A Glasgow firms’s AI software has delivered up to 100 per cent accuracy in detecting childhood sleep apnoea in ground-breaking clinical trials.
Seluna’s platform was used in sleep studies involving 500 children under 18 at NHS Greater Glasgow & Clyde, paving the way for adoption potentially across NHS hospitals.
Clinicians found the software was able to identify severe sleep apnoea with 100 per cent accuracy based on the Apnea-Hypopnea Index (AHI) severity classification, while mild cases achieved 86 per cent and moderate cases 92 per cent.
The technology processed each study in under five minutes and results fell within the normal range of variation seen when trained clinicians review the same data, known as human inter-scorer variability.
Dr Haytham Kubba, paediatric ENT surgeon at the Royal Hospital for Children in Glasgow, said: “Adenotonsillectomy is the most common procedure we perform in children, and sleep apnoea is the most common reason for it.
“Seluna’s technology has the potential to reduce delays, improve consistency in reporting, and allow us to prioritise care for the children who most urgently need intervention.”
Findings from the study, which has been supported by the West of Scotland Innovation Hub, have been independently reviewed by paediatric sleep specialists at Sheffield Children’s NHS Foundation Trust and University Hospital Southampton NHS Foundation Trust.
Dr Scott Black, CEO of Seluna, said: “I’ve heard doctors describe their sleep departments as ‘drowning in data’, with no ability to scale to meet growing demand. What this study proves is that our software works on real NHS data, with all the missing sensors and messy signals that paediatric sleep medicine involves. Our goal now is to replicate this study with hospitals across the UK and get our solution into the hands of clinicians.”
Paediatric sleep apnoea affects up to four per cent of children globally yet remains widely under-diagnosed. The condition can cause serious developmental and behavioural problems if left untreated. Sleep studies produce vast amounts of complex physiological data that doctors and physiologists must interpret manually – a process that can take up to four hours per patient and requires specialist staff who are in short supply.
The cost of not acting
Behind the diagnostic backlog this software is designed to address are the painful stories of real families, who are either stuck on NHS waiting lists, or give up pursuing a diagnosis because the process feels too slow and difficult.
People like Tristan, 41, from Edinburgh, who first suspected his daughter had sleep apnoea when she was around two or three years old.
He said: “Like most first-time parents, you have no idea what’s normal, so you don’t know if something’s wrong. She kept waking up frequently through the night and was getting viral infections quite badly – she was hospitalised three times. She also snores and mouth breathes, which we knew were signs. We did look into getting a sleep study done, but we were put off by the waiting lists and by the idea of putting her into a clinical environment that’s completely different from where she normally sleeps – how much would that actually skew the results? So, we just never went down that route.
“She’s seven now and the question is still unanswered. If the process had been faster and more straightforward, it would have been a much easier decision to make. I’d say to any other parent in the same position: don’t wait and hope it’s just going to disappear. Sometimes it just doesn’t go away.”
Seluna’s software intends to support doctors, allowing them to interpret sleep studies more efficiently, reducing strain on health services and providing faster results to families.
Currently, experts spend up to four hours per patient manually annotating these sleep studies with hopes the software can reduce that to 10 minutes per patient, although that ambition is yet to be tested.
At any rate, the next phase of the clinical validation study has been confirmed with six NHS hospitals across the UK, with the aim of the software being ‘market ready’ by the end of the year. As it is a ‘decision support’ tool, it does not seek to replace human operators, so the hope is for a quicker adoption cycle, pending the outcome of the next phase.
For people like Tristan and his daughter, that day cannot come soon enough.