Beyond auto-segmentation: the case for planning and dosimetry AI in head and neck radiation oncology

Artificial intelligence (AI) has been rapidly integrated into radiation oncology workflows, with the most visible successes occurring in auto-segmentation of target volumes and organs at risk. While these advances have delivered meaningful efficiency gains and improved standardisation, their impact on clinical outcomes in head and neck cancer remains indirect. In head and neck radiotherapy, long-term toxicity and quality of life are influenced not only by contouring accuracy but more directly by nuanced dosimetric trade-offs during treatment planning. This narrative review synthesises the current landscape of AI applications in head and neck radiotherapy planning and highlights the limitations of a segmentation-dominated paradigm. We review the evolution of planning AI from knowledge-based statistical models to deep learning-based dose prediction and emerging reasoning-driven frameworks. We also aim to discuss how these approaches may function as complementary components of a broader planning intelligence ecosystem. In addition to potential patient-level benefits, we examine the system-level implications of planning and dosimetry AI. Finally, we outline key considerations for the safe evaluation and deployment of planning AI. Together, this review positions planning and dosimetry AI as an enabling infrastructure for predictive, adaptive and equitable head and neck radiotherapy.