AI-driven IMU-based monitoring system for detecting dry powder inhaler misuseDownload PDF Download PDF ArticleOpen accessPublished: 08 September 2026Ziyi Fan1,2,Yuqing Ye1,2,Jiale Chen3,Ying Ma1,4 &…Jesse Zhu1,2 npj Primary Care Respiratory Medicine (2026) Cite this articleSave articleView saved research We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.AbstractInhaler misuse has been a persistent challenge for achieving effective pulmonary drug delivery with dry powder inhalers. Among the various misuse patterns, improper device maneuvering, especially incorrect device orientation, is one frequently overlooked yet clinically relevant misuse that impairs therapeutic efficacy. Currently, effective methods for timely and reliable detection of such misuse remain limited. To address this gap, we developed an AI-powered, motion-based digital monitoring system that could detect orientation-related misuse and other commonly observed inhaler-use errors. Specifically, a digital monitoring module was integrated into a Breezhaler® device using inertial measurement unit (IMU) sensors to capture orientation and movement signals during inhaler use. The collected data were processed and analyzed using two feature selection strategies, Boruta and Recursive Feature Elimination with Cross-Validation (RFECV), which were, for the first time, jointly applied to IMU-derived inhaler motion data to identify the most discriminative features. These selected features were then modeled using five machine learning algorithms: AdaBoostClassifier, ExtraTreesClassifier, Support Vector Classifier (SVC), XGBClassifier, and LogisticRegression. The analyses identified distinct optimal feature sets under Boruta (100 features) and RFECV (66 features) strategies, and revealed differences in model performance across predefined inhalation-related activities. The best-performing models, ExtraTrees classifier with Boruta and SVC with RFECV, illustrated excellent performance with accuracy of 0.967 and 0.950, respectively, as evaluated under a five-fold cross-validation scheme. Furthermore, SHAP-based interpretability analysis was conducted to elucidate feature importance and model decision patterns. In summary, this study demonstrates the feasibility of leveraging machine learning for accurate recognition and classification of inhalation activities and highlights the potential of this approach for future application in improving inhaler use and adherence.AcknowledgementsThe authors would like to thank Suzhou Inhal Pharma Co., Ltd., for supporting materials.FundingThis research was funded by Natural Sciences and Engineering Research Council of Canada (NSERC).Author informationAuthors and AffiliationsBiomedical Engineering, University of Western Ontario, London, ON, N6A 5B9, CanadaZiyi Fan, Yuqing Ye, Ying Ma & Jesse ZhuSchool of Biomedical Engineering, Eastern Institute of Technology, Ningbo, 315200, ChinaZiyi Fan, Yuqing Ye & Jesse ZhuNottingham Ningbo China Beacons of Excellence Research and Innovation Institute, The University of Nottingham Ningbo China, Ningbo, 315100, ChinaJiale ChenSuzhou Inhal Pharma Co., Ltd., Suzhou, 215125, ChinaYing MaAuthorsZiyi FanView author publicationsSearch author on:PubMed Google ScholarYuqing YeView author publicationsSearch author on:PubMed Google ScholarJiale ChenView author publicationsSearch author on:PubMed Google ScholarYing MaView author publicationsSearch author on:PubMed Google ScholarJesse ZhuView author publicationsSearch author on:PubMed Google ScholarCorresponding authorsCorrespondence to Yuqing Ye or Jesse Zhu.Ethics declarationsCompeting interestsZiyi Fan has a patent pending with Suzhou Inhal Pharma Co., Ltd. The author, Ying Ma, is employed by Suzhou Inhal Pharma Co., Ltd., but claims no conflicts of interest related to this work. To minimize potential bias, the study procedures, model evaluation criteria, and statistical analyses were predefined, and the results were reviewed and interpreted collectively by all authors. Both positive findings and study limitations are reported transparently. The remaining authors declare no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.Consent for publicationWritten informed consent for publication was obtained from all participants. All the authors also had approved the publication of the research to the journal.Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary informationRights and permissionsOpen Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.Reprints and permissionsAbout this article