Artificial Intelligence in Radiation Oncology: Current applications, clinical impact, and future perspectives. A Narrative Review

Abstract

Radiation oncology is a highly digital and data-intensive discipline in which artificial intelligence (AI) is increasingly being incorporated into the clinical workflow. From image analysis and automatic segmentation to treatment planning, adaptive radiotherapy, radiomics, and outcome prediction, AI has the potential to improve efficiency, reproducibility, and treatment personalization. This narrative review summarizes the principal concepts underlying machine learning and deep learning and examines their current applications across the radiation therapy pathway. Particular attention is given to automated contouring, knowledge-based and automated planning, cone-beam computed tomography (CBCT)- and magnetic resonance imaging (MRI)-guided adaptive radiotherapy, toxicity and tumor-control prediction, and the integration of radiomic, clinical, dosimetric, and molecular data. The available literature suggests that AI can reduce repetitive workload and inter-observer variability and may facilitate more consistent treatment planning. However, technical performance alone does not establish clinical benefit. Major challenges include data quality, dataset bias, limited external generalizability, interpretability, cybersecurity and data governance, and the need to define responsibility when AI-generated outputs are used in patient care. Prospective, multicenter validation and continuous quality assurance are therefore essential. The future of AI in radiation oncology is unlikely to be the replacement of the radiation oncologist, but rather a human–AI partnership in which automation supports clinical expertise. If AI successfully releases clinical time, this resource should be reinvested in patient communication, shared decision-making, research, innovation, and education.