A Dual-Branch Network for Classifying Pulmonary Disease Using Flow-Volume Loops

View at publisher Summarise with VIRAT
Authors: Jaeun Kim, Hongjun Kim
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Although flow-volume loops from spirometry are fundamental for diagnosing pulmonary diseases, distinguishing subtle morphological patterns remains challenging—even for experienced clinicians, especially when it comes to differentiating between obstructive and restrictive types. In this paper, a deep learning framework is proposed to automatically classify lung diseases based on these loops. As its core methodology, this work presents a dual-branch architecture that decomposes the complex four-class classification task—comprising normal, obstructive, restrictive, and mixed types—into two independent binary classifiers: one for detecting obstructive types and the other for restrictive types. Additionally, to ensure robust performance despite severe class imbalance in real-world medical data, a generative data augmentation strategy operating within the learned latent space of minority classes is employed. By modeling the underlying data manifold with a Convolutional Variational Autoencoder (VAE) and synthesizing high-fidelity samples, the training distribution is effectively rebalanced. This synergistic combination of a dual-branch structure and generative data augmentation significantly enhances the classifier's generalization performance, demonstrated by an increase of 0.026 in accuracy and 0.126 in F1-score over baseline models, marking a substantial advancement towards robust and scalable automated screening for pulmonary diseases.

Fields of Study & Subjects

Chronic Obstructive Pulmonary Disease (COPD) Research COVID-19 diagnosis using AI Artificial Intelligence in Healthcare

Publication details

Journal / source
International Conference on Artificial Intelligence in Information and Communication (ICAIIC)
Pages
1432–1437
Publication date
24 February 2026
Language
EN