Audio-Based Detection of COPD Exacerbations via Two-Stage Cough Detection and Classification
2026 (English)In: 2026 21st Conference on Computer Science and Intelligence Systems (FedCSIS), Polish Information Processing Society PTI , 2026, p. 555-560Conference paper, Published paper (Refereed)
Abstract [en]
Chronic obstructive pulmonary disease (COPD) is a chronic respiratory condition characterized by frequent exacerbation episodes. The paper presents an approach to detecting COPD exacerbations based on audio recordings performed by patients. In the proposed algorithm, the audio signal is processed in two stages. First, it is converted to a mel spectrogram and analyzed by a neural network to extract cough segments. Next, the detected coughs are passed to a cough classification model that determines whether the coughs are induced by a COPD exacerbation. The proposed approach was cross-validated using two datasets for cough detection and classification, and further evaluated on a synthetic dataset of recorded user narratives. The results show that effective cough detection is feasible, achieving an AUC of 0.989 and 0.967 on the synthetic dataset. COPD cough classification remains more challenging, with an AUC of 0.794, particularly due to the limited size of publicly available COPDfocused datasets.
Place, publisher, year, edition, pages
Polish Information Processing Society PTI , 2026. p. 555-560
Keywords [en]
Chronic obstructive pulmonary disease, Modeling, Recording, Signal detection, HTTP, Spectrogram, Timing, Training, Pulmonary diseases
National Category
Respiratory Medicine and Allergy
Identifiers
URN: urn:nbn:se:mdh:diva-79394DOI: 10.15439/2026F8312Scopus ID: 2-s2.0-105054703109ISBN: 978-83-977088-1-5 (electronic)OAI: oai:DiVA.org:mdh-79394DiVA, id: diva2:2114823
Conference
2026 21st Conference on Computer Science and Intelligence Systems (FedCSIS), 23-26 August, 2026, Riga, Latvia
2026-11-032026-11-032026-11-04Bibliographically approved