Development and application of an AI-empowered acoustic monitoring system for misuse detection in dry powder inhalers

Misuse and poor inhalation techniques remain persistent issues in pulmonary drug delivery via dry powder inhalation. While acoustic-based monitoring has been a feasible strategy, existing approaches often depend on smartphones for signal collection, or wired connections for data transmission, limiting their scalability and practicality in real-world settings. More importantly, few studies have specifically focused on the detection of incorrect DPI usage via digital monitoring systems, and current methods still face limitations in accuracy. Therefore, in this study, an AI-empowered acoustic monitoring system was developed, combining edge sensing and cloud analytics to support continuous signal acquisition and misuse detection. Comprehensive featuring engineering and machine learning analysis have been performed to investigate their influence on inhalation activity recognition. The results suggested that feature fusion could significantly enhance classification performance, with the Support Vector Classifier (SVC) showing 99.5 % overall accuracy during cross-validation and 100% accuracy on the test set, along with rapid training and high stability. This proposed digital system achieves a near-perfect categorization on the inhalation-related events, and effectively detects unexpected exhalation into inhalers, showing strong potential in real-life applications for improved respiratory disease management.