Artificial intelligence and radiomics in lung cancer: from imaging to clinical care

Abstract

Artificial intelligence (AI) and radiomics have emerged as promising approaches in lung-cancer imaging by extracting quantitative features from routine medical images beyond visual assessment alone. Proof-of-concept studies span pulmonary nodule characterisation, molecular biomarker prediction, treatment-response assessment and prognostication, but clinical translation remains limited. This focused narrative mini-review explains the radiomics workflow, summarises representative applications and critically examines methodological, imaging-physics and implementation barriers. Attention is given to reconstruction kernel, slice thickness, radiation dose, image noise, contrast administration, segmentation and phantom-based quality assurance. Evidence remains heterogeneous and is dominated by retrospective, single-centre studies with small or selectively curated datasets, inconsistent external validation and limited assessment of incremental clinical value. Standardisation, harmonisation, explainable AI, federated learning and multimodal integration may address specific barriers but have not demonstrated reliable benefit at scale. Future research should prioritise task-matched multicentre validation, calibration, prospective workflow studies, decision impact, cost-effectiveness and patient outcomes. AI–radiomics therefore remains a promising quantitative imaging framework with established proof of concept but insufficient evidence for routine widespread implementation.