Automated evaluation of respiratory signals to provide insight into respiratory drive

Abstract

The diaphragm muscle (DIAm) is the primary inspiratory muscle in mammals and is highly active throughout life displaying rhythmic activity. The repetitive activation of the DIAm (and of other muscles driven by central pattern generator activity) presents an opportunity to analyze these physiological data on a per-event basis rather than pooled on a per-subject basis. The present study highlights the development and implementation of a graphical user interface-based algorithm using an analysis of critical points to detect the onsets and offsets of individual respiratory events across a range of motor behaviors, thus facilitating analyses of within-subject variability. The algorithm is designed to be robust regardless of the signal type (e.g., EMG or transdiaphragmatic pressure). Our findings suggest that this approach may be particularly beneficial in reducing animal numbers in certain types of studies, for assessments of perturbation studies where the effects are relatively small but potentially physiologically meaningful, and for analyses of respiratory variability.

Keywords: Diaphragm muscle; Motor control; Motor unit; Respiratory; Statistics; Variability.

Figure 1.

Set-up for collection of EMG (blue) and P_di (orange) from anesthetized rats. Solid-state pressure transducers were inserted through the mouth into the esophagus and stomach. The pressure differential is the P_di. The abdomen was bound to stabilize the gastric pressure and signals were digitized using PowerLab 8/35.

Figure 2.

An example EMG recording of a single breath from a rat during eupnea. The orange trace represents the RMS EMG (50 ms moving window); the thick black line overlayed shows the effects of the Savitzky-Golay (3^rd order polynomial, 50.5 ms frame length) filter. Note that the Savitzky-Golay filter retains important features of the signal and does not attenuate its amplitude or appreciably change the onset or offset compared to the RMS EMG trace, but it does reduce the millisecond-scale noise.

Figure 3.

Overview of AERS algorithm. Panel I shows critical points detected by the algorithm marked with black, red, and yellow dots. The different colors represent the different amplitude clusters that each set of critical points belongs to (i.e., baseline, ECG peaks, and EMG peaks). The y-axis represents the amplitude of the signal divided by its range (i.e., the difference between the largest and smallest value in the signal). Panel II shows detection and removal of ECG-related onsets and offsets. Note that in both Panel I and Panel II, scroll bars are included to control the thresholds allow user inputs if the empirically determined thresholds are insufficient for specific signals. In Panel III, the detection of the onset (blue circles) and offset (orange circles) of individual respiratory events are shown for a segment of a Savitzky-Golay filtered RMS EMG signal during eupnea. Note that detection of ECG activity in the EMG signal can be omitted from analysis. Additionally, both the initiation and termination points can be manually adjusted (including addition or subtraction of them) to ensure that no spurious detections are included in the final analyses as shown in the inset in Panel III. Panel IV shows examples of different calculations that can be performed from the data acquired using this approach.

Figure 4.

Representative tracings of simultaneously recorded left side DIAm EMG and P_di signals across motor behaviors. Differences in the normalized amplitudes were detectable across behaviors and signal types.

Figure 5.

Ensemble average response synchronized to the start of each respiratory event (normalized to the amplitude during airway occlusion). The ensemble average response shows the overall trajectory of the respiratory signals. Note the general similarity between hypoxia-hypercapnia and eupnea. Despite a small increase in amplitude, the primary difference between hypoxia-hypercapnia and eupnea is in the respiratory rate. Note that synchronizing activity to the start of each respiratory event inherently underestimates the peak amplitude. The dashed vertical lines represent the location of the peak of the RMS EMG signal for each behavior.

Figure 6.

The relationship between P_di and RMS EMG peak amplitude (A) and 75 ms amplitude (B) for each automatically detected respiratory event. Colors represent different subjects. Across subjects, the RMS EMG generally tracks the changes in P_di peak amplitude, although there is considerable variability in the relationship. Overall, these data suggest that as a percentage of the peak amplitude during airway occlusion, RMS EMG tends to underestimate the peak amplitude compared to the P_di (note the slope in A). The difference in the relationship between the P_di amplitude at 75 ms after onset and the RMS EMG amplitude 75 ms in B after onset is even starker, likely due to the slower rate of rise in the P_di signal (evident in the ensemble average response in Figure 4 and the representative tracings in Figure 3).

Figure 7.

The relationship between respiratory rate (A), activity duration (B), and duty cycle (C) calculated from the P_di and RMS EMG signals for each automatically detected respiratory event. Colors represent different subjects. The values across ventilatory parameters tend to be similar regardless of signal type, however activity duration tends to be lower when measured from the P_di signal, likely due to the smoother transition points in the signal, which make detection of an onset and offset more difficult computationally.