AI for Affordable Prenatal Care

Reuben College continued our ‘AI for Good?’ seminar series in Week 5 of Trinity Term by discussing a topic that definitely involves technology making the world a better place. Ana Namburete, Associate Professor and Tutorial Fellow in Computer Science at Pembroke College and Group Leader of the Oxford Machine Learning in NeuroImaging Lab, joined us to speak about ‘AI for Affordable Prenatal Care: Tools for resource-constrained settings.’

Ana’s talk began with a swift overview of how AI is used in medical imaging, showing how diagnostic decisions are made using visual interpretations. AI uses pattern recognition, maths, and statistics to learn by example and mimics the ways in which a radiologist assesses the images and makes decisions. She explained how Deep Learning works in this context, where vast amounts of data are used to train complex models. Medical imaging in Ana’s lab is used to look at the brain during the two stages of life where it changes most dramatically: during pregnancy, where there is increasing structural complexity, and in late adulthood, where there is structural contamination and deterioration. In this talk, she focused on foetal brain imaging, and how this is used to predict health outcomes more accurately than just measuring size.

In wealthier countries like the UK, pregnant people are offered at least two ultrasound scans during pregnancy, allowing for time for any changes in care to be made, or for further scans to happen if needed. However, in the developing world, where the infant mortality rates are higher, many people live far away from the nearest clinics so all clinical decisions need to be made from a single visit. A lot of these sites, such as a clinic in Blantyre visited by Ana that was monitoring about 8000 pregnancies at the time, can’t rely on standard ultrasound machines for diagnostic decisions, in part because of electricity poverty causing frequent power outages. The clinic in Blantyre was using tools such as the Pinard Horn, a type of stethoscope that can listen to a foetus’s heart rate and determine it’s position, which is understandably not as accurate as an ultrasound. Fortunately, portable point-of-care ultrasound probes are being manufactured with these settings in mind.

As the images that would be produced by these portable probes are 2D, Ana’s team has put together a 3D brain atlas which is designed to serve as a population reference for healthy brain maturation, and created a tool that takes a sequence of 2D images and matches them with a 3D image. This tool can aid a sonographer to navigate during an assessment, and empower them to gain more out of existing medical equipment rather than introducing novel equipment. Combining AI with ultrasound should improve the chances of deploying affordable brain imaging in developing world settings, where sophisticated tools are unavailable.

Over dinner, we continued discussing topics raised by Ana’s talk, answering the questions ‘Would we mind Google having our healthcare records?’ and ‘How do we make sure well-intentioned donors give appropriate medical equipment to resource-constrained settings?’ As ever, these discussions were lively and thought-provoking, with one rapporteur commenting on the first question, ‘We’d trust Google more than the government’…!