Related Experiment Video
Updated: May 13, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Utilizing deep learning for automated detection of oral lesions: A multicenter study
Yong-Jin Ye1, Ying Han2, Yang Liu2
1Division of Mechanics, Beijing Computational Science Research Center, Building 9, East Zone, No.10 East Xibeiwang Road, Haidian District, Beijing 100193, China.
Oral Oncology
|June 4, 2024
Summary
A new YOLOX-based AI model accurately detects oral lesions like oral lichen planus (OLP), oral leukoplakia (OLK), and oral squamous cell carcinoma (OSCC). This AI tool outperforms experienced specialists and aids dentists in diagnosis via a mobile app.
Area of Science:
- Artificial Intelligence in Medicine
- Medical Imaging Analysis
- Oral Pathology Diagnostics
Background:
- Oral lesions, including oral lichen planus (OLP), oral leukoplakia (OLK), and oral squamous cell carcinoma (OSCC), require accurate and timely diagnosis.
- Early detection of oral lesions is crucial for effective treatment and improved patient outcomes.
- Existing diagnostic methods can be subjective and may benefit from objective, AI-driven support.
Purpose of the Study:
- To develop and evaluate a YOLOX-based convolutional neural network (CNN) for the precise detection of multiple oral lesions (OLP, OLK, OSCC).
- To assess the diagnostic performance of the AI model against experienced clinicians.
- To integrate the AI model into a mobile application for practical clinical use.
Main Methods:
- A dataset of 1419 patient photos was collected for model development and evaluation.
- A comparative analysis was performed between the AI model and a senior group of experts.
- A multicenter evaluation involving 24 participants from 14 centers assessed the model's diagnostic aid capabilities.
- The model was integrated into a mobile application for rapid diagnostics.
Main Results:
- The YOLOX-based model outperformed a senior group of experts in macro-average recall (85% vs 77.5%), precision (87.02% vs 80.29%), and specificity (95% vs 92.5%).
- In multicenter evaluations, model-assisted diagnosis significantly improved performance across dental, general, and community hospitals, matching or exceeding senior group benchmarks.
- The AI model demonstrated high proficiency in detecting oral lesions, comparable to or better than experienced specialists.
Conclusions:
- The developed YOLOX-based AI model shows high proficiency in detecting oral lesions, outperforming experienced specialists.
- The model serves as a valuable diagnostic aid for general dentists and specialists, enhancing diagnostic accuracy.
- Integration into a mobile application enables swift and precise oral lesion diagnostic procedures.
