A smartphone-based two-stage approach for early screening of pulmonary health and COPD

Wang, Shuaiqi and Li, Yuhua ORCID: https://orcid.org/0000-0003-2913-4478 2026. A smartphone-based two-stage approach for early screening of pulmonary health and COPD. Presented at: 10th International Conference, DMBD 2025, Beijing, China, 19-22 December 2025. Published in: Tan, Y. and Shi, Y. eds. Communications in Computer and Information Science. , vol.2905 Springer, 10.1007/978-981-92-0229-4_2

Full text not available from this repository.

Abstract

Respiratory diseases are a major global health burden, yet auscultation, the most widely used screening tool, suffers from clinician subjectivity and limited support for standardized or remote assessments. To enable automated, low-cost screening on widely available devices, we propose a lightweight, smartphone-based two-stage cascade for respiratory sound analysis. Stage-A classifies Healthy vs. Unhealthy, while Stage-B further differentiates Chronic Obstructive Pulmonary Disease (COPD) from non-COPD within the Unhealthy group. Using HeAR pretrained audio embeddings, we compare a lightweight Convolutional Neural Network (CNN) and a calibrated linear Support Vector Machine (SVM). Experiments on the ICBHI 2017 dataset show that the SVM consistently outperforms the CNN, achieving Stage-A accuracy of 0.920 (Macro-F1 = 0.811, AUC = 0.985) and Stage-B accuracy of 0.900 (Macro-F1 = 0.844, AUC = 0.953). These results demonstrate that combining pretrained embeddings with a simple classifier provides a robust and practical foundation for smartphone-based early screening, particularly in resource-constrained settings.

Item Type:	Conference or Workshop Item - published (Paper)
Date Type:	Published Online
Status:	Published
Schools:	Schools > Computer Science & Informatics
Publisher:	Springer
ISSN:	1865-0929
Last Modified:	02 Aug 2026 06:37
URI:	https://orca.cardiff.ac.uk/id/eprint/188031

Actions (repository staff only)

Edit Item