Year: 2026
Type: preprint
Abstract: Dental caries is the most prevalent non-communicable disease worldwide, yet lesions are often diagnosed late because conventional examination depends on clinical visits and on the experience of the examiner. This article presents Den-tAI, a cross-platform mobile application that detects and classifies dental caries in intraoral photographs and communicates the findings through an augmented-reality-style live camera overlay. A dataset of 900 clinically reviewed intraoral images with 1,504 bounding-box annotations across five lesion classes was curated from two public sources through de-duplication, expert screening, and class bal-ancing. End-to-end YOLOv8 and YOLOv11 detectors and a two-stage YOLOv8 + MobileNetV3 pipeline were compared under identical conditions. YOLOv8 achieved an [email protected] of 72.96%, an [email protected]:0.95 of 43.95%, a precision of 69.10%, and a recall of 70.97%; YOLOv11 reached a higher precision (83.43%) and [email protected] (73.90%) but a lower recall (63.54%); the two-stage pipeline reached an [email protected] of only 54.88%. YOLOv8 was deployed because of its higher recall and balanced behavior. The model runs on a Flask backend on AWS EC2 behind a Flutter app that also offers dentist chat, appointment booking, scan history, and OCR-assisted registration. A pilot session at a private dental clinic indicated that the practicing dentist found the application clinically relevant and useful for explaining findings to patients.
Source: Research Square
Institutions: German University in Cairo, City University Ajman
Language: English
Subfield: Oral Surgery
Field: Dentistry
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7236283695
