Related Experiment Video
Updated: Aug 10, 2026
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
Automated Differentiation of Oral Red-White Lesions: An Interpretable Deep Learning Approach Combining Ensemble
Mahsa Koochaki1, Amirreza Mousavie2, Maryam Basirat1
1Department of Oral and Maxillofacial Medicine, School of Dentistry, Guilan University of Medical Sciences, Rasht, Iran.
Clinical and Experimental Dental Research
|July 21, 2026
Summary
A new deep learning model accurately classifies oral lesions like Leukoplakia and Oral Lichen Planus. This AI tool shows promise for early detection of oral potentially malignant disorders (OPMDs).
Area of Science:
- Oral pathology and digital diagnostics.
- Artificial intelligence in medical imaging.
Background:
- Oral Potentially Malignant Disorders (OPMDs) pose a significant cancer risk.
- Differentiating OPMDs (e.g., Leukoplakia, Erythroplakia) from inflammatory conditions (e.g., Oral Lichen Planus, Candidiasis) is clinically challenging due to visual similarities.
Purpose of the Study:
- To develop and evaluate an interpretable deep learning framework for automated multi-class classification of pre-localized oral lesions.
- To improve the accuracy and efficiency of diagnosing oral lesions.
Main Methods:
- Utilized a dataset of 705 oral lesion images across five categories.
- Developed a Multi-Architecture Weighted Ensemble Framework integrating ResNet-50, Xception, and EfficientNet-B0.
- Employed stratified patient-level splitting, Grad-CAM for interpretability, and Decision Curve Analysis (DCA) for clinical utility assessment.
Main Results:
- The ensemble model achieved 91.2% accuracy and a 90.8% macro-averaged F1-score.
- Significantly improved detection of Oral Lichen Planus (OLP), distinguishing it from Leukoplakia.
- Grad-CAM confirmed the model's focus on relevant pathological features, and DCA indicated potential clinical benefit.
Conclusions:
- The weighted ensemble framework demonstrates high retrospective accuracy and provides visual explanations for oral lesion classification.
- This study serves as a preliminary proof-of-concept.
- Extensive external validation and prospective testing are required to confirm clinical utility in primary care settings.
