Baby love: OxNNet scoops PITCH prize

After winning this year’s PITCH start-up competition at Med-Tech Innovation Expo 2023, Sally Collins from the University of Oxford told us more about the innovation taking the first prize.

First, congratulations on winning PITCH, how did it feel when you were announced as the winner?

I was completely shocked. There were so many fantastic innovations being presented that it must have been an extremely difficult decision, but we were really honoured that the panel picked our software as their winner.

Tell us about your innovation 

Our technology is based on the OxNNet Toolkit which uses a state-of-the-art fully convolutional neural network (OxNNet) to automatically identify and segment solid organs within 3D ultrasound images. It then employs our patented measure of single vessel 3D Fractional Moving Blood Volume (3D-svFMBV) to provide the only validated quantitative measurement of tissue perfusion available for ultrasound imaging. The research leading to its development has been conducted at Oxford University over the last 12 years funded by several institutions including the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NIH), Oxford University itself and the NIHR. The OxNNet Toolkit presents significant potential for application in clinical diagnostics. It is particularly relevant in obstetrics, where ultrasound is the imaging modality of choice given its safety and availability.

What unmet need does it address?

In the UK, eight babies a day are stillborn, leaving families devastated and costing the NHS over £12 million a year. The greatest risk factor for stillbirth is foetal growth restriction (FGR), usually secondary to poor placental implantation. If they survive, a baby with FGR will usually require a prolonged stay on neonatal ITU costing around £1,445 a day. With over 42,000 babies a year born growth restricted, this places a considerable financial burden on the NHS, estimated at over £61 million a year.

The current methods to assign risk of FGR early in pregnancy are based on maternal history and clinician judgement alone. These perform badly and many women deemed ‘low-risk’ are unaware their baby is failing to thrive until they present with a compromised baby or even a stillbirth. If stratified as ‘high-risk’, women receive serial growth scans with the aim of delivery before in utero demise occurs. However, the current risk-assessment performs so poorly that many ‘high-risk’ women deliver well-grown, healthy babies after a pregnancy riddled with anxiety about their baby’s health.

We felt that a reliable, cost-effective first-trimester screening method for FGR was desperately needed. This would not only improve pregnancy outcomes but decrease unnecessary stress for the women undergoing increased surveillance for no clinical benefit. The NHS Long Term Plan aims to reduce stillbirths by 50%, so developing a robust risk assignment tool which enables targeted monitoring and timely delivery should provide a major step towards achieving this.

Many studies have shown that estimation of first-trimester placental volume can predict the growth of the foetus later in pregnancy, with FGR babies usually having abnormally small placentas at the time of the dating scan (c.11-13 weeks’). However, the only commercially available software tools to segment organs such as the placenta from ultrasound images are often inaccurate, labour-intensive, and prone to human error. These tools are only semi-automated requiring an operator to guide the segmentation so they cannot process large volumes of data in real-time, which is a crucial requirement for a population-based screening tool as would be required for prediction of FGR. Further imaging studies have indicated that poor vascularisation of the early placenta is linked to development of pre-eclampsia later in pregnancy. This aligns with the hypothesis that the underlying cause is inadequate conversion of the maternal spiral arteries failing to perfuse the developing placenta sufficiently. If the perfusion of the first-trimester placenta could also be quantitatively assessed prediction of pre-eclampsia should also be possible.

The OxNNet Toolkit can process a simple, static 3D-ultrasound image in real time and automatically identify and map-out the placenta and the placental bed (the interface between the placenta and the uterus). It then calculates the placental volume and uses our patented technique for 3D-svFMBV to measure the perfusion of the placental bed. These metrics are then combined with other risk factors to generate an individual risk-prediction estimate for foetal growth restriction and pre-eclampsia developing later in pregnancy.

If we can identify women at high-risk of FGR early in pregnancy, we can not only instigate increased monitoring of that pregnancy, but also test potential new therapies currently being developed which may increase the growth of the placenta and prevent FGR from occurring in the first place. These therapies may be as simple as high concentration beetroot juice which has shown promise in animal models. As being born with FGR increases the risk of obesity, diabetes & cardiovascular disease in adulthood, any such treatment would have significant long-term health benefits for the baby later in life as well as hopefully preventing stillbirth.

What plans do you have for it now?

Currently, the FirstPLUS and OxPLUS studies (funded by the NIHR and Sir Jules Thorn Translational Biomedical Research Award) are collecting first-trimester 3D-ultrasound images from 7,500 unselected women in two different NHS Trusts to test the utility of the OxNNet Toolkit in the real-world and facilitate a full healthcare economic analysis relevant to the NHS. These studies will form the basis for regulatory approval of the software in this clinical setting. Oxalis Medical is a proposed Oxford University spin-out company being developed with Oxford University Innovation (OUI) to take the OxNNet Toolkit to market as part of a risk-stratification algorithm for the prediction of adverse pregnancy outcomes.

Give us an insight into its potential – it focuses on babies now, but how broad a scope could it have?

Our tool's adaptive nature has proved highly effective when applied to small and diverse datasets (sparse data), outperforming networks trained from scratch. With transfer learning, it has the potential to automatically segment any solid organ or tumour from a static 3D-ultrasound image and estimate the perfusion of the target. This has numerous potential clinical applications including in a wide variety of tumours. Fundamentally, if the target can be captured within a static 3D-ultrasound and identified with the human eye we can probably segment it and estimate perfusion.

Is there anything else you feel that you need to help this innovation realise its full potential?

Money and more research. We are currently seeking initial investment for Oxalis Medical and are applying for funding from a variety of sources to expand the technology to other organs and clinical situations.

Tell us about the PITCH experience? 

The experience of pitching to both funders and an audience of highly innovative people who have all travelled a similar path with their ideas was daunting to say the least but ultimately, an extremely rewarding experience. Thank you for the opportunity.