From probing the mysteries of a sunken submarine aircraft carrier, resting upright more than 30 metres under the waters off the coast of Plymouth, UK, to attempting to apply machine learning for estimating the age of a fetus in the womb, Dr Lok Hin Lee (Croucher Scholarship 2018) seeks the challenge of the new.
“I have an explorer bent,” explained Lee, former president of the Oxford University Underwater Exploration Group, of his love of diving, which he began in 2017, the first year of his DPhil. But the same could be said of his pioneering research in obstetric ultrasound.
Hong Kong-born Lee, who first went to Oxford in 2012 after completing his secondary school education at the German Swiss International School, is part of a ground-breaking research team led by Professor Alison Noble, of the university’s Institute of Biomedical Engineering. This work, while still at the theoretical stage, could ultimately make obstetric ultrasound more readily available in countries with limited resources, improving the survival chances for unborn babies at risk.
“Ultrasound imaging hardware is getting increasingly cheap. The problem is that the expertise required to interpret ultrasound images can be lacking in regions without sufficient medical knowledge. There are places where expertise is a lot harder to get, so there is a disparity between the availability of the software and the availability of expert opinion,” he said.
“Obstetric sonography interpretation requires a lot of training and education and is inherently difficult, so it becomes a matter of whether or not we can make the task easier for sonographers or possible for the layperson.”
Lee has been working with the multidisciplinary TraCer project – a collaboration between the University of Oxford and King’s College London, Aga Khan University, London School of Hygiene and Tropical Medicine and Oxford Computer Consultants. It uses low-cost probes to capture fetal ultrasound images, which are then sent to a tablet and processed by algorithms to perform pregnancy dating. He has been creating the deep learning algorithms for those images, with the technology being trialled in Kenya, The Gambia and Mozambique.
But there are challenges yet to be overcome “The greater image variation means that it is more difficult for sonographers and our deep learning- based algorithms to do image analysis,” he said. “Moving closer to knowledge transfer means being that much closer to real-world applicability of the algorithms we develop in the lab.“
Lee joined Professor Noble’s team after completing his Master of Engineering in engineering, economics and management at the University of Oxford – but not immediately after. He worked in finance for 18 months, including as an investment analyst.But he was drawn to higher research. “I wanted to use my engineering degree properly and see what I could do in this field,” he said.
“I have always been interested in medical study. My undergraduate decision was between medicine and engineering. I chose engineering in the end, but I selected medical imaging electives in my final undergraduate year to expose myself to the healthcare field.”
Impressed by Professor Noble’s project aimed at making ultrasound more accessible, Lee’s current DPhil project uses deep learning – a kind of artificial intelligence that processes data in a way that emulates the workings of the human brain – to estimate the fetus’s gestational age.
“The first thing a doctor does when you come into hospital with a pregnancy is to try and determine how far along the fetus is. It’s easier when the fetus is small. This is very important in obstetric care because gestational age can be a marker of how well the fetus is doing. For example, we worry if it’s missing developmental stages for a given gestational age.”
The cross-disciplinary team brings together the biomedical engineers and obstetrics researchers from the university’s Nuffield Department of Women’s and Reproductive Health based at John Radcliffe Hospital in Oxford, where Lee has attended numerous obstetric ultrasounds. “The first time I went there I had no idea what to expect,” he said. “The mothers who agree to this have had it explained who I am and what they have agreed to, and my presence. The sonographer is there doing what he usually does and I am there next to him, getting a sense of the difficulties in obstetric sonography.”
Earlier this year Lee, Professor Noble and members of the team published on using their automated framework for estimating gestational age, noting that it quantifies the degree of uncertainty in the estimation. “We wanted to see the degree to which a neural network could predict fetal gestational age without any additional size information, just from the shape and position of fetal anatomies,” Lee said.
Lee, with Professor Noble, has also presented on using their network to localise the fetal heart – a difficult ultrasound task because the heart is small and indistinct – at the 2020 IEEE International Symposium on Biomedical Imaging. For obstetricians, the presentation and characteristics of the fetal heart are often a key sign of fetal health and stress.
Lee finds this work rewarding: “The application is at the forefront. You have a medical, clinical problem, how can we solve that?” he said.
He faces similar problem solving challenges in his diving. Beneath the ocean, as part of a promise to himself to become comfortable on land, in the sea, and air, the cold UK waters have led him to explore the challenges of technical diving – diving with oxygen-enriched gases and closed-circuit rebreathers. “I have a strong desire to try and become as technical and proficient as possible in any hobby I pick up.”
Having conquered land – he has cycled from Oxford to Cardiff and back – he wants to continue exploring the UK’s historical wrecks such as the HMS M2 submarine, which sank in 1932 during a training exercise and is now an official war grave. Next up, hopefully in September 2021, is the SMS Markgraf, a 175m König class German battleship resting upside down in the waters of Scapa Flow, in the Orkney Islands.
For that Lee will need more training, and this desire to continually learn and challenge himself extends to his studies. When he completes his DPhil in the first quarter of 2021, he is hoping to undertake postdoctoral work in the on-going project to make fetal ultrasound increasingly accessible.
“I like doing this with a solid application, something tangible that I’m applying engineering to. It’s a real, concrete clinical question that has immediate potential to improve lives around the world.”
Dr Lok Hin Lee graduated from the University of Oxford in 2017, with a Master’s (MEng Hons) in Engineering, Economics and Management, First Class Honours. He has since completed his DPhil degree at Oxford in 2021, and is currently reading a medical degree at the Vanderbilt University School of Medicine to continue his research. He was awarded a Croucher Scholarship in 2018 for graduate study, and is further supported in his medical studies by the Croucher Foundation by a unique grant.
- Dr Lok Hin Lee’s personal profile (The Croucher Foundation): https://scholars.croucher.org.hk/scholars/lok-hin-lee
- The scientific article Dr Lee published in Medical Ultrasound, and Preterm, Perinatal and Paediatric Image Analysis in 2021: https://www.springerprofessional.de/en/calibrated-bayesian-neural-networks-to-estimate-gestational-age-/18434974
- The presentation Dr Lee gave at a symposium in 2020: https://rc.signalprocessingsociety.org/conferences/isbi-2020/SPSISBIVID0130.html?source=IBP