Accelerometer-based measurement of exercise and mobility for pulmonary rehabilitation

The aim of this study is to develop a wearable sensor system to remotely monitor the exercise activity of chronic obstructive pulmonary disease (COPD) patients. Data were collected and analyzed from 8 COPD patients who wore accelerometers on the arms and legs while performing a series of aerobic exercises. Exercise identification was based on the output of a neural network trained with examples from each of the exercise conditions. Sensitivity (% correct classifications) of the neural network was high, ranging from 93 to 98%, and specificity (100% - % errors where one exercise classified as another) was above 97.5% in all cases. In addition, 3 patients followed an expanded protocol that included activities representing mobility in daily life in addition to the original aerobic exercise set. Planar projections of the multidimensional feature sets derived from accelerometer signals show well-defined clusters for most activities, indicating that these signals contain sufficient information to reliably discriminate among the various activities.