Background/Objectives: The early and accurate identification of impacted teeth in the maxilla is critical for effective dental treatment planning. Traditional diagnostic methods relying on manual interpretation of radiographic images are often time-consuming and subject to variability. Methods: This study presents a deep learning-based approach for automated classification of impacted maxillary canines using panoramic radiographs. A comparative evaluation of four pre-trained convolutional neural network (CNN) architectures—ResNet50, Xception, InceptionV3, and VGG16—was conducted through transfer learning techniques. In this retrospective single-center study, the dataset comprised 694 annotated panoramic radiographs sourced from the archives of a university dental hospital, with a mildly imbalanced representation of impacted and non-impacted cases. Models were assessed using accuracy, precision, recall, specificity, and F1-score. Results: Among the tested architectures, VGG16 demonstrated superior performance, achieving an accuracy of 99.28% and an F1-score of 99.43%. Additionally, a prototype diagnostic interface was developed to demonstrate the potential for clinical application. Conclusions: The findings underscore the potential of deep learning models, particularly VGG16, in enhancing diagnostic workflows; however, further validation on diverse, multi-center datasets is required to confirm clinical generalizability.

Background/Objectives: Mandibular cortical morphology has been associated with vertical facial patterns, but differences in standardized radiomorphometric indices remain unclear. This study compared five mandibular indices among vertical skeletal patterns using CBCT-derived panoramic reconstructions. Methods: This retrospective cross-sectional study included 181 adults classified as hypodivergent (n = 42), normodivergent (n = 67), or hyperdivergent (n = 72) according to the sella–nasion-to-mandibular-plane angle. Mental (MI), panoramic mandibular (PMI), gonial (GI), antegonial (AI), and mandibular cortical (MCI) indices were assessed by a blinded dentomaxillofacial radiologist, and bilateral measurements were averaged. Continuous indices were compared using ANOVA or Welch ANOVA and ANCOVA adjusted for age and sex, and the MCI using the chi-square test and logistic regression. p values across the five indices were corrected using the Holm method. Results: No index differed significantly among the groups after correction for multiple comparisons. The MI showed the most notable nominal between-group difference (hypodivergent, 3.92 ± 0.73 mm; normodivergent, 3.73 ± 0.60 mm; hyperdivergent, 3.61 ± 0.68 mm). After adjustment for age and sex, the MI was 0.35 mm greater in hypodivergent than in hyperdivergent subjects (95% CI, 0.04–0.66), but the overall group effect did not remain significant after correction (p = 0.024; Holm p = 0.118). The unadjusted association between the MCI and vertical skeletal pattern (p = 0.035) did not remain significant after correction (Holm p = 0.174) or after adjustment. The PMI, GI, and AI showed no differences. Conclusions: After correction for multiple comparisons, vertical skeletal pattern was not significantly associated with any of the five radiomorphometric indices.

Aim: This study aimed to comparatively evaluate the performance of ChatGPT-5.4, Claude Sonnet 4.6, and Gemini 3.1 Pro large language models (LLMs) in answering theoretical orthodontic questions under web search-enabled conditions. Methodology: A total of 30 open-ended questions covering the topics of growth and development, functional analysis, mixed dentition, extraoral appliances, malocclusion, and orthodontic treatment planning were prepared from Contemporary Orthodontics (7th edition, 2025). The questions were presented to all three models under standardized conditions in separate sessions in March 2026 with web search enabled. The responses were anonymized and independently evaluated in a blinded manner by two orthodontists using a 5-point Likert scale incorporating accuracy, scientific adequacy, completeness, and clarity into a single overall score. Data were analyzed using IBM SPSS Statistics (version 31.0); the Friedman test was used for model comparisons, and statistical significance was set at p < 0.05. Results: Mean performance scores were 4.40 ± 0.72 for ChatGPT-5.4, 4.43 ± 0.61 for Claude Sonnet 4.6, and 4.32 ± 0.88 for Gemini 3.1 Pro. Median scores were 4.50, 4.50, and 4.75, respectively. No statistically significant difference in overall performance was found among the models (χ² = 0.400; df = 2; p = 0.819; Kendall's W = 0.007). Subgroup analyses by topic category also revealed no significant differences in any category (all p > 0.05). Inter-rater reliability was limited (α = 0.706). Conclusion: All three models achieved high-performance scores, with no statistically significant difference in overall performance among them; this does not establish equivalence. With independent verification, these LLMs may serve as supportive tools in orthodontic education.