Related Experiment Video
Updated: Jan 6, 2026
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
3.1K
SegORG: Report Generation of Oral Potentially Malignant Disorders Image Based on Lesion Segmentation-Enhanced LLM
Rui Zhang1,2,3, Peng Huang1, Tingting Ding1
1Stomatology Hospital, School of Stomatology, Zhejiang University School of Medicine, Zhejiang Provincial Clinical Research Center for Oral Diseases, Zhejiang Key Laboratory of Oral Biomedical, Hangzhou, China.
Oral Diseases
|November 28, 2025
Summary
An automated system, SegORG, generates standardized reports for oral potentially malignant disorders (OPMDs) from images. This AI model improves documentation efficiency and aids early intervention for oral lesions.
Area of Science:
- Oral medicine
- Artificial intelligence
- Medical imaging analysis
Background:
- Oral potentially malignant disorders (OPMDs) require accurate and timely documentation.
- Manual report generation is time-consuming and can lead to inconsistencies.
- Automated systems can streamline OPMD reporting, aiding early diagnosis and treatment.
Purpose of the Study:
- To develop an automated system (SegORG) for generating standardized OPMD reports from white-light images.
- To reduce documentation workload and facilitate early intervention and lesion monitoring.
Main Methods:
- The SegORG model utilizes SegFormer for lesion segmentation and a visual encoder for feature extraction.
- Visual embeddings are mapped to a pre-trained large language model (LLM) space using a visual mapper.
- The Qwen2.5-7B model generates structured reports, enhanced by text augmentation.
Main Results:
- SegORG achieved strong performance metrics (BLEU-4: 0.291, ROUGE-L: 0.517, CIDEr: 0.578).
- The model demonstrated a clinical diagnostic F1-score of 0.695 and a median expert rating of 4/5.
- SegORG outperformed conventional models and general-purpose multimodal LLMs.
Conclusions:
- SegORG provides an effective automated pathway for OPMD report generation through enhanced visual feature extraction and text alignment.
- Further multicenter trials are necessary to validate the generalizability of the SegORG model.
