Ziyi Fan1, Yuqing Ye*1, Ying Ma1,2 & Jesse Zhu*1
1University of Western Ontario, 1151 Richmond Road, London, N6A 3K7, CANADA
2Ningbo Inhal Pharma, 108 Yuxi Road, Suzhou, 215125, China
Summary
Inhaled therapy by dry powder inhalation has been one of the most popular routes to treat lung diseases. However, suboptimal inhaler uses, such as incorrect holding gestures, have been reported as one of key concerns affecting therapeutic outcomes. As a feasible strategy, digital dry powder inhalers (DPIs) with integrated motion monitoring system have been proposed. Facilitated by artificial intelligence (AI) technologies, the system is expected to achieve providing inhaler use error (improper positioning) information. In this study, we successfully designed a novel DPI digital motion-based monitoring system and employed widely used AI techniques to detect misuse scenarios captured from participants during inhalation. Five different AI models were evaluated, with the Extremely Randomized Trees (ExtraTrees) classifier, an ensemble learning algorithm that constructs multiple decision trees with randomized feature splits to reduce variance and enhance generalization, achieving the highest average cross-validation accuracy (0.9625) and the best test accuracy (0.9667), demonstrating strong generalization and robustness in classifying inhalation-related activities. Our findings demonstrate that AI techniques can be effectively applied to misuse detection with high accuracy, offering a promising solution to enhance the efficacy of inhaled therapy.