“Tooth GenAI:” Artificial intelligence integration for periodontal regeneration and dental treatment planning

Abstract

Background:

Artificial intelligence (AI) has shown significant potential in improving diagnostic accuracy and treatment planning in dentistry. However, integrated systems combining image analysis and predictive modeling for periodontal prognosis remain limited.

Materials and Methods:

This study presents Tooth GenAI, an AI-based diagnostic framework integrating convolutional neural networks for caries detection and machine-learning regression models for the prediction of alveolar bone changes. A synthetic dataset of 10,000 patient records, including demographic and risk-factor variables, was generated. Five regression algorithms were evaluated.

Results:

Gradient Boosting Regressor demonstrated the best predictive performance for both 3-month and 6-month bone change prediction. The Convolutional Neural Network-based module achieved high accuracy in detecting carious lesions. The integrated system successfully combined image analysis and predictive modeling.

Conclusion:

Tooth GenAI demonstrates the feasibility of integrating AI-based image detection with predictive analytics to support periodontal diagnosis and treatment planning.

INTRODUCTION

Artificial intelligence (AI) is increasingly being incorporated into health care to improve diagnostic accuracy, clinical decision-making, and treatment outcomes.[1,2] In dentistry, AI-based systems have been used for image interpretation, disease detection, and predictive modeling, thereby enhancing the efficiency and reliability of clinical practice.[3,4] Machine-learning algorithms, deep-learning networks, and natural language processing tools have enabled the development of automated systems capable of analyzing complex clinical data.[5]

Periodontitis is one of the most prevalent chronic inflammatory diseases and is characterized by progressive destruction of the supporting structures of the teeth, including alveolar bone and periodontal ligament.[6,7] Accurate assessment of disease progression and prediction of treatment outcomes are essential for successful periodontal therapy.[8] Conventional diagnostic methods rely on clinical examination and radiographic interpretation, which may be subjective and prone to interobserver variability.[9]

Recent advances in AI have enabled the development of automated systems for detecting dental caries, estimating bone loss, and predicting tooth prognosis.[10,11] However, most currently available systems focus on a single diagnostic task.[12] Integrated platforms combining radiographic analysis with predictive modeling of periodontal bone changes remain limited.

Despite these advancements, most current AI applications in dentistry operate in isolation, focusing on singular tasks such as caries detection or radiographic bone loss estimation.[13] There is a paucity of integrated systems that combine visual diagnostics with predictive modeling for periodontal prognosis, specifically regarding alveolar bone regeneration.[14] Furthermore, few studies have utilized machine-learning regression models to forecast longitudinal bone growth based on patient-specific risk factors, creating a gap in tools available for personalized treatment planning.[15]

Machine-learning regression models can analyze multiple patient-related variables, including age, smoking habits, systemic health, and baseline bone levels, to predict future periodontal changes.[16] Such predictive tools may assist clinicians in selecting appropriate treatment strategies and in estimating regenerative outcomes.[17]

Therefore, the present study aimed to develop and evaluate Tooth GenAI, an integrated AI framework designed to automate caries detection and predict alveolar bone changes using machine-learning algorithms. The hypothesis of the study was that ensemble learning models would demonstrate superior performance compared to conventional linear models in predicting periodontal bone loss and regeneration.

MATERIALS AND METHODS

Study design

This study was designed as an in silico experimental study to develop and evaluate an AI-based diagnostic framework for automated caries detection and prediction of periodontal bone changes. The study involved the development of a synthetic dataset, the implementation of deep-learning and machine-learning algorithms, and the evaluation of model performance using statistical metrics [Figures 1-3].

Figure 1.

Schematic representation of the Tooth GenAI system architecture

Figure 3.

Project implementation approach

Figure 2.

Low-level design

Synthetic dataset generation

Due to the limited availability of large longitudinal periodontal datasets, a synthetic dataset was generated to simulate clinical conditions. A total of 20,000 patient records were created using a generative adversarial network model to reproduce realistic distributions of clinical and demographic variables. The generated dataset included age, gender, smoking habits, alcohol consumption, medical history, baseline bone density, and baseline alveolar bone level.

The output variables were defined as changes in alveolar bone height (mm) at 3- and 6-month intervals and were categorized as bone loss or bone regeneration.

Data preprocessing

The dataset was preprocessed before model training. Categorical variables were converted into a numerical format using one-hot encoding. Continuous variables were standardized using z-score normalization to avoid bias due to differing variable scales.

The dataset was divided into training (80%) and testing (20%) subsets using stratified sampling to maintain equal distribution of outcome variables.

Caries detection model

Figures 4 and 5 shows the layout of this AI based diagnostic framework. Automated caries detection was performed using a convolutional neural network based on the VGG16 architecture. The convolutional base was used to extract features from dental radiographs and intraoral images. The fully connected layers were modified to create a binary classification model for the detection of carious lesions.

Figure 4.

Tooth GenAI main landing page front end part 1

Figure 5.

Tooth GenAI main landing page front end part 2

The model was trained using the Adam optimizer and binary cross-entropy loss function. Early stopping and model checkpointing were used to prevent overfitting. The images of AI based diagnostic framework for caries detection have been depicted in Figures 6-8.

Figure 6.

Cavity detection module front end

Figure 8.

Cavity detection module

Figure 7.

Software detecting the presence of cavity

Prediction of alveolar bone changes

Five machine-learning regression algorithms were evaluated for the prediction of bone loss and bone regeneration:

1. Linear regression

2. Decision tree regression

3. Random forest regression

4. Support vector regression

5. Gradient boosting regression.

Each model was trained separately to predict 3-month and 6-month bone changes. Hyperparameter tuning was performed to optimize model performance. The images of AI based diagnostic framework for bone growth and bone loss pattern detection have been depicted in Figures 9-12. Figures 13 and 14 reflects the appointment scheduling feature of this diagnostic framework.

Figure 9.

Tooth bone growth prediction module

Figure 12.

Tooth bone loss module predicted results

Figure 13.

Schedule appointment feature front end

Figure 14.

Appointment confirmation

Figure 10.

Tooth bone growth/regeneration module predicted results

Figure 11.

Tooth bone loss module front end

Chatbot-based disease prediction module

A natural language processing module was developed using a pretrained Bidirectional Encoder Representations from Transformers (BERT) model. The chatbot analyzed user-reported symptoms and classified them into different dental conditions [Figures 15 and 16].

Figure 15.

Dental disease large language model chatbot

Figure 16.

Dental disease prediction via chatbot

The model was trained using a dataset of symptom-disease pairs and performed multiclass classification to predict possible diagnoses.

Evaluation criteria

Model performance was assessed using the following statistical parameters:

1. Coefficient of determination (R²)

2. Mean squared error

3. Root mean squared error

4. Accuracy, precision, recall, and F1-score for classification models.

Computational environment

All models were implemented using the Python programming language. Deep-learning models were developed using the TensorFlow and Keras libraries, and machine-learning algorithms were implemented using the Scikit-learn library. Data processing was performed using NumPy and Pandas libraries.

System integration

All modules were integrated into a unified framework to allow automated image analysis, prediction of bone changes, and disease classification. The architecture enabled simultaneous processing of image data, clinical variables, and user input to generate diagnostic output.

RESULTS

Caries detection model performance

The convolutional neural network model was evaluated for dental caries detection using a validation set of 2250 images. The model demonstrated high diagnostic performance with an overall accuracy of 94.5%. The sensitivity was 93.8%, indicating a low false-negative rate, while the specificity was 95.1%, indicating accurate identification of healthy teeth. The F1-score was 94.4%, demonstrating balanced performance across classes [Table 1].

Table 1.

Performance metrics for cavity detection (Resnet-50 architecture)

Performance metric	Value (%)	Clinical interpretation
Accuracy	94.5	The model correctly identified the presence or absence of cavities in 94.5% of cases
Precision	95.2	When the model predicted a cavity, it was correct 95.2% of the time (low false positive rate)
Sensitivity (recall)	93.8	The model successfully detected 93.8% of all actual cavities present (low false negative rate)
Specificity	95.1	The model correctly identified 95.1% of healthy teeth as cavity-free
F1-score	94.4	A balanced score indicating the model performs well even with uneven class distributions
AUC-ROC	0.96	Excellent capability to distinguish between carious and noncarious tissues

Prediction of bone growth

The performance of five machine-learning algorithms in predicting alveolar bone growth at 3 months is shown in Table 2. The Gradient Boosting Regressor demonstrated the highest coefficient of determination (R²) and the lowest mean squared error, indicating superior predictive ability compared to linear regression, decision tree, support vector regression, random forest, and k-nearest neighbors models.

Table 2.

The evaluation metrics for 5 different machine-learning/deep-learning algorithms used to predict tooth bone growth over a 3-month period

Model performance for bone growth - 3 months
ML/DL model	R-squared	Mean squared error	Root mean squared error	Graph
Linear regression model	−4.773	0.002	0.049
Decision tree model	0.310	0.000	0.017
Random forest model	0.630	0.000	0.012
Support vector machine model	−0.004	0.000	0.021
K-nearest neighbors model	0.581	0.000	0.013

ML – Machine-learning; DL – Deep learning; R – Correlation co-efficient

For the prediction of bone growth at 6 months, the results are shown in Table 3. The Gradient Boosting model again showed the best performance among all tested algorithms, with higher R² values and lower error rates.

Table 3.

The evaluation metrics for 5 different machine-learning/deep-learning algorithms used to predict tooth bone growth over a 6-month period

Model performance for bone growth - 6 months
ML/DL model	R-squared	Mean squared error	Root mean squared error	Graph
Linear regression model	−1.395	0.002	0.013
Decision tree model	−0.243	0.001	0.033
Random forest model	0.348	0.001	0.024
Support vector machine model	0.065	0.001	0.029
K-nearest neighbors model	0.223	0.001	0.026

ML – Machine-learning; DL – Deep learning; R – Correlation co-efficient

Prediction of bone loss

Table 4 shows the performance of machine-learning models in predicting alveolar bone loss at 3 months. The Gradient Boosting Regressor demonstrated the highest accuracy and lowest prediction error among the evaluated algorithms.

Table 4.

The evaluation metrics for 5 different machine-learning/deep-learning algorithms used to predict tooth bone loss over a 3-month period

Model performance for bone loss - 3 months
ML/DL model	R-squared	Mean squared error	Root mean squared error	Graph
Linear regression model	0.002	0.000	0.020
Decision tree model	0.259	0.000	0.017
Random forest model	0.628	0.000	0.012
Support vector machine model	−0.032	0.000	0.020
K-nearest neighbors model	0.561	0.000	0.013

ML – Machine-learning; DL – Deep learning; R – Correlation co-efficient

The performance for 6-month bone loss prediction is presented in Table 5. The Gradient Boosting model showed the best overall predictive performance, followed by the random forest model.

Table 5.

The evaluation metrics for 5 different machine-learning/deep-learning algorithms used to predict tooth bone loss over a 6-month period

Model performance for bone loss - 6 months
ML/DL model	R-squared	Mean squared error	Root mean squared error	Graph
Linear regression model	−0.002	0.001	0.030
Decision tree model	−0.243	0.001	0.033
Random forest model	0.391	0.001	0.023
Support vector machine model	0.025	0.001	0.029
K-nearest neighbors model	0.256	0.001	0.026

ML – Machine-learning; DL – Deep learning; R – Correlation co-efficient

Chatbot disease prediction performance

The natural language processing module based on the BERT architecture was evaluated for multiclass classification of dental diseases. The model achieved an overall accuracy of 92.1%, with macro precision of 91.5% and recall of 90.8%. The F1-score was 0.91, indicating reliable classification across different disease categories [Table 6]. Recent studies have explored the use of large language models in dental education, clinical decision-making, conversational recommendation systems, and medical knowledge encoding [Table 7].

Table 6.

Evaluation metrics for large language models -based disease prediction

Metric	Value (%)	Clinical implication
Accuracy	92.1	The model correctly identified the dental condition in over 92% of test cases
Macro precision	91.5	The model demonstrated high reliability across all disease categories, rarely misclassifying healthy symptoms as disease
Macro recall	90.8	The model successfully detected the majority of disease cases, minimizing the risk of missed diagnoses
F1-score	0.91	Indicates robust overall performance, balancing the trade-off between precision and sensitivity

Table 7.

Related work using Large Language Models in dentistry

Authors	Year	Techniques	Performance parameters	Work presented
Chang et al.	2023	LLMs evaluation	Model evaluation metrics, such as accuracy, efficiency, fluency, coherence	Evaluation of large language models in dentistry
Thirunavukarasu et al.	2023	Application of LLMs in medicine	Diagnostic accuracy, treatment recommendations, patient outcome	Utilizing LLMs for medical tasks in dentistry
Singhal et al.	2023	Encoding clinical knowledge with LLMs	Encoding accuracy, information retrieval effectiveness	Leveraging LLMs for clinical knowledge in dentistry
Safranek et al.	2023	Role of LLMs in medical education	Educational outcomes, student performance knowledge retention	Integrating LLMs in dental education for better outcomes
He et al.	2023	LLMs as conversational recommenders	Recommendation accuracy, user satisfaction and engagement	Implementing LLMs for patient consultation in dentistry

LLMs – Large language models

Overall system performance

The integrated AI framework successfully combined image analysis, predictive modeling, and symptom-based classification into a single diagnostic system. The results demonstrate that ensemble machine-learning algorithms, particularly Gradient Boosting, provided the most accurate predictions of periodontal bone changes. At the same time, the convolutional neural network showed high accuracy in detecting dental caries.

DISCUSSION

AI-based systems have been developed for automatic detection and classification of dental restorations in panoramic radiography,[18,19] as well as for caries detection and classification using deep-learning techniques.[20,21] Moreover, AI-driven approaches have been explored for automated caries detection with smartphone photography.[22] Multimodal large language models, exemplified by ChatGPT, are being explored for their applicability in various aspects of dentistry.[23]

Enhanced masked image modeling techniques have been proposed for the analysis of dental panoramic radiographs, contributing to improved diagnostics.[23,24] A comprehensive review has highlighted recent AI-driven advances in dentistry e-health, underscoring the transformative impact of AI technologies in this field.[25] Deep-learning-based registration methods for two-dimensional dental images, coupled with edge-specific loss functions, offer enhanced accuracy in dental imaging applications.[26,27] Moreover, the evolution of digital dentistry, encompassing past, present, and future trends, has been elucidated, reflecting the paradigm shift toward digital technologies in dental care.[28]

Outside the realm of dentistry, machine-learning approaches have been employed for various dental-related studies, including three-dimensional modeling of fatigue crack growth in gear teeth,[29] dynamic tooth wear prediction in gear sets,[30] automatic age and sex determination from dental images,[31] and prediction models for tooth loss.[32,33] Recent studies have focused on developing predictive models to forecast tooth loss in periodontal patients, utilizing staging and grading systems to construct nomograms for long-term prognosis.[33] Updated systematic reviews have identified predictors of tooth loss during periodontal maintenance, shedding light on factors influencing disease progression and treatment outcomes.[34] Epidemiological investigations have projected trends in tooth loss, revealing demographic shifts and predictive factors that contribute to edentulism in adults and seniors.[35] Moreover, comparative analyses have evaluated the efficacy of periodontal prognostic systems in predicting tooth loss, informing clinical decision-making and treatment planning.[34] Outside periodontics, advancements in deep learning have enabled automated detection of underground cavities using ground-penetrating radar systems,[36] and the development of algorithms for the detection of oral cavity squamous cell carcinoma from photographic images, enhancing diagnostic accuracy.[37] Reviews have also discussed the historical evolution, current applications, and prospects of AI in dentistry, emphasizing its transformative potential in patient care and education.[38] Furthermore, innovative approaches have been explored, such as utilizing virtual patients via AI chatbots to enhance dental students’ diagnostic skills[39] and integrating chatbot systems into information systems for private dental clinics to improve patient communication and service delivery.[40]

The present study evaluated the feasibility of an integrated AI-based framework for automated dental diagnosis and prediction of periodontal bone changes. The results demonstrated that machine-learning algorithms, particularly ensemble models, can effectively predict nonlinear changes in alveolar bone levels and may assist clinicians in treatment planning.

Among the evaluated regression models, the Gradient Boosting Regressor showed the highest predictive accuracy for both 3-month and 6-month bone change prediction. Periodontal disease progression is influenced by multiple interacting variables such as age, smoking habits, systemic health, and baseline bone level. These relationships are often nonlinear, which may explain the inferior performance of linear regression models. Ensemble learning methods combine multiple decision trees and are therefore better suited to model complex biological behavior.

The convolutional neural network used for caries detection demonstrated high sensitivity and specificity.[12] In clinical practice, high sensitivity is important to avoid the missed diagnosis of carious lesions. In contrast, high specificity reduces unnecessary treatment.[13] These findings support previous studies, which reported that deep-learning models can achieve diagnostic accuracy comparable to that of experienced clinicians.[14]

The chatbot-based natural language processing module also showed good performance in classifying dental diseases. Such tools may help in preliminary screening, patient education, and clinical decision support. However, these systems should be considered supportive tools, not a replacement for clinical examination.

An important feature of the present study was the integration of multiple AI modules into a single framework. Most previous studies have focused on individual tasks such as bone loss detection or caries identification. The present model combines image analysis, predictive modeling, and symptom-based classification, thereby providing a more comprehensive diagnostic approach.

Despite promising results, certain limitations must be considered. The dataset used in this study was synthetically generated due to the lack of publicly available longitudinal periodontal datasets. Although synthetic data allow controlled modeling of clinical variables, they may not fully represent real clinical variability. Therefore, the predictive accuracy observed in this study may be higher than that obtained in real-world settings.

Another limitation is the absence of validation using clinical patient data. Future studies should evaluate the model’s performance on real radiographic and clinical datasets and include longitudinal follow-up to confirm predictive accuracy.

Further research is also required to integrate such systems into clinical workflow and electronic health record platforms. With proper validation, AI-based prediction models may improve diagnosis, reduce operator variability, and support personalized periodontal treatment planning.

CONCLUSION

The present study demonstrates the potential of an integrated AI-based framework for automated dental diagnosis and prediction of periodontal bone changes. Among the evaluated machine-learning models, the Gradient Boosting Regressor showed the highest accuracy in predicting both bone loss and bone regeneration at different time intervals, indicating its suitability for modeling complex biological processes.

The convolutional neural network achieved high diagnostic accuracy in detecting dental caries, suggesting that deep-learning-based image analysis can serve as a reliable adjunct to clinical examination. In addition, the natural language processing module demonstrated strong performance in classifying dental diseases, indicating the usefulness of chatbot-based systems for preliminary screening and patient guidance.

Although the results are promising, the use of synthetic data represents a limitation, and further validation using real clinical datasets is required. Future studies should focus on clinical implementation and evaluation of AI-based tools in periodontal diagnosis and treatment planning.

AI-assisted systems may help improve diagnostic precision, reduce operator variability, and support personalized treatment strategies in periodontology.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.