Improved respiration rate estimation using a Kalman filter and wavelet cross-coherence

Introduction: Respiration rate is a common measurement in the intensive care unit (ICU) which is well correlated with patient severity. However, automated estimation of the respiration rate, especially when the patient is not intubated, is prone to large errors. Here we present a method of merging respiration estimates from the electrocardiogram (ECG) merged based on a novel signal quality index. Methods: Four lead electrocardiograms (ECGs) and capnograms were recorded for 133 patients admitted to a mixed ICU during a spontaneous breathing trial. An average of 2.93 ± 0.53 hours of data was recorded for each patient. Respiration was derived using four methods based upon respiratory sinus arrythmia and ECG amplitude modulation by respiration. A novel signal quality index (SQI), based upon the Wavelet Transform coherence (WTC) between two respiration waveforms, was used to reflect the quality of each signal. This SQI was used with a Kalman filter to provide a single robust respiration estimate for each ECG lead. These respiration estimates were compared with the reference extracted from the capnogram. Results: The root mean square error of the new approach ranged between 5.4-6. J breaths per minute across ECG leads. These errors were statistically significantly better than all component respiration estimates. Conclusions: Respiration rate can be robustly estimated from ECG leads during a spontaneous breathing trial. The use of a novel SQI within a Kalman filter allows for proper assessment of the accuracy of each component respiration estimate, even though the estimates are derived from the same underlying ECG lead.