In this ‘Behind the Research’ series post, we spoke to Kenton O’Hara, Professor of Human-Computer Interaction from the University of Bristol about his project PRISM: Prioritising Allocation of Scarce Perinatal Pathology Resource with ML-Assisted Placenta Pathology Screening.
Professor Kenton O'Hara
About the author
Kenton is a Professor of Human-Computer Interaction with 25 years’ experience leading user-centred design research and multi-disciplinary research teams in University of Bristol, Microsoft Research, CSIRO, Hewlett-Packard Labs, Rank Xerox EuroPARC and Appliance Studio.
He has PhD in Human-Computer Interaction and extensive experience using qualitative and quantitative research methods to deliver social behavioural insights with proven success translating research to product.
He has a particular interest in the design of digital and AI technologies that facilitate the work of healthcare delivery.
Give us a very short intro to your project
With pregnancy complications such as restricted foetal growth, premature birth, stillbirth, or even maternal death, the placenta will be examined by a specialist pathologist to identify potential causes and support clinical decisions relating to future pregnancies, maternal and neonatal health. However, a nationwide shortage of specialist perinatal pathologists, can result in delays and growing backlogs in placenta analysis. These delays compromise timely learning from adverse pregnancy outcomes and may limit opportunities to improve care pathways for mothers and babies.
To address this critical capacity gap, the team is looking to explore opportunities for AI-based screening of digital placenta pathology images to identify those with abnormalities and prioritise high-risk cases for urgent review by a pathologist. This will reduce workloads, accelerate turnaround times, and ultimately support safer outcomes in future pregnancies and future maternal and neonatal health.
Can you tell us a bit about the team working on your project?
In addressing these issues we aim to take a practice-oriented approach to the design of these AI-based systems. This requires multidisciplinary perspectives that brings together clinical expertise, machine learning researchers and human-computer interaction researchers.
The team have worked closely with the pathologists at Great Ormond Street Hospital to understand their current working practices and key challenges they face and used such understanding to define the machine learning questions we address and the specific ways that we need to support the human-in-the-loop and UX components of the solution.
What progress have you made so far, do you have any findings that you can share with us?
We have completed the initial user research with placenta pathology experts at Great Ormond Street Hospital. This research identified a number of additional ways to triage placentas in the allocation of reporting work as well as identifying opportunities for AI-enabled pre-population of pathology reports.
Regarding the machine learning work, we have done some initial pre-processing of the report data and are currently working on finalising the machine learning pipeline to support the planned machine learning experiments.
What would you say excites you the most about this project?
I love the multidisciplinary, practice-oriented approach to developing machine learning solutions that address the real-world challenges of the clinicians in the context of their everyday workflows. These are not just data problems but complex socio-technical problems. Through solving these problems there is the potential to translate into real world outcomes that will be impactful for pathologists, as well as maternal and child health.
What are the next steps/future plans for your project?
LEAP has given us a great opportunity to make a sensible start on a much larger problem space. I would like to build on this to develop a semi-automated reporting tool for placenta pathology.
Over and above this, by developing such AI capabilities in this domain, there will be opportunities to develop richer diagnostic signals from the placenta (beyond what is currently available). These will enable more tailored stratified interventions to better mitigate the risks in future pregnancies and future maternal and child health that that can better inform clinical interventions in future pregnancies and maternal/child health.
Can you tell us of one recent publication in the world of Digital Health research that has interested you?
I’m enjoying Barbara Duden’s book “Disembodying Women: Perspectives on Pregnancy and the Unborn”. It discusses how modern medicine, imaging and monitoring technologies transform pregnancy from the subjective, embodied experiences of the mother into an externally monitored biomedical condition.
In this respect, women’s sensations and knowledge of their own bodies are displaced by visual and numerical evidence interpreted by experts. Duden suggests that this shift “disembodies” women — situating authority outside the pregnant body and reshaping identity, responsibility, and moral judgement around the foetus rather than the woman.
While this is an old book now, I think the arguments remain relevant for how we understand the role of new AI and imaging technologies in medical contexts.
Keep up to date with the Hub
Sign up to the Hub’s newsletter for updates about the progress of our funded projects and other Hub activity.
Photo courtesy of University of Bristol