ADVANCES IN AUTOMATED LUNG DISEASE DIAGNOSIS: A SYSTEMATIC REVIEW OF MACHINE LEARNING, DEEP LEARNING, SEGMENTATION FRAMEWORKS, AND VISION TRANSFORMERS

Abstract

Lung diseases remain a leading cause of global morbidity and mortality, driving demand for rapid and accurate diagnostic support tools. Recent advancements in machine learning (ML) and deep learning (DL) have led to highly effective automated systems capable of detecting and classifying pulmonary abnormalities using chest X ray and computed tomography (CT) imaging. This survey presents a comprehensive review of traditional ML techniques, convolutional neural networks, hybrid radiomics–DL models, segmentation-based architectures, transformer frameworks, and multimodal fusion approaches. Key publicly available datasets and evaluation metrics are analysed to highlight their influence on diagnostic performance and model generalizability. The findings reveal a clear dominance of DL models, particularly CNNs and transformer-based architectures, while identifying challenges related to dataset bias, limited external validation, lack of interpretability, and inconsistencies in evaluation standards. Research gaps are discussed in the context of clinical adoption, emphasizing the need for transparent, scalable, and privacy-preserving AI solutions. Emerging directions, including federated learning, self-supervised pretraining, and multimodal integration, offer promising pathways toward robust and clinically trustworthy lung disease diagnostic systems. This survey aims to guide future research by synthesizing current progress, identifying limitations, and outlining opportunities for innovation in AI-driven pulmonary healthcare.