Nasopharyngeal carcinoma (NPC) poses significant clinical challenges during radiotherapy. Many patients experience ulceration, difficulty eating, rapid weight loss, and marked anatomical changes, which can reduce treatment accuracy over time. For some patients, these changes necessitate adaptive radiotherapy, where treatment plans are modified mid-course. However, clinicians currently lack reliable, data-driven methods to determine which patients require adaptive radiotherapy and the optimal timing for intervention.

This project addresses this gap through the development of an AI- and data analytics–driven pipeline using longitudinal CT imaging collected during treatment. By analysing continuous anatomical changes, the project seeks to identify quantitative thresholds that indicate when replanning is clinically warranted. This enables adaptive radiotherapy to shift from a reactive process to a predictive and structured approach.

The project delivers a two-pronged impact. At the treatment management and healthcare system level, early identification of patients likely to require adaptive radiotherapy supports better resource planning, scheduling, and workflow optimisation within radiotherapy services. At the patient level, the approach advances precision medicine by enabling patient-specific treatment adaptation based on individual anatomical changes, improving treatment accuracy and personalisation.

A key strength of the project is its emphasis on capacity building and interdisciplinary translation. Early-career researchers from theoretical physics backgrounds are embedded in a clinical environment, applying physics-based modelling and data analytics to address a real-world oncology problem. This integration strengthens cross-disciplinary collaboration and equips researchers to translate theory into clinically actionable solutions.

The project’s ultimate goal is to establish a clinical decision-support pipeline that can be integrated into routine radiotherapy workflows. While initially focused on NPC, the framework has translational potential for other cancer types, subject to further validation and site-specific data.