SoundCatcher: Acoustic Emission with Machine Learning for Monitoring of MDI Adherence and Detection of Non-optimal Actuation Performance
Ellinor Nilsson1, Mats Josefson2, Lisa Holmstén1, Patrik Andersson1, Lubomir Grandinarsky3,
Matthew Ferriter4, and Lars Karlsson1
1Inhalation Pharmaceutical Development, Pharmaceutical Technology & Development, Operations, AstraZeneca, Gothenburg, Sweden
2Oral Product Development, Pharmaceutical Technology & Development, Operations, AstraZeneca,
Gothenburg, Sweden
3Innovation Strategies & External Liaison, Pharmaceutical Technology and Development, AstraZeneca, Gothenburg, Sweden
4Inhalation Pharmaceutical Development, Pharmaceutical Technology & Development, Operations, AstraZeneca, Durham, USA
Summary
Acoustic Emission (AE) profiles, in combination with machine learning (ML) algorithms have been shown to provide valuable quality related information for inhalation devices and formulations. Here, the technique applicability is extended to patient adherence related monitoring. A feasibility study focusing on the attribute of non-optimal or unexpected actuation performance was carried out using pMDIs as model devices. The approach was evaluated and in an experimental set-up designed to mimic a domestic environment with a varying soundscape to properly challenge the algorithm, while using three different technical configurations to produce of non-optimal/unexpected actuation performance. Using an image-based ML model with Convolutional Neural Networks (CNN) it was possible to obtain highly accurate predictions, approximately 90%, of non-optimal actuations.
Key Message
Acoustic Emission combined with Artificial Intelligence (AI) algorithms is a very powerful and sensitive technology capable of providing a low cost and robust method to improve usability and patient compliance, also in a homelike environment with a constantly changing soundscape.