Modelling exacerbation of chronic obstructive pulmonary disease using GRU-based analysis of cough sounds

Background: Chronic obstructive pulmonary disease (COPD) is characterized by progressive airflow limitation and recurrent acute exacerbations (AECOPD) that accelerate disease progression. Although prior work has focused on classifying distinct respiratory conditions, detecting subtle physiological changes within COPD patients that can precede exacerbations remains underexplored. Cough acoustics offer a promising non-invasive biomarker, yet their ability to reflect exertion-induced respiratory state changes has not been systematically investigated. Objective: To determine whether acoustic analysis of cough signals can differentiate stable (pre-exertion) and exercise-induced hyperinflation states in COPD patients, using recordings collected before and after a standardized 6-minute walk test (6MWT). Methods: We conducted an empirical study analyzing cough recordings from 21 COPD patients (36 sessions, pre- and post-6MWT). Acoustic preprocessing and feature-based learning were applied to classify using a GRU-based attention model. External validation was performed on the ICBHI 2017 public respiratory dataset. Results: The proposed approach demonstrated high classification performance on the augmented pilot dataset. On the ICBHI 2017 dataset, the model achieved performance comparable to published benchmarks, demonstrating generalization beyond the pilot cohort. Conclusion: Cough acoustic analysis can capture physiological changes associated with exercise-induced respiratory stress in COPD. As a pilot feasibility study on a limited clinical dataset, these findings motivate larger prospective validation studies. The lightweight GRU architecture supports future deployment on mobile and edge devices for remote COPD monitoring.