscillations in the capnogram during cardiopulmonary resuscitation

摘要

Capnography is widely used by the advanced-life-support during cardiopulmonary resuscitation (CPR). Continuous analysis of the capnogram allows guidance of adequate ventilation rate, currently 10 breaths/min for intubated patients. We used 60 out-of-hospital cardiac arrest episodes containing both clean and CC corrupted capnograms. Chest compressions (CC) induce high-frequency oscillations in the capnography waveform impeding reliable detection of ventilations. Thus, a clean capnogram is essential for a better ventilation detection performance. To clean the capnogram, an adaptive noise cancellation notch filter was designed using a Least Mean Square algorithm to minimize filtering error. A fixed-coefficient low-pass filter was optimized for comparison. For the whole test set, global Se/PPV improved from 93.0/92.2% to 97.6/96.2% after adaptive filtering and to 97.7/94.8% after fixed-coefficient filtering. For the clean subset, Se/PPV maintained stable and for the corrupted subset, Se/PPV improved from 84.8/84.0% to 95.2/92.7% and 95.4/90.3%, respectively. Filtering allowed reliable automated detection of ventilations in the capnogram even in the presence of CC oscillations during CPR. Nevertheless, further evaluation of these techniques in large datasets is warranted.