Interpretable Medical Imaging
AI research in medical imaging is extensive, but translating models into routine clinical use remains difficult. Clinicians need outputs that can be assessed in context, including what a model does not know, rather than a prediction alone. Recent evidence identifies limited clinician involvement and the absence of clinically meaningful evaluation as important barriers to the use of explainable AI in routine workflows.
My work with OCT-Assist and Bayesian deep learning for OCT imaging focuses on uncertainty-aware and interpretable methods for optical coherence tomography. The aim is to produce models whose outputs support clinical judgement, including identifying uncertain segmentations, image artefacts, and the reliability of clinically relevant measurements.
Medical Outcomes Intelligence
Good health-data science depends on measuring outcomes that are meaningful, comparable, and practical to collect. Through the NIHR Outcomes Tracker project and work with the COMET Initiative, I am interested in making outcomes data more useful across research studies and healthcare settings.
Outcomes Tracker, funded through NIHR Better Methods Better Research, applies large language models and agentic methods to identify and track medical outcomes in clinical-trial literature. The project aims to surface outcome-reporting bias and provide evidence that can support the review and uptake of core outcome sets.
Core outcome sets establish a minimum set of outcomes to measure and report for a health condition or area of care. They can reduce research waste, support comparison between studies, and help ensure that the questions measured reflect what matters to patients, clinicians, and decision-makers. This work connects methodological rigour with practical data infrastructure, so that evidence can travel further and inform better health outcomes.
Learn about core outcome sets through the COMET Initiative ↗