A Comprehensive Review of Deep Learning Approaches for Multi-Class Chest Disease Classification Using X-Ray Images

Chest illnesses, which include a variety of ailments affecting thoracic architecture such as congenital heart disease, COVID-19, pneumonia, and pediatric pneumonia, are a major source of morbidity and mortality worldwide. The most popular diagnostic technique is still chest X-ray (CXR) imaging, but manual interpretation is laborious, subjective, and prone to inter-observer variability. While deep learning (DL) has shown promising results in CXR analysis, the majority of current research and reviews concentrate on binary or disease-specific classification, providing little support for true multi-class diagnosis, which is crucial in real-world clinical settings where it is necessary to distinguish between multiple or overlapping pathologies. Furthermore, previous surveys have not sufficiently addressed issues including dataset imbalance, a lack of pediatric data, poor cross-dataset generalization, and inadequate model interpretability. In order to close this gap, this review offers a methodical and comparative analysis of DL techniques for multi-class chest disease classification using CXR images. These techniques include traditional convolutional neural networks, vision transformers, hybrid architectures, and lightweight models assessed on popular public datasets. The review prioritizes clinical robustness, deployment feasibility, dataset preparation techniques, evaluation methodologies, and interpretability in addition to diagnostic accuracy. A comprehensive analysis of recent developments combining progressive feature extraction and attention mechanisms is presented, emphasizing their potential to enhance therapeutic trust and reliability. Overall, this study offers practical insights and future research objectives for accurate multi-class chest illness diagnosis, bridging the gap between algorithm-centric studies and practically deployable AI systems.