Mobile Imaging-Based Machine Learning for Dental Caries, Sealants, and Fluorosis: Protocol for a Cross-Sectional...

Assessing dental caries, sealants, and fluorosis is essential for public health surveillance, providing critical data to evaluate national prevention programs. Standard methods performed by dental professionals are often limited by affordability, accessibility, and scalability for both population-level and individualized assessments. Mobile health (mHealth) approaches to concurrently detect caries, sealants, and fluorosis have remained largely unexplored, especially at the population level.

This study leverages mHealth technologies that integrate computer vision using machine learning and deep learning with images captured by smartphone cameras and low-cost intraoral cameras. The primary objective is to develop and validate models for detecting caries lesions, identifying sealants, and quantifying fluorosis severity from standardized dental images, using standardized visual clinical examinations as the reference standard.

The proposed study population will include approximately 1000 adolescents in Colorado, United States, living in communities with naturally elevated fluoride levels in the public water system. Participants will undergo standardized clinical dental examinations and imaging using intraoral cameras and smartphones. Supervised learning models will incorporate reference chart-based color correction, radiomic spatial and textural features, and neural network classifiers. The reference standard will be standardized visual clinical examinations performed by trained and calibrated dental professionals. Two models will be developed and evaluated: one to detect caries lesions and sealants and another to assess fluorosis severity. Model performance will be evaluated against clinical assessments by dental professionals using stratified cross-validation and multiclass performance metrics while minimizing bias and accounting for confounders common to human examiners.

A standardized dental examination, an intraoral imaging protocol, and a smartphone imaging protocol are used to assess all 8 permanent molars for caries and sealants, as well as the 6 upper anterior teeth for fluorosis severity. Pilot studies were conducted to test study logistics and calibrate 3 examiners in person, supplemented by debriefings, mobile app training, and a web-based calibration module. The study was funded in September 2022 with supplemental funding awarded in June 2024. The study launched in May 2024, and as of January 2026, data have been collected from approximately 300 participants.

The integration of computer vision and mobile device imaging will enable affordable, scalable, population-level assessments for detecting caries and sealants and quantifying fluorosis severity among adolescents. mHealth technologies have been increasingly incorporated into dentistry for both clinical decision support and at-home use. This protocol will further help establish a structured methodological framework for acquiring, processing, and analyzing mobile imaging data for dental health surveillance and epidemiological studies.

DERR1-10.2196/91239.