AI-Based Quality Assessment of Peak Expiratory Flow in Spirometry

Introduction: Peak expiratory flow (PEF) is crucial to diagnose and monitor asthma, and shows a promising predictive value for asthma and COPD exacerbations in clinical trials. However, there is a lack of clear guidance on assessing PEF quality. This study introduces an AI model to automatically classify the quality of PEF curves. 

Methods: A dataset of 395 spirometry curves, including patients with asthma, COPD, IPF, and healthy individuals, was analyzed. 3 raters independently annotated PEF quality, and their consensus formed the ground truth. The dataset was divided into a training (80%) and validation (20%) set. Additionally, an expert labelled 31 complex curves for a test set. 30 features related to PEF were extracted for model training. A random forest model was trained to distinguish good and bad PEFs.

Results: In the validation/training dataset, 26% of PEF readings were of poor quality, compared to 42% in the test set. The automated model achieved 85% accuracy in distinguishing good and bad peaks in the validation set and 74% in the test set. Features related to artefacts and peak sharpness had the highest impact on classification. 

Conclusions:  This study demonstrates that the AI model can effectively classify good and bad PEF in spirometry, even in absence of clear guidelines. The model’s ability to identify PEF quality issues could significantly aid in improving the accuracy and reliability of PEF-based follow-up of respiratory patients. 

Authors: B. Cuyvers, D. Teixeira e Silva, P. Desbordes, M. De Vos, M. Topalovic