Artificial Intelligence in Pulmonary Medicine: From Diagnostic Imaging to Predictive Analytics in Clinical Practice-Applications, Challenges, and Future Directions

Abstract

Artificial Intelligence (AI) has increasingly transitioned from theoretical research to practical clinical applications in pulmonary medicine. Its integration into respiratory healthcare has been driven largely by advances in machine learning (ML) and deep learning (DL), particularly in medical imaging and predictive modeling. AI systems have demonstrated strong performance in detecting lung cancer, pneumonia, tuberculosis, and chronic obstructive pulmonary disease (COPD), while also enabling risk stratification and outcome prediction using multimodal clinical data [1,2].
However, despite promising diagnostic accuracy reported in controlled studies, real-world clinical translation remains limited. A growing body of recent literature highlights concerns regarding dataset bias, external validation failure, poor generalizability and lack of interpretability, which significantly restrict clinical adoption [6]. Furthermore, comparative studies between major architectures such as CheXNet-style convolutional models and end-to-end CT-based lung cancer systems (e.g., Ardila et al.) suggest that performance superiority in research settings does not always translate into real-world robustness [3,4].
This review critically evaluates current AI applications in pulmonary medicine, compares key methodological approaches, discusses implementation barriers, and highlights future directions including explainable AI, federated learning, and multimodal clinical intelligence systems.