TL;DR
An app that reads facial expression, speech and movement while users complete short tasks has flagged around a third of participants in a Nottingham trial of pregnant and postnatal women. The study has run since 2024 and finishes in October. Its clinicians are explicit that the tool screens rather than diagnoses.
How the screening works
Participants are asked to do something ordinary — read a few pages of a novel, write down how they feel — while software watches how they do it. Expression, speech patterns and involuntary movement are scored, and the app raises a flag once the result crosses a depression threshold. The build is a three-way effort between the University of Nottingham, Nottinghamshire Healthcare (the local NHS foundation trust) and an AI firm, BlueSky.
Dr Neil Nixon, one of the lead researchers, framed the purpose around timing rather than cleverness: “Being able to identify clinical problems early on, at a point where we can intervene for those women, could really make a big difference.” His concern is the women who would never raise it themselves — “The worry is that women don’t realise their mood is changing.”
The stakes are set by an Oxford finding published last year: suicide accounts for 39% of UK maternal deaths in the window running from six weeks to a year after birth.
The caveat the charity insisted on
Sally Bunkman of PANDAS, which supports women with perinatal mental illness, gave the technology qualified backing and then drew a hard line. Women using its services are “often dismissed” and told they have the baby blues, she said, so anything that legitimises the conversation with a GP has obvious value. But: “we don’t want women being supported by an AI. We need human to human support.”
Lizzie, a mother from Long Eaton who used the app when her youngest was seven months old, described the gap it targets — the point where hospital support stops and you are simply at home. “It’s not all rainbows and butterflies.”
Looking forward
Nixon is clear the software will not diagnose, only support clinicians. That distinction is doing heavy lifting, and UK deployments keep landing on it: the Dundee surgery running a robot through health checks, the Yorkshire GP receptionist that could not handle local accents. Screening tools move where a clinical decision starts, not whether one gets made.
The unresolved question is what happens to a flag. A third of participants is a substantial referral load for perinatal mental health services already under strain, and October is when that arithmetic stops being hypothetical.