A REVIEW OF AI-BASED MULTIMODAL TECHNIQUES FOR CHRONIC BRONCHITIS AND COPD DIAGNOSIS AND SEVERITY ASSESSMENT

Abstract

Chronic Obstructive Pulmonary Disease (COPD) and its clinical subtype Chronic Bronchitis (CB) collectively represent one of the foremost global causes of morbidity and mortality, accounting for an estimated 3.23 million deaths annually. Despite advances in spirometric assessment and radiological imaging, early and accurate severity classification of CB and COPD remains challenging owing to phenotypic heterogeneity, comorbidities, and the subjective nature of conventional diagnostic pipelines. This review comprehensively examines the emerging paradigm of AI-driven multimodal frameworks that integrate heterogeneous data sources — including high-resolution computed tomography (HRCT), chest X-rays, spirometry signals, acoustic cough biomarkers, electronic health records (EHR), and serum biomarkers — through deep learning architectures for joint detection and GOLD-stage severity classification of CB/COPD. We analyze state-of-the-art methodologies spanning convolutional neural networks (CNNs), vision transformers (ViT), recurrent sequence models (LSTM), graph neural networks (GNN), and multimodal fusion strategies. Particular attention is given to explainability (XAI) methods such as SHAP and Grad-CAM that ensure clinical trustworthiness, federated learning for privacy-preserving multi-site training, and real-world deployment considerations including regulatory pathways. The paper synthesizes published performance benchmarks, identifies critical gaps, and proposes a unified six layer reference architecture for next-generation AI-based COPD management systems.