YOLOv8-based Automatic Dental Caries Detection System: New Advances in AI-Assisted Oral Diagnosis

This project is an automated caries detection system released by Premkumar8 on GitHub on June 14, 2026. Based on the YOLOv8 deep learning model, it detects seven common dental diseases across multiple teeth and four tooth surfaces in panoramic dental X-rays, generates AI diagnostic reports, provides auxiliary diagnostic support for oral healthcare, and features scalability and a localized web application interface.

Dental caries is a common oral disease, and early detection is crucial; traditional detection relies on doctors' experience and is prone to missed diagnoses. This project breaks through the limitations of traditional single-tooth and single-surface detection, enabling simultaneous detection across multiple teeth and four tooth surfaces in panoramic X-rays, which is more aligned with real-world clinical application scenarios.

The system is trained based on the YOLOv8 object detection model, outputting disease labels, confidence scores, and AI diagnostic reports. It uses a dataset-driven design that can automatically read category names and has a built-in knowledge base for seven diseases (including caries, deep caries, impacted teeth, etc.), enhancing interpretability and scalability.

It includes four main modules: dataset download (automatic acquisition from Kaggle + retry mechanism), model training (outputs optimal model and training history), prediction report (single X-ray inference + JSON report saving), and web application (localized UI for upload and real-time result viewing).

Requires Python 3.10 environment and dependent packages; dataset download supports resumable transfer; training allows specifying paths and epochs, with models saved to the artifacts directory; single prediction requires specifying image and model paths to output summaries and JSON reports; the web application starts a local service via command.

This system is an AI-assisted diagnostic tool, not a clinical diagnostic tool; its performance depends on dataset completeness, balance, and image quality; results need to be verified by professional doctors in production environments; more diverse clinical data is needed for validation and optimization to improve generalization ability.

Possible directions include expanding disease detection categories, improving the ability to handle complex cases; exploring integration with electronic medical record systems; introducing attention mechanisms to enhance key area localization; developing a mobile version to improve accessibility for primary healthcare institutions.