1. Diagnosis and caries detection
   AI-assisted radiograph interpretation, early caries detection, periapical lesion identification, crack/fracture detection, and pulp-status prediction.
2. Treatment planning in conservative dentistry
   Caries-risk prediction, restoration selection, minimally invasive treatment planning, and prognosis prediction.
3. Restorative procedures
   AI support for shade matching, cavity design, restoration quality assessment, marginal-gap or recurrent-caries detection, and CAD/CAM workflows.
4. Endodontic diagnosis and imaging
   Identification of root-canal anatomy, canal number and curvature, calcifications, resorption, missed canals, and CBCT interpretation.
5. Endodontic treatment and outcomes
   Working-length estimation, apical-foramen localization, treatment planning, detection of obturation errors, post-treatment assessment, and prediction of endodontic success.
6. Education and clinical workflow
   Simulation, automated feedback for students, documentation, decision support, and patient communication.
7. Limitations, ethics, and future directions
   Dataset bias, data privacy, validation, clinician responsibility, explainability, regulation, and the need for prospective clinical evidence.

Artificial Intelligence in Conservative Dentistry and Endodontics

Introduction

• Machine learning (ML): Algorithms learn from labelled data.
• Deep learning (DL): A subset of ML using multilayer neural networks.
• Convolutional neural networks (CNNs): DL models particularly suited to analysing radiographs, CBCT scans, photographs, and other images.
• Artificial neural networks (ANNs): Computational systems designed to learn complex relationships between inputs and outcomes.

A. AI in Conservative Dentistry

1. Diagnosis of Dental Caries

• Early enamel caries
• Dentin caries
• Proximal caries
• Occlusal caries
• Root caries
• Recurrent or secondary caries adjacent to restorations

Clinical uses

• Automated marking of suspicious radiolucencies
• Estimation of lesion depth
• Caries classification and documentation
• Assistance in caries-risk assessment
• Monitoring lesion progression on serial images

Advantages

• May improve detection of subtle proximal lesions.
• Provides rapid and consistent image screening.
• Can be useful as a second reader, especially in high-volume practice or teaching.
• May reduce inter-examiner variation.

Limitation

2. Detection of Existing Restoration Defects

• Overhanging restorations
• Open or defective margins
• Proximal contact errors
• Voids and inadequate contour
• Recurrent caries
• Fracture or chipping of restorative material
• Marginal staining or possible microleakage

Importance

3. Treatment Planning and Caries-risk Prediction

• Individual caries risk
• Probability of new carious lesions
• Likely progression of early lesions
• Need for preventive, non-operative, or operative management
• Prognosis of a restoration

Applications

• Selection of preventive measures such as fluoride therapy, sealants, diet counselling, and recall intervals
• Decision support for minimally invasive dentistry
• Identification of patients needing closer review

4. Shade Selection and Esthetic Restorative Dentistry

• Shade matching
• Colour analysis
• Detection of tooth colour variations
• Smile analysis
• Prediction of esthetic outcomes
• Digital smile design
• Selection of composite shade, opacity, and layering approach

Benefits

• More objective than visual shade selection alone
• Can reduce variation due to lighting conditions, eye fatigue, and operator experience
• Useful in anterior composite restorations and ceramic shade communication with the laboratory

5. CAD/CAM and Digital Restorative Workflows

• Intraoral scanning
• Automatic tooth segmentation
• Margin detection
• Restoration design
• Occlusal-contact analysis
• Design of inlays, onlays, veneers, crowns, and indirect restorations
• Detection of preparation defects

Potential advantages

• Faster workflow
• Improved standardisation
• Improved communication between clinician and laboratory
• Potentially more conservative preparation and restoration design

6. AI in Dental Education

• Virtual simulation of cavity preparation
• Automated assessment of preparation dimensions, taper, undercuts, and depth
• Immediate feedback during training
• Case-based diagnostic exercises
• Personalised learning plans
• Assessment of radiographic interpretation skills

B. AI in Endodontics

1. Diagnosis of Periapical Pathosis

• Periapical radiolucency
• Apical periodontitis
• Persistent periapical disease after root-canal treatment
• Periapical bone changes
• External and internal root resorption
• Root fractures
• Perforations

Advantages

• May identify small lesions overlooked on routine interpretation.
• Provides rapid and reproducible screening.
• Can assist in follow-up assessment of healing.

Limitation

2. Detection of Root-canal Anatomy

• Number of roots
• Number of canals
• Additional canals, including the mesiobuccal second canal in maxillary molars
• Canal curvature
• Canal bifurcation or merging
• C-shaped canals
• Calcified or obliterated canals
• Root anomalies
• Dens invaginatus and other developmental variations

Clinical significance

3. Working-length Determination

• Root apex localisation
• Apical foramen localisation
• Estimation of working length from radiographs
• Identification of the cemento-dentinal junction region

Role in practice

1. Electronic apex locator use
2. High-quality working radiographs
3. Tactile feedback and clinical judgment
4. Knowledge of root anatomy and apical morphology

4. Detection of Root Resorption, Root Fracture, and Perforation

• Internal inflammatory root resorption
• External inflammatory root resorption
• Cervical root resorption
• Vertical root fracture
• Horizontal root fracture
• Iatrogenic perforation
• Furcation perforation

Clinical value

5. Treatment Planning

• Non-surgical root-canal treatment
• Retreatment
• Endodontic surgery
• Referral to an endodontist
• Restorability and prognosis
• Extraction versus tooth preservation
• Need for CBCT imaging in selected cases

6. Assessment of Root-canal Treatment Quality

• Missed canals
• Inadequate canal preparation
• Short, long, or non-homogeneous obturation
• Voids in obturation
• Inadequate coronal seal
• Overextension of filling material
• Underfilling
• Procedural errors
• Persistent periapical radiolucency

Uses

• Quality assurance
• Audit of clinical procedures
• Student feedback
• Standardisation of treatment-evaluation criteria

7. Prognosis and Outcome Prediction

• Preoperative periapical status
• Lesion size
• Tooth type and location
• Root-canal anatomy
• Quality of cleaning, shaping, and obturation
• Coronal restoration
• Retreatment status
• Periodontal condition
• Patient-related factors

8. AI in Endodontic Education

• Identification of canal anatomy on radiographs and CBCT
• Simulation of access cavity preparation
• Training in working-length determination
• Automated feedback on obturation length and density
• Assessment of procedural errors
• Virtual and case-based learning

Advantages of AI in Conservative Dentistry and Endodontics

1. Improved diagnostic support
   AI may identify patterns and subtle abnormalities that can be missed by the human observer.
2. Speed and efficiency
   Rapid analysis of radiographs and CBCT scans can streamline screening and documentation.
3. Consistency and reproducibility
   It can reduce variation among clinicians and observers.
4. Improved treatment planning
   AI may integrate multiple clinical and radiographic factors to support prognosis estimation.
5. Better education and training
   It enables simulation-based learning with immediate, objective feedback.
6. Audit and quality assurance
   It can help evaluate restorative and endodontic treatment outcomes systematically.
7. Patient communication
   Visual AI annotations may help patients understand caries, periapical lesions, or treatment needs.

Limitations and Disadvantages

1. Dependence on data quality
   Poor radiographs, motion artefacts, inadequate CBCT quality, and inaccurate labels can produce unreliable output.
2. Dataset bias
   An AI model trained on one population, device, imaging protocol, or institution may not perform equally well elsewhere.
3. Lack of external validation
   Many studies are retrospective and based on limited datasets. Real-world multicentre validation remains insufficient.
4. False-positive and false-negative findings
   AI may mark normal anatomy as disease or fail to recognise subtle disease.
5. Black-box problem
   Some deep-learning systems do not clearly explain how a conclusion was reached.
6. Ethical and legal responsibility
   The clinician, not the AI system, remains accountable for diagnosis and treatment decisions.
7. Privacy and data security
   Patient radiographs, CBCT scans, photographs, and clinical records must be stored and processed securely.
8. Cost and infrastructure requirements
   Acquisition, software subscriptions, training, and integration into practice workflows may be costly.
9. Risk of over-reliance
   AI must not replace clinical examination, pulpal and periapical testing, or sound biological principles.

Ethical Principles for AI Use in Dentistry

• Obtain appropriate consent for use and transfer of digital patient data.
• Protect confidentiality and comply with data-protection laws.
• Use validated and regulated systems where available.
• Check AI findings against clinical and radiographic evidence.
• Document when AI-assisted software has influenced decision-making.
• Ensure that the final decision remains with the treating dentist.
• Avoid using AI in a discriminatory way or without considering bias in the underlying dataset.

Future Directions

• Better integration of radiographs, CBCT, clinical notes, pulp-test findings, and photographs
• Real-time assistance during access preparation and canal negotiation
• Improved prediction of pulpal healing and endodontic prognosis
• Personalised caries-prevention programmes
• More reliable automated detection of missed canals, fractures, and resorptive defects
• Greater use of explainable AI, where the clinician can understand why the system produced a result
• Prospective multicentre validation before widespread routine use

Conclusion