Background/Objectives: Deep learning models for cephalometric landmark detection report near-ceiling accuracy on single benchmarks, yet most are trained and tested on the same dataset. Whether the best in-domain architecture remains best out-of-domain has not been systematically quantified. Methods: Four architecture families (heatmap CNN, two-stage cascade, coordinate-regression Vision Transformer, pretrained ResNet-50) were each trained on two independently sourced datasets—ISBI 2015 (400 images, one device) and Aariz (1000 images, seven devices)—and evaluated on both, over their 19 shared landmarks in millimeters. All axes used three seeds (mean ± SD): the cross-dataset matrix, leave-one-device-out shift, balanced joint training, a pretrained-versus-scratch ablation, and a landmark-level breakdown. Results: In-domain mean radial error (MRE) was 2.47–3.04 mm (Aariz) and 4.12–5.55 mm (ISBI); cross-dataset error rose steeply (Generalization Drop—the relative increase in error out-of-domain—155–426%). The most accurate model in-domain (a from-scratch heatmap CNN, 2.47 mm) showed the largest drop, and every pretrained estimate fell below every from-scratch estimate (family means 248% vs. 380%): in-domain ranking did not predict cross-domain ranking. Leave-one-device-out revealed reproducible device-specific shift (held-out MRE 1.89–7.94 mm). Inter-observer variability was 0.53 mm, so cross-domain errors were 15–44× the human band. Balanced joint training reduced the cross-domain gap for all four architectures (both domains ≈ 2.1–3.5 mm) without harming in-domain accuracy. Pretraining more than halved cross-domain error from Aariz to ISBI (7.73 vs. 16.36 mm) but not in reverse, supporting the mechanism in one direction rather than uniformly. A-point, B-point, Nasion, and Menton all exceeded the 2 mm clinical threshold out-of-domain, including points localized to sub-millimeter accuracy in-domain. Conclusions: Single-dataset accuracy substantially overstates clinical generalizability, and the best in-domain architecture is not the most transferable, so in-domain leaderboards are an unreliable basis for selecting a model for clinical deployment; balanced multi-source training recovers most of the loss across the architectures tested.

Background/Objectives: Large language models (LLMs) are consulted for clinical decision support, and their reliability is judged by hallucination prevalence—a measure treating every unsupported element as equivalent, although a fabricated citation and a fabricated protocol differ in what a clinician acting on them would do. We tested whether prevalence tracks clinical risk. Methods: Five LLMs answered a 100-item orthodontic benchmark validated by three external orthodontists (content validity index 0.923). The reference standard was fixed by construction for the 45 items naming a non-existent entity, so any substantive elaboration is unsupported by design; citations were adjudicated against PubMed and CrossRef. Responses were coded with a seven-category taxonomy and assigned to clinical, operational, or epistemic severity tiers in a post hoc exploratory stratification, pre-specified rather than prospectively registered. Because all models answered the same items, comparisons used Cochran’s Q with pairwise McNemar tests and generalised estimating equations clustered on item. Results: Of 500 responses, 449 (89.8%) contained a hallucination but only 65 (13.0%; 95% CI 10.3–16.2) were clinically consequential: prevalence was roughly sevenfold greater than the rate of clinically consequential output as the authors defined it. Between-model differences were large for undifferentiated prevalence (73–100%; Q = 50.51, p < 0.001) and contracted at the clinical tier (10–16%; Q = 10.00, p = 0.040), where no pairwise contrast survived adjustment. Item-level clustering was far stronger for clinically consequential output than for undifferentiated prevalence (intra-class correlation 0.83 versus 0.07). Conclusions: Prevalence and clinically consequential error are not interchangeable and rank models differently. The stratification is exploratory and author-defined, and the benchmark stress-tests susceptibility to fabricated premises rather than surveying natural use.

Direct-print aligners are widely presented as the environmentally preferable alternative to thermoforming, but which part of the process governs the carbon released during manufacture has not been established. Four commercial printer–resin workflows were modelled: Graphy Tera Harz resin on a Uniz NBEE (A) and an Asiga MAX 2 (B), the LuxCreo iLux Pro as a closed chain (C), and Graphy resin on a DentaFab Sega Pro (D). Energy, material and throughput data came from the manufacturers. Production-phase CO₂e per aligner was propagated through 10,000 Monte Carlo draws, with the resin emission factor sampled across its full published range (3.0–6.5 kg CO₂e/kg) and the thermoformed comparator represented as a distribution. Mean footprints were 0.043, 0.036, 0.089 and 0.020 kg CO₂e per aligner (95% uncertainty intervals were 0.032–0.056, 0.026–0.047, 0.077–0.101 and 0.016–0.025), against a thermoformed median of 0.190. Direct printing was the lower of the two in every paired iteration. Resin dominated, contributing 41–79% of the total; solvent added a further 41% in the one workflow using it, while electricity of all kinds accounted for 19–36%. Thermoforming would match direct printing only at series lengths of 45 to 202 aligners, far above the 14–30 of clinical practice; under the most unfavourable resin factor the margin narrowed but did not close. Material efficiency, not energy sourcing, is where reductions are available.

Ortodontide Kanıta Dayalı Klinik Uygulamalar, çağdaş ortodonti pratiğini şekillendiren temel ve yenilikçi konuları bir araya getiren kapsamlı bir akademik eserdir. Bu kitapta; MARPE uygulamalarından fonksiyonel ortodontiye, Sınıf III maloklüzyonların yönetiminden ağız solunumunun dentofasiyal etkilerine, ortodontik diş hareketinin hızlandırılmasından kök rezorpsiyonuna kadar geniş bir yelpazede güncel bilgiler sunulmaktadır. Ayrıca yapay zekâ destekli tanı ve tedavi planlaması, doğrudan 3D baskılı alignerlar, 4D ortodontik materyaller ve şeffaf plak sistemlerine ilişkin en güncel bilimsel gelişmeler ayrıntılı biçimde değerlendirilmektedir. Kanıta dayalı yaklaşımı merkeze alan bu eser; klinik karar verme süreçlerinde güvenilir bilimsel verileri öne çıkarırken, geleneksel ve modern tedavi yöntemlerini karşılaştırmalı olarak ele almakta ve güncel literatür ışığında yorumlamaktadır. Ortodonti uzmanları, akademisyenler, araştırmacılar, uzmanlık öğrencileri ve diş hekimleri için hazırlanan bu kitap, hem günlük klinik uygulamalarda yararlanılabilecek pratik bilgiler hem de bilimsel perspektif kazandıran kapsamlı değerlendirmeler sunmaktadır. Bilimsel kanıtlarla desteklenen güncel bilgiler, yenilikçi teknolojiler ve geleceğin ortodontisine yön veren yaklaşımlar bu eserde bir araya getirilmiştir.

Thermoforming remains the predominant fabrication route for clear thermoplastic orthodontic appliances, yet it reduces and redistributes sheet thickness, which governs force delivered to teeth. Simultaneous thermoforming of two models may reduce fabrication time and material waste; however, its effect on final appliance thickness has not been evaluated. This in vitro study compared appliance thickness after thermoforming over a single model or two models simultaneously, using ten commercial aligner brands with different polymer composition and initial thickness. Thickness was measured with a digital caliper at anterior, canine and posterior sites, and analyzed using three-way ANOVA with Tukey post hoc tests. A significant three-way interaction (measurement point × number of models × brand; p < 0.001) was found. Significant single- versus dual-model differences emerged in some brand–region combinations; in most, dual-model appliances were thicker than single-model appliances. Four brands showed no significant difference at any site. Within the limits of this study, simultaneous fabrication of two appliances in a single thermoforming cycle produced appliances with thicknesses comparable to, and often slightly greater than, those fabricated over a single model. Simultaneous thermoforming maintained comparable appliance thickness and may improve manufacturing efficiency.

Background/Objectives: Patients increasingly use large language models (LLMs) to understand orthodontic conditions, yet little is known about how leading models compare on the same patient-communication task or how performance varies with the recipient addressed. Methods: In this blinded pilot, five LLMs—Claude (claude-opus-4-7), GPT (gpt-5.5-2026-04-23), Gemini (gemini-2.5-pro), DeepSeek (deepseek-v4-flash), and Grok (grok-4.3)—each explained three orthodontic malocclusions while adopting three clinician personas and addressing three recipient profiles of differing professional background in a complete balanced 5 × 3 × 3 × 3 design (135 responses in Turkish). Three orthodontists independently scored each response on a seven-axis rubric (1–5). Differences used Kruskal–Wallis, Friedman, and Scheirer–Ray–Hare tests with Holm corrections, epsilon-squared effect sizes, and bootstrap confidence intervals. Results: Inter-rater reliability was good (composite ICC [2, 1] = 0.79). Composite scores differed modestly across models (p = 0.044), but no pairwise model difference survived correction; separation was driven by terminology (p = 0.005). Among factors, the recipient profile showed the largest effect size (epsilon-squared = 0.058 vs. 0.045 for model); the engineer profile scored lower than a healthcare worker or layperson (both corrected p < 0.05), with no statistically significant model × recipient interaction detected. Neither persona nor malocclusion affected quality. Conclusions: Because these were prompted profiles and outcomes were expert-rated, deployment and oversight may benefit from attention to audience fit and terminology rather than model choice alone.