Multimodal LLM Agents for Early Detection of Brain Tumour Recurrence

About the Project

Brain tumour recurrence is a leading cause of morbidity and mortality, especially in high-grade gliomas where recurrence rates exceed 70% within two years of treatment. Early detection is critical to enable timely surgical re-intervention, radiotherapy, or systemic therapy. However, recurrence diagnosis is challenging due to overlapping radiographic features between post-treatment changes (e.g., pseudoprogression) and true recurrence, often leading to delays or unnecessary interventions.

This project will develop a multimodal Large Language Model (LLM) Agent that integrates MRI imaging, histopathology slides, and longitudinal clinical notes to identify early signs of recurrence. The agent will employ vision-language transformers to extract joint representations from medical images and textual records, enabling holistic reasoning over spatial tumour features, cellular morphology, and clinical trajectories. Recent advances in function-calling LLM agents will be leveraged, allowing the model to dynamically query specialised sub-models for MRI lesion segmentation, volumetric growth rate estimation, and pathology grading, before synthesising an interpretable recurrence risk report.

A large multi-institutional dataset will be assembled, incorporating pre- and post-treatment MRI scans, pathology data, and anonymised clinical notes. The system will be trained and validated against radiological and histopathological ground truth. Furthermore, uncertainty quantification and automatic failure detection will be embedded to enhance clinical trustworthiness and safety.

The student will gain expertise in multimodal LLM architectures, clinical agent workflow design, and real-world medical AI deployment. They will work closely with neuro-oncology teams at partner hospitals, learning how tumour recurrence is diagnosed and managed in clinical pathways. By combining cutting-edge AI with multi-source clinical data, the project aims to produce a decision-support agent capable of detecting recurrence earlier than current practice, improving survival outcomes and reducing the physical and psychological burden of delayed intervention.

Candidates are expected to hold (or be about to obtain) a minimum upper second class honours degree (or equivalent) in computer science, mathematics or equivalent technical areas. Candidates with experience in Machine Learning, Computer Vision, and large Language Models are encouraged to apply.

Applicants must make direct contact with preferred supervisors before applying. It is your responsibility to make arrangements to meet with potential supervisors, prior to submitting a formal online application.

To be considered for this project you MUST submit a formal online application form – on the application form select PhD Biomedical Imaging Sciences Programme. Full details on how to apply can be found on the Website: How to apply for postgraduate research at The University of Manchester

If you have any queries regarding making an application please contact our admissions team FBMH.doctoralacademy.admissions@manchester.ac.uk