A baby’s heartbeat is a barely perceptible signal that is easily lost in the biological cacophony of the womb. But finding and monitoring that tiny pulse is essential to tracking foetal health, detecting potential problems early, and acting quickly to save one – and possibly two – lives.
For those with high-risk pregnancies, this monitoring often means exhausting hospital trips several times a week in the later months. Juggling family, work and medical care becomes almost impossible.
This is one of a growing number of scenarios where artificial intelligence (AI) is improving the lives of patients.
At the University of Melbourne start-up, Kali Healthcare, electrical engineer Dr Emerson Keenan and his colleagues are developing an AI-based system for ongoing remote foetal monitoring. It is designed to help pick up early warning signs of problems and could potentially be used to predict when a woman is likely to go into labour.
“So rather than having people go in for intense monitoring procedures and be in hospital for long durations, we can have it happen in the home setting and get the same clinical accuracy.”
The technology is not only investing in prevention but also improving health equity by improving the accessibility of high-quality care.
Dr Keenan’s interest in medical monitoring technology started during his honours year when he worked on a smartphone app to diagnose sleep apnoea based on breathing sounds. Soon after, he learned that foetal heartbeat monitoring technology had barely changed in decades – it still relied on an ultrasound probe being moved across the abdomen to track the heartbeat.
Joining a University of Melbourne research group, Dr Keenan began a PhD exploring whether foetal monitoring could be achieved using electrical sensors, similar to those used to monitor electrical activity in the heart or brain.
The challenge was to distinguish the relatively faint foetal heartbeat amid the background noise, which offered a unique opportunity for AI.
"We spent a lot of time working out exactly where the sensors would need to be so that we could accurately pick up the foetal signals, as well as then developing algorithms that could interpret those signals to give us that foetal heart rate," explains Dr Keenan.
Kali Healthcare uses neural networks: systems that ‘learn’ from existing data and apply those learnings to new data sets, much like the human brain.
The aim is to provide hospital-in-the-home care for pregnancies at risk of complications.
The first step is monitoring the foetal and maternal heart rate and uterine contractions. However, Dr Keenan also hopes the device could be used to predict the likelihood that a woman will go into labour in the next seven days.
Currently in clinical trials, the device must show agreement with existing clinical standards to satisfy medical device regulators such as the US Food and Drug Administration. Early results suggest it will exceed the required benchmarks – an exciting future prospect.
Proving accuracy is a major hurdle for AI-based medical technologies. At the Mass General Brigham and Harvard Medical School in Boston, Assistant Professor James Hillis (BMedSc 2007, MBBS 2009) is on the frontline of navigating that challenge.
After graduating from the University of Melbourne Medical School, Assistant Professor Hillis saw the impact that artificial intelligence was starting to have in many parts of society, which sparked his interest in the field.
In Boston, he joined a team developing AI algorithms for stroke-related applications and started to think about the evidence needed to apply for FDA authorisation. It was a steep learning curve, and Professor Hillis came to appreciate the unique skill set involved in assessing such algorithms.
“Our team very quickly realised there was a lot of interest from companies throughout the world – especially companies outside the US who wanted to get into the US market – to have a group that was dedicated to the evaluation of AI algorithms,” Assistant Professor Hillis says. And he was part of a team that pivoted to provide the service.
Since 2020, Assistant Professor Hillis and his colleagues have worked on studies that have led to the FDA authorisation of 17 different AI-based devices. This includes one that flags brain scans with possible haemorrhage for more urgent review to another that interprets cardiac measurement on point of care ultrasound.
“With models that help to prioritise radiology scans, you've got the AI model working to do that, but you still have a human radiologist or neurologist or another clinician saying, ‘Okay, I agree with the AI, let's move it to the next step’,” says Assistant Professor Hillis.
AI’s greatest strength is its ability to parse huge data sets and find subtle signals or associations that human eyes and minds might miss.
“It's not possible for a human to compute everything that's going on in today’s clinical environment,” says Assistant Professor Hillis. “There is so much data that are generated as part of healthcare, with the potential for myriad insights that could benefit care.”
