Monitoring asthma medication adherence through content based audio classification

Chronic respiratory diseases, such as asthma, are very common around the world and have been shown to have a significant effect on the quality of life of patients. A crucial component for the effective management of asthma is the adherence of patients to their medication prescription, which can be separated into two distinct and equally important components, i) the adherence of patients to the time schedule for the use of their inhaled medication and ii) their competence in using the inhaler correctly and effectively. Aiming in this direction the current paper investigates three different algorithmic approaches not only for the detection of Metered Dose Inhaler actuations but for the understanding of the overall inhaler technique of patients. More specifically, Short Time Fourier Transform is used as the basis for the extraction of features that are then used for the classification of 4 events (inhaler actuation, patient inhalation, patient exhalation, background noise) using three distinct algorithmic approaches (Support Vector Machines, Random Forests and AdaBoost). The final experimental results demonstrate that Adaboost outperforms the alternative approaches leading to accuracies above 96%.