Automatic identification and accurate temporal detection of inhalations in asthma inhaler recordings.

Asthma is chronic airways disease characterized by recurrent attacks of breathlessness and wheezing. Adherence to medication regimes is a common failing for asthmatic patients and there exists a requirement to monitor such patients' adherence. The detection of inhalations from recordings of inhaler use can provide empirical evidence about patients' adherence to their asthma medication regime. Manually listening to recordings of inhaler use is a tedious and time consuming process and thus an algorithm which can automatically and accurately carry out this task would be of great value. This study employs a recording device attached to a commonly used dry powder inhaler to record the acoustic signals of patients taking their prescribed medication. An algorithm was developed to automatically detect and accurately demarcate inhalations from the acoustic signals. This algorithm was tested on a dataset of 255 separate recordings of inhaler use in real world environments. The dataset was obtained from 12 asthma outpatients who attended a respiratory clinic over a three month period. Evaluation of the algorithm on this dataset achieved sensitivity of 95%, specificity of 94% and an accuracy of 89% in detecting inhalations compared to manual inhalation detection.