Deep learning in fetal neurosonography: Technical challenges and clinical opportunities - Ana Namburete, University of Oxford

The abundance of data created by an increasingly digitised medical system promises new insights into human physiology and tools for diagnostic support. Machine learning-based methods show great potential in deriving predictive biomarkers from medical images but have yet to see widespread clinical adoption due to major bottlenecks. Specifically, (1) shortage of expertly labelled data for model training; (2) large memory footprints required by deep learning models limit deployability on standard hospital devices; and (3) the fact that models optimally adapt to the distribution represented by the training dataset, at the risk of becoming biased and failing to generalise to new datasets. I will present solutions developed by my group to tackle these challenges. Our scientific interest is in understanding structural changes in the human brain during development and aging. We have built the first population atlas of the fetal brain from clinical ultrasound data, depicting spatial and temporal in-utero maturation, and revealing previously undescribed structural features. Our tools enable neurodevelopmental assessment from early pregnancy and deployment in resource-constrained settings.

Ana Namburete is a Royal Academy of Engineering Research Fellow at the Institute of Biomedical Engineering and an Associate Research Fellow at St. Hilda’s College at the University of Oxford. Her research focuses on machine/deep learning with applications to human brain imaging. She holds a BASc in Engineering Science (Electrical and Biomedical Engineering) from Simon Fraser University, and a DPhil in Engineering Science from the University of Oxford with support from a Commonwealth Scholarship, supervised by Professor Alison Noble. Following a postdoctoral fellowship funded by a Grand Challenge Explorations grant from the Bill and Melinda Gates Foundation, she established the Neuroimage Analysis Group at Oxford’s Engineering Department. Her group’s interest is in developing algorithms to characterise the spatiotemporal dynamics of brain maturation and deterioration. Ana serves on the Management Committee of the EPSRC Centre for Doctoral Training in Biomedical Imaging, run jointly between the University of Oxford and University of Nottingham. She has served as a Program Chair of the Medical Image Understanding and Analysis (MIUA) conference (2020 and 2021). She is on the Organising Committee of the 2021 Workshop on Advances for Simplifying Medical Ultrasound (ASMUS) at the Medical Image Computing and Computer-Assisted Interventions (MICCAI) conference. In 2020, she was awarded a Springboard Award from the Academy of Medical Sciences.

Professor Dr. Manon Benders is director of the Departement neonatology of the Wilhelmina Children’s hospital of the University Medical Center in Utrecht, the Netherlands. She was working as a senior clinical lecturer, King’s college London and honary consultant in Neonatology at Guy’s and St Thomas' NHS Foundation Trust in 2014. During her training she did a junior research fellowship at UCLA in 1997, USA and a neonatal neurology fellowship at University of Geneva (Prof. dr. P.S. Hüppi) in 2006-2007.
Her research focus is on brain development and neonatal brain injury using advanced quantification MRI techniques predicting neurodevelopmental outcome and developing neuroprotective/neuroregenerative strategies from bench-to-bedside. In this research field she is supervising several PhD students and clinical research fellows, working in different national and international research projects, as well in clinical as experimental studies.
The talk will focus on brain development and perinatal brain injury using advanced quantification MRI techniques predicting neurodevelopmental outcome and using perinatal neuroimaging to evaluate neuroprotective/neuroregenerative strategies in newborn babies.