Artificial intelligence in adaptive radiotherapy for head and neck cancer: a systematic review of synthetic CT, automated segmentation and prediction of anatomical change and adaptive RT requirement

Artificial intelligence (AI) is increasingly being explored to overcome the technical and workflow challenges of adaptive radiotherapy (ART) in head and neck cancer (HNC). However, the available evidence remains heterogeneous, with varying degrees of technical and clinical validation. This review evaluated the current role and clinical maturity of AI-enabled ART in HNC. A systematic literature search of PubMed, Scopus and EMBASE was performed from inception to 1 August 2026. Original prospective or retrospective human studies published between 1 January 2020 and 1 August 2026 were eligible. Studies were restricted to photon-based radiotherapy using CBCT, kVCT or MVCT. Eligible studies were classified into four domains: synthetic CT/image enhancement; AI-based auto-segmentation and deformable image registration; anatomical change prediction and adaptive planning; and prediction of ART requirement. Reporting quality was assessed using CLAIM 2024, QUADAS-2 was applied where appropriate. Out of 62 records screened, 25 studies were included. Sixteen studies addressed synthetic CT/image enhancement, seven addressed auto-segmentation/deformable registration, two evaluated anatomical change prediction/adaptive planning, and three investigated predictions of ART requirement, with some overlap between domains. AI consistently demonstrated technical improvements in image quality, CT-number accuracy, segmentation performance and dosimetric agreement. However, evidence for prospective prediction of anatomical change and ART requirement was considerably limited. CLAIM adherence ranged from approximately 68% to 88%, with a mean of 79.9%±5.8%. AI demonstrates substantial technical potential across HNC ART workflows, particularly for image synthesis, segmentation and registration. However, multicentre evidence demonstrating improved clinical decision-making and patient outcomes remains insufficient.