Yonsei University Dental School Uses AI to Identify Optimal Timing for MRI Scans in Dental Patients

Development of an AI Model Based on Panoramic X-ray Images

AI technology is expected to improve diagnostic efficiency by quickly identifying patients who require MRI (Magnetic Resonance Imaging).

On March 19, Professor Yeonjeong Park of the Department of Oral Medicine at Yonsei University Dental Hospital, Professor Hyojung Jung of the Oral Science Research Institute at Yonsei University College of Dentistry, Professor Seongjae Hwang of the College of AI Convergence at Yonsei University, and Master’s student Dayoon Joo at the Graduate School of AI Convergence announced the development of an AI model that analyzes both panoramic X-ray images and clinical information to pre-screen for abnormalities that are typically detected in temporomandibular joint (TMJ) MRIs.

Professor Youn Jeong Park, Department of Oral Medicine, Yonsei University Dental Hospital; Professor Hyojung Jung, Institute of Oral Science, Yonsei University College of Dentistry; Professor Sungjae Hwang, College of Artificial Intelligence Convergence, Yonsei University; Researcher Dayoon Joo, Master's Program, Graduate School of Artificial Intelligence Convergence, Yonsei University Health System

The research findings were published in the international journal 'npj Digital Medicine' (Impact Factor 15.1).

TMJ disorders are major oral and maxillofacial conditions that affect the function of the jaw joint, which is used for chewing and speaking. Many patients visit medical facilities due to symptoms such as TMJ pain, difficulty opening the mouth, or joint noises.

However, abnormalities in the position of the TMJ, internal inflammation, or excessive accumulation of synovial fluid within the joint can only be accurately identified through MRI.

The challenge is that the high cost and limited accessibility of MRI examinations make it difficult to perform MRIs on all patients. In actual clinical settings, the decision to conduct an MRI is often based on the medical team's experience, which can result in some patients undergoing unnecessary tests, while others may experience delays in receiving needed examinations.

To address these issues, the research team developed an AI model that predicts the likelihood of TMJ abnormalities on MRI by utilizing panoramic X-ray images, which are the most commonly taken in dental clinics, along with patient clinical information.

The team analyzed 1,355 patients who visited the Department of Oral Medicine at Yonsei University Dental Hospital from January 2021 to December 2023 with TMJ symptoms and underwent both panoramic X-ray and MRI scans. A total of 2,710 TMJs were included in the analysis, and the AI model was trained and validated based on MRI readings.

Additionally, the model incorporated panoramic X-ray images taken with the mouth both closed and open, enabling the AI to reflect changes in joint position according to TMJ movement. The system was designed to focus on the condylar region, which is clinically significant for diagnosis, and it also integrated clinical information such as limited mouth opening and joint noises.

As a result, the AI model achieved an accuracy metric (AUC) of 0.86 in cross-validation and 0.84 in independent testing. The AUC (Area Under the Curve) indicates higher prediction accuracy as it approaches 1. This demonstrates that the likelihood of TMJ abnormalities detectable on MRI can be meaningfully predicted using only panoramic X-ray images and basic clinical information, without the need for three-dimensional MRI scans.

In particular, it was also confirmed that the AI model enables visual identification of the specific regions in the images used for decision-making, making it easier for clinicians to interpret and utilize the results.

Professor Yeonjeong Park stated, "This study is significant not because it replaces MRI, but because it presents a new clinical system for pre-selecting patients who truly need MRI, which could reduce unnecessary advanced imaging and minimize diagnostic delays."

Professor Hyojung Jung added, "This research proposes a new approach to TMJ diagnosis that connects screening and precision testing, moving beyond conventional diagnosis focused solely on advanced imaging. Since panoramic X-ray imaging is performed in most dental clinics, the model has strong potential for real-world clinical application."

This study was co-first authored by Professor Hyojung Jung of the Oral Science Research Institute at Yonsei University College of Dentistry and Master’s student Dayoon Joo at the Graduate School of AI Convergence at Yonsei University, with Professor Yeonjeong Park serving as the corresponding author and leading the research.

Meanwhile, the research team received the Best Oral Presentation Award at the 2024 Annual Scientific Meeting of the American Academy of Orofacial Pain (AAOP), and also presented their findings at the 2025 European Pain Federation Congress (EFIC).

This content was produced with the assistance of AI translation services.