Feed

Artificial intelligence (AI) demonstrates potential throughout the cancer care continuum, with evidence supporting its application in medical imaging for detection, staging, treatment planning, and prognostic evaluation. However, clinical translation is hindered by challenges, data curation and annotation, model interpretability, generalizability, and integration into workflows. To address these barriers and provide guidance, a national multidisciplinary expert panel in China developed this consensus. A modified Delphi approach was employed to achieve expert consensus, involving 81 specialists in radiology, nuclear medicine, oncology, and imaging AI from university hospitals across China. These experts completed a survey containing 30 core statements addressing AI applications in clinical cancer imaging, spanning cancer screening, diagnosis, staging, treatment planning, response assessment, prognostic prediction, data governance, and implementation. Consensus was defined as a mean score ≥ 7 on a 9-point Likert scale, with ≥ 80% of experts scoring ≥ 7. All 30 statements fulfilled these thresholds, with mean scores ranging from 8.06 to 8.58 and the proportion of experts scoring ≥ 7 ranging from 86% to 98%. This expert consensus summarizes key AI application scenarios in cancer imaging and delivers recommendations on data acquisition and annotation, model development and validation, interpretability, multicenter generalizability, privacy-preserving collaboration, clinical workflow integration, and post-deployment monitoring, while contextualizing these statements across major clinical application domains and key implementation challenges in practice. It further identifies priority research directions, including the integration of multimodal and multi-omics data, longitudinal modeling of treatment response, and prospective validation in clinical settings, to support the safe, effective implementation of AI technologies in cancer imaging. KEY POINTS: Question AI translation in oncologic imaging remains constrained by limitations in rigorous validation, actionable interpretability, standardization, governance, and workflow integration. Findings Eighty-one Chinese experts reached consensus on 30 clinically practical statements covering AI applications from early detection to deployment. Critical relevance statement Recommendations highlight expert-supervised labeling, multicenter validation, subgroup evaluation, interpretable outputs, privacy-secured collaboration, integrated workflows, and post-implementation surveillance.