• Research
• Expressions Spring 2026

AI at the Tooth Level

HKU-Led Team Cracks the Code of Early Childhood Decay

The research team, led by Professor Shi Huang (left). The AI system achieved a 93% accuracy rate in predicting caries that would develop two months later.

Early childhood caries (ECC) is the most common chronic disease of childhood, affecting more young children than asthma or obesity. Yet one basic puzzle has remained unanswered - why do some tooth decay and cavities develop rapidly while other teeth, in the very same mouth, stay relatively healthy?

A collaborative team led by the Faculty of Dentistry at The University of Hong Kong (HKU) has now brought that mystery into sharp focus. Working with partners from the Chinese Academy of Sciences Qingdao Institute of Bioenergy and Bioprocess Technology (CAS-QIBEBT), Qingdao Stomatological Hospital, and Qingdao Women and Children’s Hospital, they have developed the world’s first artificial intelligence (AI) system that can predict cavity risk for individual teeth based on their microbial fingerprints.

The system, called Spatial-MiC, reaches more than 90% accuracy and, in some cases, can foresee decay two months before it becomes visible. The study, published in the prestigious journal - Cell Host & Microbe, marks a major step towards truly “tooth-specific” precision dentistry for children.

The research team was led by Professor Shi Huang, Assistant Professor in Microbiology in the Division of Applied Oral Sciences and Community Dental Care at HKU Dentistry. The core group included PhD student Yufeng Zhang, Professor Jian Xu from CAS-QIBEBT, Dr Fei Teng from Qingdao Stomatological Hospital, and Dr Fang Yang from Qingdao Women and Children’s Hospital.

To understand why ECC targets particular teeth, the team undertook the most comprehensive analysis to date of tooth-specific microbial communities in preschoolers. They followed 89 children aged 3 to 5 over nearly a year, collecting 2,504 plaque samples from individual teeth. Using cutting-edge 16S rRNA sequencing alongside shotgun metagenomics, they examined not only which microbes were present, but also what they were capable of doing.

A striking pattern emerged in healthy mouths: a natural anterior-to-posterior microbial gradient. Front teeth (incisors) tended to host one type of bacterial community, while back teeth (molars) hosted another, forming a consistent spatial arrangement shaped by saliva flow, chewing forces and tooth anatomy.

In children who developed cavities, this orderly pattern began to break down before any visible holes appeared. Microbes normally associated with incisors started to appear on molars, and molar-associated microbes migrated forward. These subtle but consistent shifts signalled that specific teeth were moving towards disease.

Spatial-MiC was trained to read these patterns. By analysing the microbial community on each tooth, together with information from neighbouring teeth, the AI system achieved 98% accuracy in detecting existing cavities and 93% accuracy in predicting which teeth would develop cavities within two months. This represents a major advance over current approaches, which largely rely on whole-mouth assessments and often miss early, localised warning signs.

HKU Dentistry developed Spatial-MiC, the world’s first AI system for early childhood caries detection at the single-tooth resolution.

“These findings fundamentally change how we understand tooth decay,” said Professor Huang. “We have moved from seeing cavities as inevitable to being able to predict and prevent them at the microbial level, tooth by tooth.”

The implications are especially significant for countries like China, where ECC affects more than 70% of five-year-olds, and for global child health more broadly. Although clinicians have long observed that some teeth are more prone to decay, preventive strategies still tend to treat all teeth alike. The new research opens the door to targeted protection for high-risk teeth before irreversible damage, pain and infection set in.

The team now plans to validate the Spatial-MiC approach in more diverse populations and to explore how it might be translated into practical tools for clinics. Their vision is that, one day, a simple plaque sample could help dentists identify which teeth in a child’s mouth need the most intensive preventive care.

“This isn’t just about better dental care,” noted Dr Yang, the first author. “It’s about giving children healthier starts in life by preventing pain, infections, and the developmental impacts of severe tooth decay in a more precise manner.”