Respiratory rate estimation during walking/running activities using principal components estimated from signals recorded by a smart garment embedding piezoresistive sensors

The aim of this work was to investigate the use Principal Component Analysis (PCA) applied to signals recorded by 4 textile piezoresistive sensors embedded within a smart garment (SG) to estimate respiratory rate (RR) during walking/running tasks. To this aim, we enrolled 6 subjects who were asked to perform 7 trials on a treadmill at different speeds: one static condition (quiet breathing at rest), three walking trials at 1.6 km/h, 3.0 km/h and 5.0 km/h, and three running trials at 8.0 km/h, 10.0 km/h and 12.0 km/h. We recorded breathing activity using both the SG embedding 4 piezoresistive sensors, located on lower thorax and abdomen, and a reference flowmeter. We estimated RR using three different signals from SG: i) the average along sensors of the band-pass filtered piezoresistive signals (Xbp), the first principal component (P1st) and the average along components of the sub-set of principal components needed to obtain an accounted variance of 0.95 (P95). On the basis of the RR computed with the reference flowmeter, we obtained that, on average, the error committed relying on P95 is 1.02 bpm, the error obtained considering Xbp is 1.04 bpm and the error obtained using P1st is 1.13 bpm. However, the use of P95 allows obtaining a better estimation at high speed tasks than both Xbp and P1st. This finding may suggest that the use of PCA and, specifically, of the signal obtained as the average along all those components required to obtain an accounted variance of 0.95 may be conveniently used to selectively discard breathing-unrelated components, thus improving the estimation of RR.