# Decadentry

> Decadentry translates the newest peer-reviewed research on artificial intelligence in dentistry into clear, honest, fact-checked articles for patients and clinicians. Every article is built on one specific, named peer-reviewed study, checked against the primary source before publication, and credits the original authors with the DOI (and ORCID iDs where available). Decadentry is an independent educational publication led by Hossein Boustani Hezarani, DDS; it is not affiliated with the studies' authors or any commercial AI platform. Content is educational, not medical advice.

## About
- [About Decadentry](https://decadentry.com/about/): Mission, principles and who is behind the publication.
- [Editorial Standards & Methodology](https://decadentry.com/about-editorial-standards/): How studies are chosen, fact-checked against the primary source, credited and corrected; disclosure of AI assistance.
- [Author: Hossein Boustani Hezarani, DDS](https://decadentry.com/hezarani/): Dentist (Azerbaijan Medical University) and AI-in-healthcare researcher; ORCID 0009-0007-8481-0498.
- [Publications & Books](https://decadentry.com/publications/): The author's peer-reviewed articles, contributed books and open-source dental software, each with a DOI.
- [AI in Dentistry Glossary](https://decadentry.com/glossary/): Plain-English definitions of 56 terms used in dental AI research (sensitivity, AUC, CNN, U-Net, YOLO, CBCT, RAG, external validation and more).
- [Medical Disclaimer](https://decadentry.com/medical-disclaimer/): Educational content only; not a substitute for professional dental or medical advice.

## Topics
- [All research](https://decadentry.com/research/): Every explainer, newest first.
- [Diagnostics & Imaging](https://decadentry.com/research/diagnostics-imaging/): AI reading radiographs, CBCT and photos — caries, oral cancer, osteoporosis, TMJ, age estimation.
- [Periodontology](https://decadentry.com/research/periodontology/): AI for gingivitis, periodontitis staging and risk prediction.
- [Orthodontics](https://decadentry.com/research/orthodontics/): Cephalometric landmarks, crowding, aligner outcomes and treatment simulation.
- [Implants & Oral Surgery](https://decadentry.com/research/implants-oral-surgery/): Implant planning and identification, nerve detection, surgical robots.
- [Endodontics & Restorative](https://decadentry.com/research/endodontics-restorative/): Root canals, pulp therapy, crown design and shade matching.
- [Generative AI & Chatbots](https://decadentry.com/research/generative-ai-chatbots/): ChatGPT, large language models and retrieval-augmented generation in dentistry.
- [Ethics, Education & Patient Care](https://decadentry.com/research/ethics-education-patient-care/): Trust and explainability, cost-effectiveness, education and patient experience.

## Contact
- [Contact](https://decadentry.com/contact-us/): Research collaboration, media and corrections.

## Articles
- [The Dentist in Your Pocket: Can AI Turn a Smartphone Photo Into a Cavity Detector?](https://decadentry.com/ai-smartphone-cavity-detection/): A 2025 systematic review of AI smartphone apps for detecting dental caries — strong on obvious cavities (up to 99%), weak on early lesions (as low as 37%).
- [AI in Dental Education: Better Than a Lecture — or a Tutor?](https://decadentry.com/ai-dental-education-evidence/): A 2025 meta-analysis of AI-assisted problem- and case-based learning — beats lectures, not human tutors.
- [AI vs. Endodontist: Who Reads a Dental X-Ray Better?](https://decadentry.com/ai-vs-endodontist-dental-xray/): AI vs expert endodontists for detecting apical lesions, with 3D CBCT as the gold standard (AI 47.9% vs humans 65.3% sensitivity).
- [Explainable AI in Dentistry: Real Trust or Just a Story?](https://decadentry.com/explainable-ai-dentistry-trust/): A 2026 systematic review of explainable AI in dentistry — 14 of 19 studies at high risk of bias; only one tested human understanding.
- [AI Implant Planning: Better Citations, Worse Spatial Sense?](https://decadentry.com/ai-implant-planning-rag/): A 2026 study of retrieval-augmented (guideline-grounded) AI for implant planning — better citations, no better clinical accuracy, a bias toward overtreatment.
- [AI Oral Cancer Screening: Early Warning or False Alarm?](https://decadentry.com/ai-oral-cancer-screening/): A 2026 Scientific Reports study of a deep-learning cytology test that scores brushed oral-lesion cells for cancer risk — near-perfect at telling a healthy mouth from cancer (AUROC 0.99), but only 0.78 at catching the earliest dangerous lesions.
- [AI Periodontitis Staging: Real Diagnosis or Guesswork?](https://decadentry.com/ai-periodontitis-staging/): A 2025 BMC Oral Health study of a YOLOv8 system that stages and grades gum disease from panoramic X-rays — near-expert on the bone-loss landmarks (0.95–0.97 precision) but weaker at identifying individual teeth (0.82), with a rulebook, not the AI, making the final call.
- [Can AI Cephalometric Landmark Detection Match a Specialist?](https://decadentry.com/ai-cephalometric-landmark-detection/): A 2025 European Journal of Medical Research multicentre study of a lightweight 3D deep-learning model that places craniofacial landmarks on CT and cone-beam CT (1,190 scans) — matched a senior specialist (~1.2 mm error) and beat a junior, but retrospective, validated on one region's scans, and not yet clinic-proven.
- [ChatGPT vs. Radiologist: Can AI Read a Dental Radiograph?](https://decadentry.com/ai-dental-radiograph-chatgpt-grok/): A 2026 Annals of Medicine study comparing ChatGPT, Grok and the healthcare-tuned MANUS against two oral radiologists on 120 dental radiographs — the AIs matched 88.3–95.0% of expert diagnoses vs the radiologists' 96.7%, but only on curated textbook images and with no equivalence test.
- [AI Crown Design: Better Than a Master Technician?](https://decadentry.com/ai-crown-design/): A 2025 Journal of Dentistry in-vitro study comparing AI-designed crowns with an experienced technician's on 12 premolars — AI matched natural-tooth width where the technician's crowns ran narrower and tied on overall 3D shape, but couldn't reproduce the functional wear facets that govern the bite.
- [AI Osteoporosis Screening: Can a Dental X-Ray Catch It First?](https://decadentry.com/ai-osteoporosis-screening-dental-x-ray/): A 2025 Journal of Dentistry meta-analysis of 24 studies training AI to detect osteoporosis on dental panoramic X-rays — pooled ~88% sensitivity and ~82% specificity, but per-study accuracy ranged 50–99% with signs of publication bias, making it an early-warning flag confirmed by a DXA scan, not a diagnosis.
- [AI Wisdom Tooth Surgery: Can It Spot the Nerve in Time?](https://decadentry.com/ai-wisdom-tooth-surgery-nerve/): A 2026 BMC Oral Health study of a dual-stream deep-learning model that grades wisdom-tooth-to-nerve (M3M–IAC) contact from a panoramic X-ray plus radiologist-selected CBCT slices — 0.9478 macro F-score and ~0.98 AUC on 250 single-center cases, but it still needs the CBCT scan, showed no significant gap between the four AI backbones, and isn't externally validated.
- [AI Pulpotomy: Can a Machine Save a Tooth's Nerve?](https://decadentry.com/ai-pulpotomy-pulp-assessment/): A 2026 Journal of Dentistry study of a self-supervised DINOv2 model that judges whether an exposed tooth pulp is healthy enough to cap during a pulpotomy — matched expert endodontists (95.7% internal accuracy) and beat novices, but fell to 85.7% on just 21 external images.
- [Is AI in Dentistry Cost-Effective, or Just Costly?](https://decadentry.com/ai-dentistry-cost-effectiveness/): A 2026 Frontiers in Medical Technology systematic review of the health economics of AI oral-disease screening — only 4 studies qualified, all German cavity studies; cost models favored AI (up to 64 vs 62 tooth-years, €298–€378 vs €322–€419), but the one randomized trial found identical cost (€330 vs €330) and tooth retention (49 vs 49 years) because sharper detection led to more drilling.
- [Black Triangles After Clear Aligners: Can AI Predict Them?](https://decadentry.com/black-triangles-clear-aligners-ai/): A 2025 Progress in Orthodontics study of a machine-learning nomogram that predicts "black triangles" (open gingival embrasures) before clear aligner treatment — pairing the Invisalign ClinCheck simulation with clinical factors reached 0.88 AUC, but it is one retrospective, single-clinic, Invisalign-only model with no external validation that still misses about one in three real cases (66% sensitivity).
- [AI Dental Age Estimation: Can It Tell If You're 18?](https://decadentry.com/ai-dental-age-estimation-18/): A 2026 Diagnostics study of YOLO deep-learning models reading wisdom teeth on panoramic X-rays for forensic age estimation — near-perfect at locating the teeth (94.5% accuracy) but, at the legally decisive 18-year threshold, correctly flagging only 23.5% of people who were truly 18 or older while catching 96.5% of under-18s; single-center, one population, no external validation.
- [ChatGPT in Pediatric Dentistry: Helpful or Half-Right?](https://decadentry.com/chatgpt-pediatric-dentistry/): A 2025 BMC Oral Health study in which 30 pediatric dentists graded ChatGPT-4's answers to 60 questions (30 parent FAQs, 30 dental-school questions) across six topics — about 4/5 for accuracy overall and as strong on parents' questions as on exam questions, but weakest on fluoride (3.99/5) and baby-tooth pulpal therapy (3.93/5), with the authors warning the high scores reflect perceived quality, not clinical safety.
- [AI for TMJ Disorders: Diagnosis or Educated Guess?](https://decadentry.com/ai-tmj-disorders-diagnosis/): A 2026 Diagnostics systematic review and meta-analysis of AI for diagnosing temporomandibular (jaw-joint) disorders — pooled ~79% sensitivity and ~87% specificity for TMJ osteoarthritis on imaging, but only 3 of 1,471 screened papers had comparable enough data to combine, so AI reads as decision support, not a replacement for MRI, CBCT, or a DC/TMD clinical exam.
- [AI Gingivitis Detection: Can an App Read Your Gums?](https://decadentry.com/ai-gingivitis-detection/): A 2025 Journal of Clinical Medicine study of a consumer intraoral-camera AI filter for gingivitis — its calls matched the clinical gold standard (bleeding on probing) only at chance (Cohen's κ 0.055), and a dentist reading the same photos did almost as poorly (κ 0.087), because gum inflammation is felt on probing, not seen in a surface photo; it also had to exclude pigmented gums, a real equity gap.
- [AI Dental X-Ray Enhancement: Real Detail or Illusion?](https://decadentry.com/ai-dental-xray-enhancement/): A 2026 International Dental Journal study testing whether "super-resolution" AI that sharpens blurry periapical X-rays helps other AI tools read them — boundary precision for outlining structures roughly doubled (boundary IoU 23.6%→48.2%, near the 50.0% true-high-res ceiling) and detection recall rose 4.2 points, but overall detection precision barely moved and caries detection dipped; tested only on artificially blurred images from one hospital, by algorithms, with no dentist reader study, so it's an algorithm-level proof of concept.
- [AI Dental Trauma Advice: Grounded Fact or Confident Guess?](https://decadentry.com/ai-dental-trauma-advice/): A 2026 Journal of Dentistry study of DT-RAG, a curated retrieval-augmented AI that answers dental-trauma questions only from five authoritative guidelines (IADT, ESE, AAE, Krastl, Cochrane) — it scored 96.0% on 99 yes/no questions vs 87.9% for the best general chatbot, lifted the base model's sound-reasoning rate from 49.5% to 98.0% and cut made-up rationales from 21 to 0, and was ranked first by all seven blinded specialists on ten scenarios; but it was tested on benchmarks and vignettes, not real patients, and the authors say clinical safety still needs prospective validation.
- [AI Teeth Crowding Check: Can a Photo Replace the Mold?](https://decadentry.com/ai-teeth-crowding-photo/): A 2026 BMC Oral Health study of an HRNet neural network that measures lower-front-teeth crowding (Little's Irregularity Index) from a single intraoral photo — tight per tooth (0.42 mm average error, 90% of landmarks within 1 mm), but the summed index overshot, overestimating crowding by 1.07 mm with a −1.36 to +3.50 mm error band that blew past the ±2 mm the team set as clinically acceptable; single-center, one annotator, so the authors call it a complementary tool for remote monitoring, not a replacement for the mold.
- [AI Dental Implant Identification: Sharper Than a Dentist?](https://decadentry.com/ai-dental-implant-identification/): A 2026 Journal of Dentistry in-vitro study of a two-stage AI (Mask R-CNN + ResNet-50) that tells four dental implant systems apart on X-rays — two original brands and two near-identical generics — classifying every test radiograph correctly (100%) vs 86.6% for 61 dentists and students (p<0.001); but it only chose among four trained brands (the closed-set trap), on standardized ex-situ films, so the authors call for real-world clinical validation.
- [AI MB2 Canal Detection: Real Fix or Just a Prototype?](https://decadentry.com/ai-mb2-canal-detection/): A 2026 International Dental Journal scoping review of AI for finding the easily-missed MB2 canal in upper molars on CBCT — lab scores were strong (F1 up to 0.93, sensitivity/accuracy above 0.80 across all five studies), but only 5 of 1,218 records qualified and every model stalled at readiness level 4 of 9 (prototype) with none externally validated, so it's a decision-support adjunct, not a standalone tool.
- [AI for Dental Anxiety in Kids: Calm or a Slower Pulse?](https://decadentry.com/ai-for-dental-anxiety-children/): A 2026 European Archives of Paediatric Dentistry randomised controlled trial of an AI-personalised video self-modelling app vs a standard distraction video for 80 children (6-12) facing fillings - both cut dental fear and anxiety, but on the primary anxiety questionnaire (CFSS-DS) the AI app was no better than the ordinary video; its only clear edge was a larger heart-rate drop (7.65 vs 2.18 bpm, p<0.001, Cohen d=0.89), a physiological signal, not the main measure of fear.
- [AI Orthodontic Simulation: Real Preview or a Guess?](https://decadentry.com/ai-orthodontic-simulation/): A 2026 Journal of Dentistry study of a diffusion model (Stable Diffusion + ControlNet) that synthesizes a photorealistic post-treatment facial profile from a target soft-tissue contour — realistic enough that laypeople told AI from real faces only 51.3% of the time (chance) and orthodontists just 71.0%, with 95% of landmarks within 2 mm; but it renders a contour it is given rather than predicting the outcome, and was built on only 10 adult Asian women, single-center, with no external validation.
- [AI Implant Planning: A Surgeon's Blueprint or a Rough Draft?](https://decadentry.com/ai-implant-planning-cbct/): A 2026 Journal of Dentistry study of ImplantPlanNet, a deep-learning framework that drafts single-tooth dental-implant plans straight from a CBCT scan — it aimed within about 1.5–1.7 mm and 5.6–6.9° of specialists' reference plans, but chose the right implant diameter only 75% of the time on new scans (100% on familiar ones) and the right length just 65% in both sets, on 144 scans from one center; the authors present it as a clinician-supervised proposal, not an autopilot.
- [Autonomous Dental Implant Robot: Precision or Peril?](https://decadentry.com/autonomous-dental-implant-robot/): A 2026 Journal of Dentistry in-vitro study of a fully autonomous robot (Yakebot) placing zygomatic implants in resin jaw models — tight precision (mean 0.87 mm deviation at the neck, 1.26 mm at the deep apex, 1.22° angulation) but on rigid plastic copied from a single patient, with the model cracking at 8 of 40 sites and no comparison to human surgeons and no clinical validation.
- [AI Gum Disease Prediction: Forecast or Shaky Guess?](https://decadentry.com/ai-gum-disease-prediction/): A 2026 Journal of Clinical Periodontology systematic review and PROBAST critical appraisal of 27 periodontitis prognostic models (1979–2025) — of the 20 scorable models, all but one were at high risk of bias (19 of 20 high in the statistical Analysis domain), driven by weak validation, missing-data handling and predictors assessed without blinding; data-driven models are better documented than old scorecards but mostly lack external validation, so a gum-disease risk score is guidance, not a verdict.
- [AI Dental Shade Matching: Can a Photo Pick Your Tooth Color?](https://decadentry.com/ai-dental-shade-matching/): A 2026 Journal of Dentistry study of PerceptShade, an AI that estimates tooth shade from routine smartphone or camera photos under everyday lighting — its single best pick matched a three-expert consensus 76.0% of the time (95%CI 61.8-86.9) vs 72.7% agreement among the experts themselves, with the correct shade in its top three 96.44% of the time (Top-1 79.12% on the main test set); but it was judged against photographs and an expert panel rather than a spectrophotometer or the finished restoration, still needs a physical reference tab in the frame, and came from a single institution with no external validation.