Automated Classification of Whole Body Plethysmography Waveforms to Quantify Breathing Patterns

Figures

• 

  FIGURE 1. Method for breath detection and respiratory waveform clustering. (A) Example plethysmography flow trace. (B) Inspiratory efforts were detected with a threshold of 0.1 ml/s to avoid accidental noise detections, the starting point of each breath was then set to the previous zero-crossing and ended at the start of the next breath. (C) Breath detection windows were shifted back in time 20% of the breath length (max of 100 ms) to capture the entire inspiratory effort. Two windows are shown at their original position (dashed box) and final position (solid line). Positive values indicate inspiratory activity. Breaths were separated into four time domains to allow for cluster sorting; these time domains correspond to distinct respiratory behaviors. (D) Example waveforms from several animals for each of the four time domains (sniffing, short tidal, long tidal, and very long breaths). (E) Cluster methodology. Long duration (1.2–4 s) breaths are used as an example to illustrate clustering method. The relationship between cluster number and variance is plotted to show how we determined the number of clusters within each time domain. A 30 cluster hierarchical linkage tree with the six cluster…

• 

  FIGURE 2. Experimental protocol and example plethysmography traces during baseline and fentanyl periods. (A) Experimental timeline, rats were lightly anesthetized with isoflurane to place an i.v. port in the tail vein, then rats were placed in whole body flow through plethysmography chamber (21% O2 balance N2), after a 32 ± 7 min (mean ± SD) baseline period fentanyl was infused until minute ventilation dropped to 50% of baseline levels. A rescue dose (3, 10, or 30 mg/kg) of ampakine CX1942 or control (saline) was then injected. (B) Example flow traces showing the baseline and fentanyl period from three rats. Under baseline conditions, the varied nature of rat respiration produced waveforms indicative of switching between multiple behaviors (e.g., quiet breathing and sniffing). However, injection of fentanyl caused respiratory waveforms to exhibit less variation and decrease in amplitude. (C) Expanded traces from panel (B).

• 

  FIGURE 3. Clustering allows for behavioral annotation and quantification. (A) Example plethysmography traces from the baseline and fentanyl periods of the same animal. 20-breath bins are annotated to indicate sniffing (blue), common tidal breathing (green), uncommon tidal breathing (red), and augmented breaths (*cluster 19). Epochs were only marked if the majority of breaths met the inclusion criteria (the majority of waveforms belonging to a specific set of clusters, see text). (B,C) Percentage of breaths belonging to clusters 5–15 that represent breaths that have lengths associated with tidal breathing. (B) Breaths that had waveforms common at BL (clusters 5, 6, 11, 12, and 14) decreased following fentanyl infusion and (C) breaths with waveforms uncommon at baseline increased during fentanyl infusion. (D) Percentage of breaths belonging to clusters 1–3 (sniffing behavior) during the baseline and fentanyl periods. (E) Pile plot of all breaths classified as belonging to cluster 19 during baseline and fentanyl infusion. (F) Percentage of breaths during baseline (BL) and fentanyl infusion (F) which were classified as cluster 19. Cluster 19 breaths were completely absent during the last 3rd…

• 

  FIGURE 4. Traditional ventilation metrics. (A) Respiratory rate (RR), (B) Tidal volume (VT), and (C) Minute ventilation (VE) expressed as a percentage of baseline. Two-way ANOVA. Post hoc comparisons (Tukey–Kramer) are shown (*p < 0.05) if the interaction p < 0.05; further only the comparisons within a time point across groups, or within a group but across time points are shown. n = 8 animals/treatment group; 32 animals stratified by treatment condition.

• 

  FIGURE 5. Cluster prevalence throughout experimental protocol stratified by rescue condition. (A) Peri-event raster plot, aligned to the rescue injection. Each line in the raster plot represents a single animal in each experimental condition. Rasters mark the beginning of inspiration for each breath; raster colors match cluster colors from Figure 1. The beginning of the fentanyl infusion is marked with an asterisk (*). Two animals excluded from plot due to prolonged fentanyl infusion period. The proportion of each cluster during the (B) baseline (n = 32, all animals in study pooled together), (C) fentanyl (n = 32, all animals in study pooled together), (D) rescue periods (n = 8 animals/treatment group). The average waveform and cluster number is plotted next to the pie segment if the cluster prevalence was greater than 5%.

• 

  FIGURE 6. Prevalence of respiratory waveform clusters that were strongly affected by ampakine CX1942 injection. Comparison of the prevalence of clusters that were either (A) reduced (n = 8/treatment group) or (B) increased (n = 8/treatment group) by ampakine CX1942; between the control (saline) and high dose (30 mg/kg) ampakine groups. These six clusters were identified within the high dose (30 mg/kg) ampakine group as being affected by ampakine (see text). All comparisons were 2-way ANOVAs with Tukey–Kramer post hoc tests to control for multiple comparisons (*p < 0.05) were run if the interaction p < 0.05; if there was a significant (p < 0.05) main effect of time (BL, Fent, Rescue), a ‡ or a # indicates a significant (p < 0.05) pairwise comparison between baseline (BL) or fentanyl (Fent), respectively.