Blog

AI detection of oral lichen planus from clinical photographs

A Deep Learning Algorithm for Classification of Oral Lichen Planus Lesions from Photographic Images: A Retrospective Study

Authors: Gaye Keser, İbrahim Şevki Bayrakdar, Filiz Namdar Pekiner, Özer Çelik, Kaan Orhan

Summary
This retrospective study developed a deep learning model to classify oral lichen planus (OLP) lesions from clinical photographic images of the buccal mucosa.
A total of 137 intraoral photographs were retrospectively collected using the CranioCatch platform (Eskisehir, Turkey): 65 healthy mucosa and 72 OLP lesions, histopathologically confirmed.
Images were re-verified by experts in Oral Medicine and Maxillofacial Radiology and divided into training (51/58), validation (7/7), and test (7/7) sets for healthy and OLP classes.

The model used the Google Inception V3 architecture implemented with TensorFlow, trained over 2,000 epochs on 1024×1024 images.
On the test set, the algorithm achieved 100 % accuracy for both healthy and diseased mucosa.

This study is the first Turkish investigation to apply deep learning for OLP classification using photographic images and demonstrates the strong diagnostic potential of CNNs for mucosal-lesion analysis.

DOI: https://doi.org/10.1016/j.jormas.2022.08.007

Key Words

Oral Lichen Planus, Photographic Images, Deep Learning, Inception V3, TensorFlow, Image Classification, CranioCatch, Computer Vision, Oral Medicine, Artificial Intelligence in Dentistry

Extracted Data

Year: 2023
Modality: Clinical Photographs (Intraoral – Buccal Mucosa)
Dataset: 137 images (65 healthy / 72 OLP)
Dataset Split: Training = 51 + 58; Validation = 7 + 7; Test = 7 + 7
Network Architecture: GoogleNet Inception V3 (CNN, TensorFlow implementation)
Metrics: Accuracy = 100 % (test set)
AP – Strategy: Retrospective image collection + expert re-verification + histopathologic confirmation
AP – Professional Qty: 3 oral medicine and radiology specialists
AP – Supervisor Presence: Yes (Marmara University and Eskisehir Osmangazi University)
AP – Experience Level: Experienced faculty members
AP – Expertise Area: Oral Medicine, Maxillofacial Radiology, Medical AI
AP – Tool or System: CranioCatch Annotation Platform + TensorFlow Pipeline
Task: Classification (Healthy vs OLP Lesion)
Project Objective: To develop a deep learning approach for detecting oral lichen planus lesions in clinical photographs and evaluate its diagnostic performance

Clinical Relevance

Clinical importance: Oral lichen planus is a chronic inflammatory mucosal disease that can be challenging to diagnose clinically.
Innovation: The proposed AI model achieved 100 % accuracy, highlighting the feasibility of using standard intraoral photographs for lesion classification without radiation exposure.
Practical impact: Such AI tools may assist clinicians in screening and early recognition of oral mucosal pathologies, reducing diagnostic variability and workload in dental practice.

YouTube Video: https://www.dentalai.pt/en/blog/ai-detection-of-oral-lichen-planus-from-clinical-photographs