A Scoping Review of Multimodal Artificial Intelligence Integration in Dental Diagnostics and Treatment Planning: Foundations for the DAIS Model.

Abstract

Background: Artificial intelligence (AI) has demonstrated remarkable potential across dental practice; however, applications remain predominantly fragmented into single-modality, specialty-specific tools. This fragmentation limits comprehensive diagnostic and treatment planning that reflects the inherently multimodal nature of clinical dentistry.

Objectives: Guided by the Population-Concept-Context (PCC) framework, this scoping review mapped current AI applications in dentistry to: (1) characterize the extent of multimodal data integration; (2) synthesize evidence on diagnostic accuracy across specialties; (3) identify the availability of clinical implementation and economic evaluation evidence; and (4) map research gaps to inform the development of integrated frameworks, using the proposed Dental-AI Synergy (DAIS) model as a potential exemplar.

Eligibility Criteria: Studies published from January 2018 to August 2026 focusing on AI, machine learning, or deep learning in clinical dentistry, prioritizing those incorporating multiple data modalities (imaging, structured clinical data, molecular/biomolecular data, or textual data).

Sources of Evidence: A systematic search was conducted across PubMed/MEDLINE, Scopus, Web of Science, Embase, IEEE Xplore, Cochrane Library, and CINAHL.

Charting Methods: Standardized data extraction, piloted on five studies, captured study characteristics, AI models, data modalities, performance metrics, and key findings. Data were synthesized qualitatively using thematic analysis.

Results: Of 1,825 records identified, 45 studies met inclusion criteria. The majority (76%) relied on single-modality approaches, predominantly imaging-based. Only 18 studies (40%) integrated two or more modalities. Convolutional neural networks were most common (22 studies). Diagnostic performance was high, with AUC values frequently exceeding 0.90. However, external validation was limited (18%), economic evaluations were sparse (3 studies), and implementation research was minimal.

Conclusions: While AI demonstrates substantial potential to enhance diagnostic accuracy, the field remains characterized by fragmentation and a limited evidence base for implementation. Identified gaps, limited multimodal integration, insufficient external validation, thin economic evidence, and scarce implementation research provide clear rationale for developing unified, multimodal frameworks. The DAIS model is proposed as one such framework to guide future research and development.