Clean-Breathing: a Novel Sensor Fusion Algorithm Based on ICA to Remove Motion Artifacts from Breathing Signal

Although smart-textile solutions based on piezoresistive technology have emerged as a tool to assess unobtrusively breathing activity, their signals are affected by the artifacts related to the subjects' movements during common activities (e.g. walking or running). In order to remove such artifacts, we implemented a novel algorithm combining the information recorded by four piezoresistive textile sensors, which allowed to measure rib cage movements due to the breathing activity, with the data synchronously recorded by an inertial measurement unit. Specifically, by using an Independent Component Analysis (ICA), our algorithm allowed to blindly reduce movement artifacts from the signals recorded by the piezo sensors, leading to highlight the breathing activity. We tested our algorithm in a pilot study, in which we enrolled one healthy subject during a free-running task. In order to assess our approach, we compared the signal spectrum obtained applying our algorithm with the one computed after a standard band-pass filter at 0.05-3 Hz. To this aim, we compared the average amplitude of the Power Spectral Densities (PSDs), computed after both approaches, along three frequency ranges: i) [0–1] Hz, related to the breathing activity; ii) [1-2] Hz, related with the torso rotation, during a running a task; iii) [2-3] Hz, related with the pace, during a running a task. Although the study was performed on one single subject, the results obtained seem to be promising. Indeed, within the range [0–1] Hz, the average reduction of the Power Spectral Density (PSD) is only about 4%, while it is considerably higher within frequency range related to the walking/running activity. Specifically, considering the ranges [1-2] Hz and [2-3] Hz such a reduction is around 39% and 36% respectively.