Non-supervised recognition of chest pattern through accelerometry during tidal breathing.

Chest and abdominal wall movements have important implications for monitoring patient routines. In this regard, the use of wearable technologies combined with data analysis can be a low cost and effective option to achieve this end. Here we set out to determine the feasibility of using accelerometry and machine learning to detect chest-abdominal wall patterns during tidal breathing. Eleven healthy participants were included in the study. The participants adopted a seated position using an array of accelerometers placed over respiratory auscultation points during 10 min. Following signals pre-processing, principal component and clustering analysis were performed. The Euclidean distances respect centroids were compared between the clusters. Two clusters corresponding to anatomically costal-superior and costal-abdominal patterns were identified. This result suggests the possibility of detecting regional differences of breathing by using accelerometry.