We analyse the respiration trace from a respiratory belt in a study on emotions. We want to calculate the average respiratory rate across a given block (1 minute). An automated analysis with LabChart detects most respiratory cycles correctly. However, some peaks in the respiration trace seem to be abnormal / artifacts (e.g., very short or very shallow breaths). Do you know any guidlines that inform how to include only the meaningful data into the analysis and rule out breathing artifacts?
I used a respiratory belt for a project a few years ago. When we saw suspicious respiration trace values we looked at other physiological signals. If the participant fidgets or moves in such a way that the respiration signal is damaged, odds are the artifacts will appear in other signals as well. In our case, we had ECG leads mounted on the chest above the belt, so respiration and ECG artifacts tended to coincide.
Hi Lukasz,
I'm studying about breathing and pain and in my study I record lots of respiratory and other physiological data. I don't know about such guideline but following is my comment based on my little experience.
First you need to distinguish an artifact which is usually due to movements not associated with respiration from a respiratory cycle which is real but largely different from other cycles ("abnormal" as you mentioned). Unfortunately most of the respiratory belts are very prone to movement artifacts. You can manage such artifacts by:
- Filtering of the belt signal with low pass of 0.5 to 1 Hz give you more smooth signal. If the rate is the main outcome this is helpful. If the amplitude is also important make sure that filtering doesn't change the amplitude a lot. You need to play with the filter.
- Looking at EtCO2, I guess you don't have it.
- Looking at other signals like ECG and EMG may help to find movements, if they are not already filtered for movement-related artifacts.
- If you align the belt signal and RR-interval (after interpolation) in parallel and then look at the respiratory sinus arrhythmia that can help since you expect variation in RR-interval associated with respiration.
- The amplitude of the R-wave in the ECG also changes with respiration. If the high pass cut off is set to 0.05-0.1 Hz and not 0.5-1 Hz, then respiration-related changes in R-wave amplitude and also in the baseline in the ECG signal are more visible.
- The peak of the HF band of the HRV is usually corresponding to respiration frequency. You can test its accuracy in a sample of your subjects which you are sure you don't have artifacts. The Kubios software provides ECG-derived respiration data as separate output in addition to HF peak.
But, even with measuring air flow at mouth level (kind of gold standard for respiration) you sometimes see a suspicious respiratory cycle very different from other cycles but not artifact. These cycles either have very low amplitude (tidal volume) or are very short. I don't call them "abnormal". How to manage them depends on the context you are studying. For example, there are usually a period of apnea/hypopnea after a sigh. So, in case that you have sighs you may have few very shallow inspirations right after that. In case you do not expect large variability in respiratory parameters (time, tidal volume) screening for outliers (e.g. based on mean +- 2/3 SD) can help.
At the end, for accurate measure, you need to visually check the signal and correct the data. There is no optimal automatic method to do that (especially with the belt), to my knowledge. Any criteria you chose for excluding "abnormal" cycles make sure that you don't get biased toward your study hypothesis since the probability of those "abnormal" cycles may be affected by the study intervention.
I'm following this question to see if someone can introduce a guideline for this issue.
Good luck with your research