Professor Meng Liuyan's team from the Hospital of Stomatology at Wuhan University (WHU) has made remarkable progress in advancing AI-assisted dental imaging, with their latest research published online in Nature Biomedical Engineering.

The paper, Towards clinical-level interpretation of dental panoramic radiography using an instance-guided vision-language model, introduces an innovative solution for automated diagnosis and structured report generation of dental panoramic radiographs.

Panoramic radiography is a widely used tool for diagnosing and planning treatments for oral diseases. However, with increasing demand for imaging tests, there is a global shortage of resources for oral imaging diagnosis.

This imbalance can lead to delays in comprehensive interpretation, increasing the workload for clinicians and delaying treatment decisions. Imaging reports in clinical practice often suffer from incomplete records, focusing only on the patient's chief complaints, which can hinder a comprehensive assessment of oral health and increase the risk of missed or incorrect diagnoses.

To tackle these clinical challenges, the team developed a large-scale oral imaging dataset covering over 101,000 patients and proposed an instance-guided vision-language model named DentFound.

DentFound integrates tooth instance information into visual representation learning, guiding the model to focus on specific tooth positions, lesion areas, and post-treatment changes. This allows for integrated analysis for lesion localization, disease diagnosis, and structured report generation.

The model enhances its fine-grained understanding of complex oral images through multi-level knowledge resampling and progressive learning , enabling AI to move beyond single-disease recognition toward comprehensive, clinical-level interpretation of oral imaging.

Experimental results show that DentFound outperforms existing medical vision-language models in automated report generation and disease diagnosis. Expert evaluations reveal that the quality of reports generated by DentFound is superior to or comparable with those written by radiologists.

The study also acknowledges the significant contributions of researchers including Professor Du Bo from WHU's School of Computer Science, Professor Liu Zhonghao from Yantai Stomatological Hospital, and Chief Physician He Li from Shiyan People's Hospital.