AlphaDent: A dataset for automated tooth pathology detection

A research team has released AlphaDent, a significant new open-source dataset designed to train AI models for automated dental pathology detection. The dataset is built from over 1,200 high-resolution DSLR photographs of teeth from 295 patients. Crucially, each image is meticulously annotated for instance segmentation, a computer vision task where an AI must identify and outline each distinct object (like a specific tooth or cavity) within the image. The data is categorized into 9 distinct classes, covering various dental conditions, providing a robust foundation for model training.

The team, which includes authors from institutions like the Moscow Institute of Physics and Technology, has also published the results of their own experiments training neural networks on AlphaDent, reporting high prediction quality. The entire package—dataset, training and inference code, and pre-trained model weights—is available under open licenses. This comprehensive release lowers the barrier to entry for developers and researchers aiming to build computer vision applications for dentistry, from automated screening tools to assistive diagnostics for dental professionals.