DPhil researcher · University of Oxford

Machine learning for reproductive and maternal health.

I develop computational methods for objective, reproducible analysis of the placenta and its pathology.

Explore my research View publications

Big Data Institute · EPSRC Health Data Science CDT
Jesus College · University of Oxford

Oxford, United Kingdom

Published HAPPY pipeline figure showing hierarchical analysis of placental whole slide histology — How HAPPY worksAnalysis at the level of nuclei, cells and tissue.
YOLODetect every nucleus
Locate nuclei across the whole slide image.
CNNClassify cell identity
Assign detected nuclei to biologically defined placental cell types.
GNNInfer tissue structure
Build a cell graph and classify the tissue around each cell.
Explore the open research pipeline

Featured research and presentation

Making the placenta computationally legible.

HAPPY analyses whole slide histology at three linked levels: nuclei, cell identity and the cell to tissue graph. It produces quantitative measurements of placental composition and spatial structure.

Nature Communications · open sourceEmma Walker with Christoffer Nellåker

Watch the presentationHAPPY AI to understand the placenta and diseases

Research question

How can tissue images produce measurements that reflect biology and support clinical research?

I combine computational pathology, representation learning and women’s health.

Selected research

Measuring placental structure and disease.

I develop methods, validate their biological meaning and test their relevance to clinical research.

Computational pathology

HAPPY

HAPPY measures the cellular composition and spatial organisation of placental histology. Its biologically inspired pipeline links tissue measurements with pathology and clinical outcomes.

Digital histologyComputer visionOpen source

Deep phenotyping

Placental composition

Quantitative analysis of tissue composition across major placental lesion types, producing an organ level view of disease.

Representation learningGraph methodsWomen’s health

Health data science

Learning from rare outcomes

Methods for analysing rare outcomes in population cohorts using proteomic and clinical data.

UK BiobankProteomicsModel evaluation

How I work

Biology first.

I design models around the organisation of tissue. Their measurements should be biologically coherent, reproducible and useful to researchers and clinicians.

Model how cells are organised within tissue.

Check that the model’s findings reflect real biology.

Build methods that others can reproduce and clinicians can use.

Selected publications