Development of a AI-driven caries risk assessment and preventive personalized treatment planning platform based on panoramic radiographs and clinical data.

Project Abstract

This study presents the development and validation of an AI-driven platform for caries risk assessment and preventive personalized treatment planning, integrating panoramic radiographs with comprehensive clinical data to support evidence-based decision-making in dentistry. The platform architecture combines a robust image analysis module, a clinical data fusion engine, and an advanced optimization module for personalized treatment recommendations. A multi-stage approach was employed, starting with data collection from a diverse cohort of patients across multiple clinics to ensure heterogeneity in demographics, caries experience, oral health behaviors, and radiographic presentation. Preprocessing steps included standardization of panoramic radiographs, artifact reduction, and normalization of clinical variables such as age, sex, dietary habits, fluoride exposure, salivary flow, oral hygiene index, previous caries history, and socio-economic factors. The image analysis component leverages a deep convolutional neural network trained to detect incipient and overt carious lesions, quantify lesion depth, and assess radiographic features correlated with caries risk, such as enamel hypoplasia, restoration density, and adjacent dentition status. The clinical data fusion engine integrates image-derived features with temporal clinical indicators to produce a comprehensive risk profile, employing ensemble learning techniques to optimize predictive performance. The risk assessment model outputs probabilistic risk scores for future caries development over defined horizons (1-3 years) and identifies the most impactful risk factors for each patient. The treatment planning module utilizes a decision-support framework that translates risk stratification into personalized preventive strategies and restorative interventions. It incorporates evidence-based guidelines, patient preferences, cost considerations, and resource availability to generate an actionable treatment plan with prioritized steps, schedule recommendations, and patient-specific preventive measures such as fluoride interventions, sealants, professional cleaning, remineralization protocols, and behavioral modifications. The platform also features an explainable AI component that highlights the contribution of imaging and clinical variables to the risk score, supporting clinicians in shared decision-making with patients. A rigorous validation strategy included internal cross-validation and external validation across independent datasets, with metrics including area under the receiver operating characteristic curve (AUC), calibration plots, F1-score for lesion detection, and decision-curve analysis to evaluate clinical utility. The results demonstrated superior predictive accuracy for caries incidence compared with baseline models that relied solely on radiographs or clinical data, and the treatment planner showed improved alignment with guideline-concordant care and patient adherence in simulated scenarios. Usability testing with dental professionals indicated favorable acceptance, intuitive workflow integration, and potential reductions in unnecessary interventions. Ethical considerations addressed data privacy, informed consent, and bias minimization across population subgroups. The platform is designed for interoperability with existing electronic health record systems and imaging modalities, enabling scalable deployment in general practice, community health centers, and dental schools. Limitations acknowledged include dependence on data quality, the need for ongoing model retraining to accommodate evolving epidemiology, and the requirement for prospective longitudinal studies to assess long-term clinical outcomes. The study contributes a validated, explainable, AI-assisted framework for caries risk stratification and personalized preventive planning, aiming to enhance early detection, tailor interventions, and improve patient-specific oral health trajectories.

Project Overview

What This Project Is About

---

A straightforward study to create a tool that uses X-ray pictures of teeth (panoramic radiographs) and basic patient information to predict who is at higher risk of developing cavities (caries) and to help plan personalized prevention and treatment steps.

---

---

The Problem It Addresses

---

Caries are common and can be prevented, but many people do not get timely, tailored advice. Dentists often rely on experience and simple checks. This project aims to combine simple data with image analysis to improve risk prediction and treatment planning for individual patients.

---

---

Objectives of the Project

---

1. Develop a user-friendly risk score using dental images and clinical data.
2. Integrate a decision-support module that suggests personalized preventive steps.
3. Test the tool on sample patient data to check accuracy.
4. Assess how the tool could fit into routine dental visits.

---

---

What You Will Do Step by Step

---

1. Review basic literature on caries risk factors and image analysis.
2. Collect de-identified panoramic radiographs and clinical data.
3. Extract simple features from images and combine them with clinical information.
4. Build a basic predictive model to estimate caries risk.
5. Develop a user-friendly interface for clinicians.
6. Evaluate the model’s predictions against known outcomes.
7. Refine the model based on feedback from mentors.

---

---

Expected Outcome

An accessible, tested tool that provides caries risk scores and personalized prevention plans, with clear notes for clinicians on how to apply recommendations in patient care.