Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Respiratory Inhaler Sound Event Classification Using Self-Supervised Learning

Davoud Shariat Panah, Alessandro N Franciosi, Cormac McCarthy, Andrew Hines

TL;DR

This paper addresses automatic classification of inhaler sounds to assess adherence to inhaler technique across different devices and capture hardware. It adopts the wav2vec 2.0 self-supervised learning framework, pre-training on unlabeled inhaler sounds and fine-tuning on labeled segments, using DPI-Watch (watch-recorded DPI data) and RDA (pMDI data). The results show high within-inhaler performance (UAR around 0.98) when training and testing on the same inhaler, but poor cross-inhaler generalization from pMDI to DPI, with rapid adaptation possible using minimal target-data (as little as 15 seconds). The study also demonstrates the feasibility of smartwatch-based personalized monitoring and coaching, suggesting practical pathways for mobile adherence tools and large-scale clinical trials.

Abstract

Asthma is a chronic respiratory condition that affects millions of people worldwide. While this condition can be managed by administering controller medications through handheld inhalers, clinical studies have shown low adherence to the correct inhaler usage technique. Consequently, many patients may not receive the full benefit of their medication. Automated classification of inhaler sounds has recently been studied to assess medication adherence. However, the existing classification models were typically trained using data from specific inhaler types, and their ability to generalize to sounds from different inhalers remains unexplored. In this study, we adapted the wav2vec 2.0 self-supervised learning model for inhaler sound classification by pre-training and fine-tuning this model on inhaler sounds. The proposed model shows a balanced accuracy of 98% on a dataset collected using a dry powder inhaler and smartwatch device. The results also demonstrate that re-finetuning this model on minimal data from a target inhaler is a promising approach to adapting a generic inhaler sound classification model to a different inhaler device and audio capture hardware. This is the first study in the field to demonstrate the potential of smartwatches as assistive technologies for the personalized monitoring of inhaler adherence using machine learning models.

Respiratory Inhaler Sound Event Classification Using Self-Supervised Learning

TL;DR

This paper addresses automatic classification of inhaler sounds to assess adherence to inhaler technique across different devices and capture hardware. It adopts the wav2vec 2.0 self-supervised learning framework, pre-training on unlabeled inhaler sounds and fine-tuning on labeled segments, using DPI-Watch (watch-recorded DPI data) and RDA (pMDI data). The results show high within-inhaler performance (UAR around 0.98) when training and testing on the same inhaler, but poor cross-inhaler generalization from pMDI to DPI, with rapid adaptation possible using minimal target-data (as little as 15 seconds). The study also demonstrates the feasibility of smartwatch-based personalized monitoring and coaching, suggesting practical pathways for mobile adherence tools and large-scale clinical trials.

Abstract

Asthma is a chronic respiratory condition that affects millions of people worldwide. While this condition can be managed by administering controller medications through handheld inhalers, clinical studies have shown low adherence to the correct inhaler usage technique. Consequently, many patients may not receive the full benefit of their medication. Automated classification of inhaler sounds has recently been studied to assess medication adherence. However, the existing classification models were typically trained using data from specific inhaler types, and their ability to generalize to sounds from different inhalers remains unexplored. In this study, we adapted the wav2vec 2.0 self-supervised learning model for inhaler sound classification by pre-training and fine-tuning this model on inhaler sounds. The proposed model shows a balanced accuracy of 98% on a dataset collected using a dry powder inhaler and smartwatch device. The results also demonstrate that re-finetuning this model on minimal data from a target inhaler is a promising approach to adapting a generic inhaler sound classification model to a different inhaler device and audio capture hardware. This is the first study in the field to demonstrate the potential of smartwatches as assistive technologies for the personalized monitoring of inhaler adherence using machine learning models.

Paper Structure

This paper contains 12 sections, 3 equations, 3 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. Methods
  3. wav2vec 2.0
  4. Data Pre-processing
  5. Model Training
  6. Model Evaluation
  7. Results and Discussion
  8. DPI Models Performance
  9. pMDI to DPI Generalization
  10. Model Adaptation
  11. Overall Analysis
  12. Conclusion

Figures (3)

  • Figure 1: Correct medication delivery technique by two types of respiratory inhalers. Top: pMDI, Bottom: DPI. Adapted from pmdi_inhaler and dpi_inhaler, respectively.
  • Figure 2: wav2vec 2.0 model architecture (pre-training phase). Adapted from Baevski2020wav2vec.
  • Figure 3: Examples of sound waveforms colored with ground truth labels, and their corresponding spectrograms captured using (a) DPI and (b) pMDI inhalers. Sound events include actuation (green), exhalation (blue), and inhalation (red).