BEYOND AUTOMATED DETECTION: ARTIFICIAL INTELLIGENCE IN MULTIMODAL CRANIOFACIAL IMAGING, MORPHOLOGICAL PHENOTYPING, AND CLINICAL DECISION SUPPORT—A NARRATIVE REVIEW

Argatu Daniela, Balcos Carina, Norina Forna, Bamboi Irina, Corneaga Andrei, Doriana Agop-Forna

DOI : 10.62610/RJOR.2026.3.18.5

Abstract

Background: Artificial intelligence (AI) has expanded the analytical capabilities of craniofacial radiology, enabling automated detection, anatomical segmentation, quantitative image analysis, and increasingly integrated computational modeling. Its clinical significance, however, remains heterogeneous and is rarely demonstrated beyond technical performance. Objective: To critically synthesize AI applications in panoramic radiography, cephalometric imaging, cone-beam computed tomography (CBCT), conventional computed tomography (CT), and magnetic resonance imaging (MRI), and to organize them within a functional framework that distinguishes detection, quantification, morphological phenotyping, and clinical decision support. Methods: Structured narrative review. PubMed/MEDLINE, Scopus, Web of Science, and IEEE Xplore were searched for studies published between January 2018 and June 2026, supplemented by foundational earlier work and methodological guidance. Studies were prioritized according to reference-standard quality, independence of test data, multicenter or external validation, and the presence of clinical or longitudinal endpoints. The proposed Craniofacial AI Diagnostic Continuum was positioned against the hierarchical model of diagnostic imaging efficacy. Evidence synthesis: Representative studies show mature performance for cephalometric landmark localization, fracture and lesion detection, and tooth, bone, and mandibular-canal segmentation, including multicenter datasets of several thousand examinations. In contrast, morphological phenotyping and multimodal decision support rest on few cohorts, mostly single center, with prospective longitudinal validation reported only exceptionally (e.g., temporomandibular joint osteoarthritis progression). External validation, calibration, and evaluation of clinical impact remain uncommon. Conclusions: The continuum adds a phenotyping level that has no direct counterpart in classical efficacy hierarchies and makes explicit which evidence each level requires. Increasing algorithmic complexity does not confer clinical value; future research should prioritize reproducible morphometry, external and prospective validation, and outcome-based evaluation.