Can AI spot tooth decay in toddlers? Singapore is finding out
Tooth decay in a two-year-old is easy to miss. The child can’t tell you which tooth feels odd, nobody is peering at the back of their mouth, and dental problems tend to get dealt with only once they show up. By then, as Singapore’s National University Hospital puts it, “treatment can be more complex.”
That is the problem a small programme in western Singapore has been chipping away at since January 2023. It is worth a parent’s attention for two reasons. The early results are good. And the next step is AI, which makes it one of the more honest tests of what AI can actually do for a very young child’s health.
The gap is bigger than most parents think
NUHS says around 48 per cent of Singapore preschoolers aged 1.5 to four have tooth decay, and that 90 per cent of the affected children go untreated. One study cited by NUH found prevalence rising from 17.8 per cent among two-year-olds to 42.9 per cent by age three. So the jump happens in a single year, at an age when most children have never sat in a dentist’s chair.
Access is uneven too. A 2018/19 NUHS study found that among preschoolers from lower-income families, “fewer than half had ever visited a dentist, compared with 75.4 per cent of their peers from higher-income families.”
How the screening works today
The programme sits inside HEADS-UPP, a community health initiative for lower-income families run with Care Corner Singapore and PCF Sparkletots. The dental part is run by NUH, NUS Dentistry and the National University Centre for Oral Health, Singapore (NUCOHS).
The steps are simple, which is the point:
- NUH nurses and case management officers visit the preschool and photograph each child’s teeth. NUHS says they take five intraoral views.
- Parents fill in a questionnaire adapted from the American Academy of Pediatric Dentistry’s caries risk assessment tool.
- A paediatric dentist from NUS Dentistry reviews the photos and answers remotely, then writes a personalised report with a risk level and what to do first.
- Higher-risk children are referred to NUCOHS. Lower-risk children go to primary care. HEADS-UPP staff walk parents through the report.
- Six to nine months later, a follow-up questionnaire checks what changed.
As of 20 January 2026, 355 children aged 18 months to six years from 17 preschools had been through it.
The result that matters is not the diagnosis
Here is the number we would underline. Among children found to have decay, 57.5 per cent got specialist follow-up within four to six months. In the earlier 2018/19 NUHS study, that figure was 13.3 per cent. Regular primary-care dental visits rose too, from 28.9 to 51.3 per cent. And the share of children rated moderate to high risk fell from 93.3 per cent at the start to 75.6 per cent at follow-up.
These are preliminary figures from the programme’s own team, not a randomised trial, so treat them as encouraging rather than settled. But the shape of the gain is telling. Spotting the decay was never really the hard part. Getting a busy family from “your child has a cavity” to an actual appointment was. The photo and the plain-language report seem to close that gap.
So where does AI come in?
Not yet, and we think that is the honest part of this story. NUH says the next phase aims to “enhance the dental screening process through the integration of artificial intelligence (AI)”: AI would help review the photos, read the questionnaire answers and draft the oral health reports. The stated goal is to “reach more children and expand to additional preschools” while cutting manual work.
In other words, the AI’s job would be to stretch one paediatric dentist’s eyes across many more preschools. It would not replace the nurse who visits, or the staff member who sits with a parent and explains the report. That is the right job to give it.
Can it do that job? The research is promising, with real caveats. A 2024 systematic review in Oral Diseases looked at 19 studies of AI finding caries in photographs. Accuracy (measured as an F1 score) ranged from 68.3 to 95.4 per cent with professional cameras, but only 42.8 to 80 per cent with smartphone cameras. The authors called for “more robust study designs” before drawing firm conclusions.
A 2026 meta-analysis focused on children, in the journal Children, was warmer: one photo-based study of early childhood caries reported 91 per cent sensitivity and 97 per cent specificity. Even so, the authors wrote that “AI is not yet ready for routine standalone use,” and described it as an adjunct that works under a clinician’s supervision. They also warned that most of the data came from single centres, which can make results look better than they will in the wild.
So the evidence supports exactly the setup Singapore is planning: trained staff take good photos, AI sorts and drafts, a dentist signs off.
What to take from this tonight
Two things, one practical and one about AI generally.
The practical one: if your child’s preschool offers a dental screening, say yes, and treat the referral as the important part. The programme’s own numbers suggest the screening only helps when the follow-up visit actually happens.
The general one is a useful test for any AI that touches your child’s health. Ask who takes the input, who checks the output, and who picks up the phone afterwards. A consumer app reading a blurry phone photo at bedtime answers none of those, and the smartphone figures above are the weakest in the review. A programme where AI speeds up a dentist, inside a system that books the appointment, answers all three. That is what AI genuinely helping a child looks like: less a clever diagnosis, more a few hundred more toddlers seen before anything hurts.
Was this useful?
Sources
- New tele-dentistry programme supports early intervention for children at risk of caries · National University Hospital (NUH)
- Tele-dentistry pushes dental care earlier for children · NUHS+ (National University Health System)
- Detecting dental caries on oral photographs using artificial intelligence: A systematic review · Oral Diseases (Wiley), via PubMed
- Artificial Intelligence in Pediatric Dentistry: A Systematic Review and Meta-Analysis · Children (MDPI), via PubMed Central