Related Experiment Video
Updated: May 5, 2026
05:42
Author Spotlight: Unlocking the Mysteries of Oral Potential Malignancies
Published on: August 11, 2023
1.6K
A Multimodal Large Language Model Framework for Clinical Subtyping and Malignant Transformation Risk Prediction in
Ali Robaian1, Fatma E A Hassanein2, Mohamed Talha Hassan3
1Conservative Dental Sciences Department, College of Dentistry, Prince Sattam bin Abdulaziz University, Al-Kharj, Saudi Arabia.
International Dental Journal
|January 1, 2026
Summary
A new AI model demonstrates expert-level accuracy in classifying oral lichen planus (OLP), oral lichenoid lesions (OLL), and related oral cancers. This AI tool can also reliably identify high-risk malignant transformation features, aiding oral medicine specialists.
Area of Science:
- Oral medicine and digital health advancements.
- Artificial intelligence in medical diagnostics.
- Clinical pathology and oncology.
Background:
- Oral lichen planus (OLP), oral lichenoid lesions (OLL), and squamous cell carcinoma on a lichenoid background (SCC-over-LP/LLP) present overlapping clinical features.
- This clinical overlap can delay the recognition of malignant transformation.
Purpose of the Study:
- To compare the diagnostic accuracy of a multimodal large language model (ChatGPT-5) against oral medicine specialists.
- To evaluate the AI's ability in tripartite classification (OLP/OLL/SCC-over-LP/LLP) and flagging malignant risk.
Main Methods:
- A cross-sectional, paired diagnostic accuracy study was conducted.
- 262 retrospective, anonymized cases (OLP, OLL, SCC-over-LP/LLP) were independently evaluated by ChatGPT-5 and oral medicine specialists.
- A reference standard panel of OM experts established diagnoses using full clinical data and histopathology.
Main Results:
- Overall diagnostic accuracy was comparable between ChatGPT-5 (84.7%) and OM specialists (85.5%).
- High sensitivity was observed for OLP (0.99) and SCC-over-LP/LLP (0.85), with OLL sensitivity at 0.70 and specificity at 1.00.
- ChatGPT-5 correctly identified malignant-risk features in 88% of SCC-over-LP/LLP cases, comparable to OM specialists (92%).
Conclusions:
- A multimodal large language model achieved expert-level accuracy for classifying OLP, OLL, and SCC-over-LP/LLP.
- The AI reliably flagged malignant transformation risk, supporting its use as an adjunctive decision-support tool in oral medicine.
