Every year, over 12,000 people in the UK are diagnosed with head and neck cancer, and when the disease returns after treatment, most patients survive only 6 to 9 months. Current follow-up scans catch recurrences in symptom-free patients at low rates, wasting clinic time while the cancer silently advances. This project tests whether artificial intelligence can do better. Researchers will build a repository of CT and PET scans from 698 patients in the PETNECK2 trial, linking each image to detailed clinical and genomic data. They will then test dozens of published radiomic models—algorithms that extract subtle patterns from medical images invisible to the human eye—to see which ones best predict which patients will relapse. The goal is a combined clinical-radiomic classifier that outperforms today’s standard risk assessment. If successful, the tool could shift follow-up from blanket clinic visits to a targeted, patient-initiated system guided by AI, reducing strain on cancer services while catching recurrences earlier. The team will also create a patient education resource explaining how AI works in cancer prediction, addressing a gap in public understanding that currently limits uptake of such tools.
View original technical description
Background: Head and Neck Cancer (HNC) affects over 12,000 individuals annually in the UK, has a rising incidence and has one of the highest disease burdens of any cancer type. The prognosis for recurrent HNC remains extremely poor, with overall survival of between 6-9 months for most patients. Current problem: Routine HNC surveillance is inefficient, with low rates of cancer recurrence detection in asymptomatic patients through routine clinical surveillance. This, coupled with increasing incidence, means an escalating demand on cancer services. Our PETNECK2 trial is testing a novel active surveillance strategy, consisting of PET-CT guided, patient-initiated, symptom-based follow-up and assesses overall survival, effect on fear of cancer recurrence and cost-effectiveness. Rationale for development proposal: Workstream 6 (WS6) of the PETNECK2 programme is investigating alternatives to PET-CT for stratifying patients by recurrence risk, specifically using clinical multi-variable outcome predictive models (OPMs). Recently, interrogation of complex imaging features (radiomics) has rapidly developed because of powerful Artificial Intelligence (AI) algorithms and is now established as a significant new frontier in cancer prediction research. In our recent systematic reviews, adding radiomic features to clinical models outperformed those relying solely on clinical or radiomics data for predicting HNC outcomes. There is a strong rationale therefore to now explore the added benefit of combining radiomic features to the models already being developed in WS6, as current evidence suggests this is likely to outperform clinical-only models. Aims: Create an unrivalled HNC imaging repository with linked prospective clinical and genomic data. Independently evaluate and validate published prognostic HNC radiomic models and benchmark performance. Develop and validate a new clinical-radiomic prognostic classifier. Increase patient awareness and understanding of AI and radiomics in HNC. Address capacity building within areas of unmet need in academic clinical radiology and surgery. Development work plan: Phase 1) (Months 1-19) Collate pre-, on- and post-treatment radiology scans from 698 patients recruited into the PETNECK2 trial. Uniquely, this will be coupled with prospective genomic and clinical data within the setting of a randomised trial. Phase 2) (Months 4-23) a) Externally validate a range of published radiomic models in HNC using the image dataset from phase 1. b) Evaluate the performance of these radiomic models when integrated into the WS6 clinical OPM. c) Develop a meta-model from these individual radiomic models – and re-test performance. Phase 3) (Months 1-23) Develop a patient-focussed educational resource, to improve understanding of the way AI can be applied to cancer prediction models. Impact and future developments: The PETNECK2 repository of clinical data, imaging, digitised pathology slides and biological samples will be a unique resource for: Development of dynamic risk stratification, treatment selection, or post-treatment follow-up for head and neck cancer patients generated by deep learning and neural networks. External validation of existing AI digital pathology, radiomics and clinical prediction algorithms through federated learning via international consortia. Development of this pipeline into a clinical decision support tool through industry collaboration and integration into existing software. Clinical academic career development in surgery and radiology with skill expansion in AI and bioinformatics.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know