Categories
Change Password!
Reset Password!
Artificial intelligence accurately segments and classifies oral lichen planus, oral leukoplakia, and benign oral ulcers, advancing precision diagnosis and clinical decision support in oral healthcare.
One of the biggest barriers to applying artificial intelligence (AI) in oral healthcare has not been the algorithms themselves, but the shortage of large, standardized clinical datasets with reliable expert annotations. To address this gap, investigators developed a comprehensive image dataset including oral lichen planus, oral leukoplakia, and oral benign ulcers. They assessed how accurately leading deep learning models could localize lesions and distinguish between these clinically important oral mucosal disorders.
The project brought together 808 high-resolution intraoral images from 350 patients, each independently reviewed by three oral medicine specialists. Every image was assigned a diagnostic label and detailed pixel-level lesion annotation before being used to train and validate five state-of-the-art deep learning architectures, enabling direct comparison of their diagnostic performance under standardized conditions.
The benchmark results demonstrated that the dataset supported robust model development across all evaluated architectures. The highest-performing model achieved a mean Dice coefficient of 0.768 ± 0.031 for lesion segmentation, while disease classification reached an overall accuracy of 94%, reflecting strong performance in both lesion localization and diagnostic prediction. The expert-generated pixel-level annotations also provided a reliable reference standard for evaluating and comparing different AI models.
Beyond improving algorithm performance, the study establishes an important resource for future research in AI-driven oral diagnostics. A standardized, high-quality dataset enables more reproducible model development, facilitates fair benchmarking across emerging technologies, and brings AI-assisted diagnosis a step closer to routine clinical implementation. As these tools continue to evolve, they may enhance diagnostic consistency and support clinicians in the early detection of common oral mucosal diseases.
International Dental Journal
Artificial Intelligence-Driven Segmentation of Three Oral Diseases: Enabling Precision Diagnosis and Decision Support
Baowen Cheng et al.
Comments (0)