Artificial intelligence for chest radiography: an overview of techniques, challenges, and future directions.
Authors
Affiliations (4)
Affiliations (4)
- Department of Radiology, Kobe University, Kobe, Japan. [email protected].
- Department of Radiology, Kobe University, Kobe, Japan.
- Department of Advanced Imaging in Medical Magnetic Resonance, Kyoto University, Kyoto, Japan.
- Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland.
Abstract
This paper presents a critical analysis of AI advancements in chest radiograph (CXR) analysis, tracing its evolution from conventional machine learning to deep learning and multimodal approaches. Early models relied on hand-crafted features, while recent CNNs and transformer-based architectures now achieve diagnostic accuracies exceeding or comparable to radiologists for various thoracic conditions. The recent integration of large language models and multimodal systems-combining imaging with clinical text-has further improved performance and interpretability. Despite notable success, challenges still remain, including model bias, limited generalisation across institutions, and explainability. Solutions such as data sharing, domain adaptation, and explainable-AI techniques are actively being explored. Looking forward, AI systems trained on diverse patient data streams promise enhanced clinical integration and diagnostic precision.