OPTIMAL-RT - Optimising and Personalising Treatment though multi-modal analysis of radiotherapy for rare cancers

Description

The vision of Leeds Cancer Research UK Radiation Research Centre of Excellence (RadNet Leeds) is to develop smarter, kinder, personalised radiotherapy cancer treatments. Leeds is the largest single-site UK radiotherapy centre (7000+ patients treated annually), that provides a large pool of data and clinical experience. Access to detailed imaging (e.g. CT and MRI scans) and treatment data allows in-depth analysis.

As a global leader in federated learning, a crucial technique for analysing sensitive medical data while maintaining patient privacy across institutions, that addresses a critical ethical concern in medical research. A particular benefit of this approach is the ability to study rarer tumours across multiple centres. We have already demonstrated the power of a federated approach by successfully creating the largest worldwide real-world anal cancer dataset.

The project will support a 0.5WTE data scientist (combined with 0.5WTE CRUK RadNet Leeds funding) who will co-ordinate federated analysis of more complex, rare tumour datasets including anal, liver and brain tumours, exploiting the previously established international network. The focus will be on linking routine radiotherapy, imaging and other medical data to cancer outcomes allowing a better understanding of treatment responses and side effects. This will inform design of future clinical trials of radiotherapy treatments tailored to individual patient needs. We will foster international collaboration, leveraging global scale using data-driven approaches to optimise and personalise future radiotherapy treatments.

Methodology

We will undertake four distinct but interlinked subprojects to answer these four research questions:

1. Linking Radiotherapy and Multimodal Data to Outcomes
2. Immune Resistance and Imaging Biomarkers
3. Establishing an International Anal Cancer Cohort 
4. HCC and Glioblastoma Biomarker Integration and Validation

Method

Linking Routine Radiotherapy and Cancer Outcome Data with Radiology and Digital Pathology (research question 1)

We will establish a robust framework integrating routine radiotherapy and cancer outcome data (via our existing radiotherapy research database, LeedsCAT) with radiology (MRI, PET-CT) and digital histopathology datasets. This will initially focus on anal cancer cohorts but subsequently expand to other cancer types. Collaborating with the Scientific Computing team in Medical Physics and Clinical Engineering at Leeds Teaching Hospitals NHS Trust, we will build on our prior work on standardised extraction of imaging biomarkers for cancer outcome prediction. Additionally, we will analyse diversity (demographic and clinical differences) between routine clinical patient populations and trial cohorts to evaluate their potential impact on treatment response and outcomes.

Exploring Correlates of Immune Resistance and Imaging Biomarkers for Response and Toxicity (research question 2)

We will investigate correlations between immune resistance mechanisms (derived from blood and tissue samples) and imaging biomarkers of treatment response and toxicity. Initial studies will utilise existing trial cohorts, including the multi-centre PLATO trial in anal cancer, and will be separately funded by CRUK RadNet Leeds. These studies will focus on signatures from baseline PET-CT and MRI imaging and their relationship with tissue / blood markers (particularly HPV status and biology), with an emphasis on identifying early predictors of treatment response and resistance (at 3- and 6-month post-treatment). This will allow us to explore mechanisms and predictive markers of resistance in HPV-driven cancers. The findings will subsequently be validated in real-world Leeds patient cohorts and through international collaborations enabled by federated learning (see below).

Establishing an International Cohort for Anal Cancer via Federated Learning (research question 3)

Through the international atomCAT consortium, we will assemble a global cohort of anal cancer patients with multimodal imaging, blood, and tissue sampling data. Collaborating with 16 radiotherapy centres across the UK, Europe, and Australia, we will identify and link existing datasets using our established federated learning platform. Our collaboration with Maastro (Maastricht University, Netherlands) will expand the platform’s capacity to support imaging biomarker selection and validation. Imaging biomarkers for non-invasive phenotyping of immune response and hypoxia, developed in trial cohorts (see above), will be validated as markers of treatment response in these multi-centre cohorts.

Integrating MRI-Derived and Pathological Biomarkers in Glioblastoma and HCC (research question 4)

We will combine MRI-derived imaging biomarkers with pathological data to further develop predictive response signatures for glioblastoma, augmenting prior work on survival prediction using MRI features. These will be validated across international cohorts using federated learning, ensuring their generalisability and clinical utility in diverse populations.

Impact

Our work aims to personalise the use of radiotherapy for patients with rare and poor outcome disease. We will link deep multimodal data which currently sit siloed across radiotherapy, radiology and pathology, allowing us to learn from every patient receiving radiotherapy at Leeds Teaching Hospitals NHS Trust, while also ensuring that these data are connected to and feed into international collaborations and cohort studies. We will use the insights generated from this to design new treatment strategies, to be tested in clinical trials, and – conversely – to evaluate learnings from clinical trials in real-world Yorkshire patients. This will enable the delivery of the long-term vision for the RadNet Leeds - to deliver physically and biologically informed, stratified and personalised radiotherapy for patients.