Identification of eupneic breathing using unsupervised machine learning.

ABSTRACT

The diaphragm muscle (DIAm) is unique to mammals and the primary muscle involved in breathing. In awake animals, considerable heterogeneity in DIAm electromyographic (EMG) activity reflects varied ventilatory and non-ventilatory behaviors. Experiments in awake animals are an essential component to understanding the neuromotor control of breathing; thus, it is paramount to unambiguously identify DIAm EMG activity that in fact reflects breathing. Current strategies for doing so in a reproducible, reliable, and efficient fashion are lacking. The present study used machine learning to evaluate DIAm EMG from awake rats using hierarchical clustering across four-dimensional feature space to classify eupneic breathing. Our model, which can be implemented with automated threshold of the clustering dendrogram, successfully identified eupneic breathing with high F1 score (0.92), specificity (0.70), and accuracy (0.88), indicating that it is a robust and reliable tool for investigating the neural control of breathing.