Research

By developing computational tools, the OMNI Lab aims to address key needs in modern medicine: (i) automated analysis relieves the requirement for highly-skilled radiologists; (ii) discovery of novel markers for diagnostic screening; (iii) software for portable devices, thus broadening access to high-quality care in the developing world; and (iv) investing in sophisticated software while leveraging existing imaging hardware provides a large cost benefit to an already constrained healthcare system. To achieve these goals, our research focuses on the following topics:

Characterising the Fetal Brain
⁠⁠2D-3D US Reconstruction
Clinical Translation
⁠⁠Deep Learning Methodology

Characterising the Fetal Brain

Normal development of the human brain can be characterized by precisely timed growth and folding of its surface (cortex), with deviations often associated with poor cognitive outcomes. Advances in ultrasound (US) imaging technology now make it possible to visualize the cortex and screen for brain abnormalities before birth, from as early as 14 gestational weeks (GW). Working closely with the INTERGROWTH-21st Consortium and healthcare professionals, we develop a range of computational tools for assessing fetal health from US images.

Normative spatiotemporal fetal brain maturation with satisfactory development at 2 years

Namburete et al. Nature (2023)

Subcortical segmentation of the fetal brain in 3D ultrasound using deep learning

Hesse et al. NeuroImage (2022)

BEAN: Brain Extraction and Alignment Network for 3D Fetal Neurosonography

Moser et al. NeuroImage (2022)

Fully-automated alignment of 3D fetal brain ultrasound to a canonical reference space using multi-task learning

Namburete et al. Medical Image Analysis (2018)

Cortical Plate Segmentation Using CNNs in 3D Fetal Ultrasound

Wyburd et al. PIPPI (2021)

Learning-based prediction of gestational age from ultrasound images of the fetal brain

Namburete et al. Medical Image Analysis (2015)

Prototype learning for explainable brain age prediction

Hesse et al. WACV (2024)

Assessment of regional cortical development through fissure based gestational age estimation in 3D fetal ultrasound

Wyburd et al. MIUA (2021)

Multi-task CNN for Structural Semantic Segmentation in 3D Fetal Brain Ultrasound

Venturini et al. MIUA (2021)

2D to 3D Reconstruction

In neuroimaging research, 3D image data is the mainstay for representing anatomical details. However, in conventional clinical practice, pre- and post-natal assessments are performed with 2D video or static US images. The sonographers need to interpret the relationships between the 2D views and 3D brain anatomy and mentally reconstruct a 3D image given just the 2D information. Our goal is to develop methods to reconstruct a 3D brain scan from 2D freehand video acquisitions. We foresee that this would have applications in perinatal clinics, and in LMICs.

Sensorless volumetric reconstruction of fetal brain freehand ultrasound scans with deep implicit representation

Yeung et al. Medical Image Analysis (2024)

Learning to map 2D ultrasound images into 3D space with minimal human annotation

Yeung et al. Medical Image Analysis (2021)

Geometric Transformation Uncertainty for Improving 3D Fetal Brain Pose Prediction from Freehand 2D Ultrasound Videos

Ramesh et al. MICCAI (2024)

Adaptive 3D Localization of 2D Freehand Ultrasound Brain Images

Yeung et al. MICCAI (2022)

RapidVol: Rapid Reconstruction of 3D Ultrasound Volumes from Sensorless 2D Scans

Eid et al. ISBI (2025)

Sli2Vol: Annotate a 3D Volume from a Single Slice with Self-supervised Learning

Yeung et al. MICCAI (2021)

Clinical Translation

Deep learning has shown great promise in the research domain, but significant barriers still exist which limit the clinical translatability of deep learning models. Our research aims to tackle these barriers focusing on the challenges of: distributed data and data privacy, model compression, and the development of clinically meaningful biomarkers.

Challenges for machine learning in clinical translation of big data imaging studies

Dinsdale et al. Neuron (2022)

Distributed Data and Data Privacy

UniFed: A unified deep learning framework for segmentation of partially labelled, distributed neuroimaging data

Dinsdale et al. bioRxiv (2024)

SFHarmony: Source Free Domain Adaptation for Distributed Neuroimaging Analysis

Dinsdale et al. ICCV (2023)

FedHarmony: Unlearning Scanner Bias with Distributed Data

Dinsdale et al. MICCAI (2022)

Deep learning-based unlearning of dataset bias for MRI harmonisation and confound removal

Dinsdale et al. NeuroImage (2021)

Model Compression

STAMP: Simultaneous Training and Model Pruning for Low Data Regimes in Medical Image Segmentation

Dinsdale et al. Medical Image Analysis (2022)

Low-Memory CNNs Enabling Real-Time Ultrasound Segmentation Towards Mobile Deployment

Vaze et al. IEEE Journal of Biomedical and Health Informatics (2020)

Biomarker Development

Learning patterns of the ageing brain in MRI using deep convolutional networks

Dinsdale et al. NeuroImage (2021)

Is Your Style Transfer Doing Anything Useful? An Investigation Into Hippocampus Segmentation and the Role of Preprocessing

Kalabizadeh et al. MLCN (2024)

Brain Ages Derived from Different MRI Modalities are Associated with Distinct Biological Phenotypes

Roibu et al. SDS (2023)

Deep Learning Methodology

Fundemental to to our research is the developement of state of the art DL methodology. Key topics include segmentation, interpretability and domain adpatation.