Development and evaluation of an AI model for dental implant type detection: A comparison of diagnostic accuracy between a deep learning model and dental professionals
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
PURPOSE: To develop a deep-learning system for identifying five dental implant brands from periapical radiographs and compare its diagnostic accuracy with dental professionals and evaluate successive You Only Look Once (YOLO) architectures (v7-v12) to justify model selection. MATERIALS AND METHODS: A dataset of 5851 periapical radiographs was compiled and divided into training, validation, and test partitions (80/10/10). After filtering to five brands (Adin, Dentium, Noris, OSSTEM, and Straumann), YOLO-based object detection models (versions 7-12) were trained and tested under identical conditions. The YOLOv12x model was adopted for final evaluation based on its optimal balance of accuracy and inference speed. Human performance was assessed using 100 held-out test images (20 per brand) via a multiple-choice web survey distributed to six clinician groups. Diagnostic metrics included mean average precision at IoU 0.50 (mAP@50), mAP@50-95, precision, recall, and accuracy. RESULTS: The model achieved an mAP@50 of 0.989 (98.9%), an mAP@50-95 of 0.900 (90.0%), a precision of 0.969 (96.9%), and a recall of 0.977 (97.7%) across brands. Across YOLO generations, performance improved from mAP@50-95 = 0.817 (YOLOv7) to 0.905 (YOLOv12x). Fifty-two clinicians completed 5,200 image evaluations; the model significantly outperformed all clinician subgroups (one-way ANOVA with Tukey HSD, p < 0.001). Transformer-based DF-DETR achieved mAP@50-95 = 0.878, confirming YOLOv12x's superior efficiency-accuracy trade-off. CONCLUSIONS: A high-performing model identified implant brands on periapical radiographs and outperformed clinicians across experience levels. Comparative analysis across YOLO architectures validated its measurable advantage in accuracy and speed. Lack of external validation and dataset imbalance are important limitations; future work will include external, multisite data and human-AI workflow evaluation.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it