A Systems Approach to Modelling Tumour and Tissue Response in Radiotherapy for Head-and-Neck Cancer

The heterogeneity of patient responses to radiotherapy poses significant challenges in head-and-neck cancer (HNC), with outcomes ranging from severe toxicities to loco-regional recurrence (LRR). Multi-modal biomarker data—encompassing genetic, dosimetric, and imaging domains—offers an opportunity to address this variability by leveraging all routinely collected radiotherapy data to guide personalised treatment. However, building comprehensive computational frameworks from these diverse data sources is complicated by ethical considerations, data heterogeneity, and reproducibility challenges. To address these issues, this thesis adopts a systems approach to understanding radiotherapy response, focusing on biomarker discovery at the patient, treatment, and tumour levels. This research employed a multi-faceted approach, beginning with the ethical evaluation of computational frameworks for multi-modal prediction of radiotherapy response. Key Data Hazards, such as interpretability and linkage of sensitive datasets, were identified alongside proposed strategies to mitigate these risks in current and future research. Patient-level analyses implemented genome-wide association studies (GWAS) to explore genetic contributors to radiotherapy toxicities. Common genetic variants associated with acute xerostomia and late dysphagia were identified, providing insights into the genetic basis of treatment response; however, larger national and international cohorts are needed to validate these findings. At the treatment level, a novel spatial mapping methodology classified patterns of loco-regional recurrence relative to planned dose distributions. The results revealed that most HNC recurrences occurred in high-dose regions, suggesting radioresistance and the potential need for targeted therapies. This methodology proved both reproducible and generalisable, with successful application to both HNC and breast cancer datasets. Tumour-level analyses applied a radiomic pipeline to clinical target volumes (CTVs) extracted from planning CT scans, for the development of predictive models for recurrence risk. These models achieved up to 82% accuracy for internal datasets, demonstrating the potential of intensity-based radiomic features for classifying recurrence. The results indicate the potential for computational frameworks to utilise multi-modal data collected across the radiotherapy treatment pathway for biomarker discovery and response prediction. By independently analysing patient, treatment, and tumour-level datasets, this research lays the foundation for future integration of radiotherapy response biomarkers into comprehensive predictive models. This work contributes methodological approaches and insights into radiotherapy response while seeking to address the challenges of reproducibility and data heterogeneity, paving the way for personalised radiotherapy strategies.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2025

Author dc:creator

• Welsh, Ceilidh

Advisors dc:contributor.advisor

• Jena, Rajesh
• Barnett, Gillian C

• oncology
• head-and-neck cancer
• radiotherapy
• local recurrence
• tissue toxicity

Licence

• https://www.repository.cam.ac.uk/bitstreams/cc23fa5c-017e-4f79-b736-274225baceb1/download
• http://purl.org/NET/rdflicense/allrightsreserved

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.121839

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/390214