Non-intrusive real-time breathing pattern detection and classification for automatic abdominal functional electrical stimulation.

Abdominal Functional Electrical Stimulation (AFES) has been shown to improve the respiratory function of people with tetraplegia. The effectiveness of AFES can be enhanced by using different stimulation parameters for quiet breathing and coughing. The signal from a spirometer, coupled with a facemask, has previously been used to differentiate between these breath types. In this study, the suitability of less intrusive sensors was investigated with able-bodied volunteers. Signals from two respiratory effort belts, positioned around the chest and the abdomen, were used with a Support Vector Machine (SVM) algorithm, trained on a participant by participant basis, to classify, in real-time, respiratory activity as either quiet breathing or coughing. This was compared with the classification accuracy achieved using a spirometer signal and an SVM. The signal from the belt positioned around the chest provided an acceptable classification performance compared to the signal from a spirometer (mean cough (c) and quiet breath (q) sensitivity (Se) of Se(c)=92.9% and Se(q)=96.1% vs. Se(c)=90.7% and Se(q)=98.9%). The abdominal belt and a combination of both belt signals resulted in lower classification accuracy. We suggest that this novel SVM classification algorithm, combined with a respiratory effort belt, could be incorporated into an automatic AFES device, designed to improve the respiratory function of the tetraplegic population