AI-Based Caries and Plaque Detection and Caries Risk Prediction in Children

Dental caries is a multifactorial disease, and accurate assessment of both current disease status and future caries risk is important for planning individualized preventive care. Conventional caries risk assessment relies on the combined evaluation of clinical findings, oral hygiene, dietary habits, fluoride exposure, previous caries experience, and other risk or protective factors. However, this process may vary between clinicians and may be influenced by subjective interpretation. The present study is designed to investigate whether artificial intelligence (AI)-assisted analysis of digital dental images can contribute to a more objective, standardized, and reproducible approach to caries and dental plaque assessment in children.

Clinical dental examinations will be performed using standardized criteria. Dental caries will be recorded according to ICDAS II criteria, caries experience will be documented using DMFT/dft indices, and dental plaque will be assessed using a standardized plaque index. In addition, standardized intraoral photographs and three-dimensional intraoral digital scans will be obtained. Information related to oral health behaviors, including toothbrushing frequency, sugar consumption, fluoride exposure, and dental attendance, will be collected through a structured parental questionnaire.

The digital data will be used to develop an AI-based image analysis system for the automatic detection and segmentation of dental caries and plaque. An instance-segmentation approach is planned so that both localized carious lesions and irregularly distributed plaque areas can be identified at the pixel level rather than only by bounding boxes. The model will be developed using transfer learning from pretrained weights and will be retrained using the intraoral image and scan dataset generated within the study.

Reference annotations for model development will be created manually according to predefined clinical criteria. The annotated dataset will be divided into training, validation, and test subsets. Model performance will be evaluated using detection and segmentation metrics such as precision, recall, F1-score, mean average precision, Dice coefficient, and Intersection over Union. Additional validation procedures, including cross-validation and comparison of different convolutional approaches, are planned to improve model robustness and generalizability. Visualization methods such as Grad-CAM will also be used to improve the clinical interpretability of model decisions.

After model training and testing, the optimized system will be integrated into a clinical decision-support interface. For newly evaluated participants, intraoral scan data, with photographs when necessary, will be analyzed automatically. The system will provide outputs including the number, location, and severity of detected carious lesions and the estimated extent of dental plaque. These AI-derived findings will be combined with clinical indices and behavioral information to generate an individualized caries risk assessment.

Participants will be classified into predefined caries-risk categories using a combination of clinical disease indicators, plaque findings, AI-derived lesion and plaque measurements, and behavioral risk and protective factors. The resulting risk classification will be used to support individualized preventive recommendations. Repeated clinical and digital evaluations will allow changes in caries findings, plaque levels, and risk status to be assessed over time.

The study is intended to determine whether combining AI-based analysis of intraoral digital data with established clinical and behavioral indicators can improve the consistency and clinical usefulness of caries risk assessment in pediatric dentistry. The proposed system is intended as a clinical decision-support tool rather than a replacement for professional dental examination and judgment.

Children classified as low caries risk, with no disease indicators, no or minimal risk factors, and predominant protective factors. This group includes children with no AI-detected lesion, plaque percentage \<10% or plaque index 0, twice-daily brushing with fluoride toothpaste, low sugar consumption, and regular dental attendance. Preventive recommendations will include recall every 6 months, brushing twice daily with fluoride toothpaste, dietary counseling, and fissure sealants when indicated.

Behavioral: individualized preventive recommendations

Children without disease indicators but with at least one risk factor, including a single AI-detected non-cavitated lesion, plaque percentage of 10-29% or plaque index 1-2, increased daily sugar consumption, brushing less than twice daily, irregular dental attendance, or low fluoride exposure. Preventive recommendations will include recall every 3-6 months, oral hygiene instruction, fluoride support when needed, brushing twice daily with fluoride toothpaste, professional fluoride varnish every 6 months, dietary counseling, and fissure sealants.

Behavioral: individualized preventive recommendations

Children classified as high caries risk based on the presence of at least one disease indicator or increased disease burden, including ICDAS ≥3, DMFT/dft \>0, restoration or tooth loss due to caries within the previous 12 months, ≥2 AI-detected lesions, AI-detected lesion with plaque percentage ≥30%, or plaque index ≥2. Preventive recommendations will include recall every 1-3 months, professional plaque control, recommendation of a remineralizing agent (Tooth Mousse or R.O.C.S. Medical Minerals gel), oral hygiene instruction, brushing twice daily with fluoride toothpaste, professional fluoride varnish every 3 months, dietary counseling, and fissure sealants.

Behavioral: individualized preventive recommendations

Participants will be classified into predefined caries-risk categories using a combination of clinical disease indicators, plaque findings, AI-derived lesion and plaque measurements, and behavioral risk and protective factors. Based on the assigned caries risk level, individualized preventive recommendations such as oral hygiene instruction, dietary counseling, fluoride applications, professional plaque control, and follow-up visits will be provided. Participants will be followed according to their risk level, and digital scans and clinical assessments will be repeated to evaluate changes in caries, plaque, and caries risk over time.

Time frame: Baseline

Accuracy will be calculated as the proportion of correct predictions among all evaluated predictions, using expert-annotated reference data. Model predictions will be compared with manually annotated reference data for dental caries. Accuracy will be reported as a single percentage value (%).

Time frame: Baseline

Sensitivity will be calculated as the proportion of correctly identified positive findings among all positive findings in the expert-annotated reference data. Model predictions will be compared with manually annotated reference data for dental caries. Sensitivity will be reported as a single percentage value (%).

Time frame: Baseline

Specificity will be calculated as the proportion of correctly identified negative findings among all negative findings in the expert-annotated reference data. Model predictions will be compared with manually annotated reference data for dental caries. Sensitivity will be reported as a single percentage value (%).

Time frame: Baselina

Precision will be calculated as the proportion of correct positive predictions among all positive predictions made by the artificial intelligence model. Model predictions will be compared with manually annotated reference data for dental caries. Precision will be reported as a single percentage value (%).

Time frame: Baseline

The F1-score will be calculated as the harmonic mean of precision and recall. Values range from 0 to 1, with higher values indicating better detection performance. Model predictions will be compared with manually annotated reference data for dental caries. The F1-score will be reported as a single percentage value (%).

Time frame: Baseline

Intersection over Union (IoU) will be calculated as the ratio of the overlapping area to the combined area of the predicted and expert-annotated segmentation masks. Values range from 0 to 1, with higher values indicating better segmentation performance. Model predictions will be compared with manually annotated reference data for dental caries.

Time frame: Baseline

Accuracy will be calculated as the proportion of correct predictions among all evaluated predictions, using expert-annotated reference data. Model predictions will be compared with manually annotated reference data for dental plaque.

Time frame: Baseline

Sensitivity will be calculated as the proportion of correctly identified positive findings among all positive findings in the expert-annotated reference data. Model predictions will be compared with manually annotated reference data for dental plaque.

Time frame: Baseline

Specificity will be calculated as the proportion of correctly identified negative findings among all negative findings in the expert-annotated reference data. Model predictions will be compared with manually annotated reference data for dental plaque.

Time frame: Baseline

Precision will be calculated as the proportion of correct positive predictions among all positive predictions made by the artificial intelligence model. Model predictions will be compared with manually annotated reference data for dental plaque.

Time frame: Baseline

The F1-score will be calculated as the harmonic mean of precision and recall. Values range from 0 to 1, with higher values indicating better detection performance. Model predictions will be compared with manually annotated reference data for dental plaque.

Time frame: Baseline

Intersection over Union (IoU) will be calculated as the ratio of the overlapping area to the combined area of the predicted and expert-annotated segmentation masks. Values range from 0 to 1, with higher values indicating better segmentation performance. Model predictions will be compared with manually annotated reference data for dental plaque.

Time frame: At baseline assessment.

Participants will be classified into low, moderate, or high caries risk categories according to predefined AI-integrated criteria, combining AI-derived dental caries and plaque findings with clinical indices and behavioral risk factors. The outcome will be reported as the number of participants in each risk category.

Time frame: Baseline and up to 6 months

The number of carious lesions will be assessed at baseline and follow-up using clinical examinations and digital intraoral images. The change will be calculated by subtracting the baseline number of carious lesions from the follow-up number for each participant. Follow-up assessments will be performed according to the participant's caries risk category.

Time frame: Baseline and up to 6 months

The change in the percentage of tooth surface area covered by dental plaque from baseline to follow-up will be evaluated using digital intraoral images and the artificial intelligence model. Follow-up assessments will be performed according to the participant's caries risk category.

Time frame: Baseline and up to 6 months

Changes in caries risk classification from baseline to follow-up will be evaluated using the predefined low, moderate, and high caries risk categories. The outcome will be reported as the number of participants whose risk category changes during follow-up.

Detection of Dental Caries and Plaque Using AI-Based Imaging and Prediction of Caries Risk in Children