Reproduce “Towards Generalist Intelligence in Dentistry: Vision Foundation Models for Oral and Maxillofacial Radiology” on OpenResearch

About this paper

Xinrui Huang, Fan Xiao, Dongming He, Anqi Gao, Dandan Li, Xiaofan Zhang, Shaoting Zhang, Xudong Wang

Researchers from Shanghai Jiao Tong University and Shanghai Ninth People’s Hospital introduce DentVFM, the first family of vision foundation models specifically for oral and maxillofacial radiology, leveraging a new large-scale multi-modal dataset called DentVista. DentVFM achieves generalist intelligence across diverse dental tasks, demonstrating up to 13% improvement over baselines and outperforming experienced dentists by 3.3-13% in cross-modality diagnoses, while also showing exceptional label efficiency.

Key insights

• Scaling laws established in general computer vision extend to the dental domain, with larger models benefiting from more extensive and diverse pre-training data.
• Self-supervised learning on a large, multi-modal dental imaging dataset effectively yields robust, generalist visual representations, outperforming single-modality pre-training.
• Foundation models can enable accurate cross-modality diagnosis, allowing low-cost modalities like panoramic X-rays to detect conditions typically requiring expensive examinations such as MRI or pathology.

Summary by alphaXiv.