Recognition of breathing patterns helps clinicians to understand acute and chronic adaptations during exercise and pathological conditions. Wearable technologies combined with a proper data analysis provide a low cost option to monitor chest and abdominal wall movements. Here we set out to determine the feasibility of using accelerometry and machine learning to detect chest-abdominal wall movement patterns during tidal breathing. Furthermore, we determined the accelerometer positions included in the clusters, considering principal component domains. Eleven healthy participants (age: 21 ± 0.2 y, BMI: 23.4 ± 0.7 kg/m2, FEV1: 4.1 ± 0.3 L, VO2: 4.6 ± 0.2 mL/min kg) were included in this cross-sectional study. Spirometry and ergospirometry assessments were performed with participants seated with 13 accelerometers placed over the thorax. Data collection lasted 10 min. Following signal pre-processing, principal components and clustering analyses were performed. The Euclidean distances in respect to centroids were compared between the clusters (p < 0.05), identifying two clusters (p < 0.001). The first cluster included sensors located at the right and left second rib midline, body of sternum, left fourth rib midline, right and left second thoracic vertebra midline, and fifth thoracic vertebra. The second cluster included sensors at the fourth right rib midline, right and left seventh ribs, abdomen at linea alba, and right and left tenth thoracic vertebra midline. Costal-superior and costal-abdominal patterns were also recognized. We conclude that accelerometers placed on the chest and abdominal wall permit the identification of two clusters of movements regarding respiration biomechanics.
Keywords: Abdominal wall; Chest wall; Clustering; Machine learning; Wearable technology.
Copyright © 2019 Elsevier Ltd. All rights reserved.