Artificial Intelligence-Based Diagnosis of Oral Lichen Planus Using Deep Convolutional Neural Networks

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Oral lichen planus (OLP) is a chronic inflammatory condition affecting oral mucosa.
• Accurate clinical differentiation between OLP and non-OLP lesions is crucial for appropriate management.
• Histopathological confirmation is the gold standard but can be invasive.

Purpose of the Study:

• To evaluate the efficacy of artificial intelligence (AI), specifically convolutional neural networks (CNNs), in distinguishing OLP from non-OLP using clinical photographs.
• To compare the performance of different CNN models for OLP diagnosis.

Main Methods:

• A dataset of 609 OLP and 480 non-OLP clinical photographs, histopathologically confirmed, was utilized.
• Data augmentation techniques were applied to enhance the training dataset.
• Multiple CNN models were trained and validated, with performance assessed using accuracy, sensitivity, specificity, and F1-score.
• Gradient-weighted class activation mapping (Grad-CAM) was employed for model interpretability.

Main Results:

• All evaluated CNN models successfully diagnosed OLP and non-OLP lesions from clinical images.
• The Xception model exhibited the highest performance metrics, including accuracy and F1-score.
• The study demonstrated CNN model accuracies ranging from 82% to 88%.

Conclusions:

• CNN models show significant potential for the non-invasive diagnosis of OLP from clinical photographs.
• The Xception model is a promising AI tool for differentiating OLP from other oral lesions.
• Further validation with larger datasets is warranted to integrate this technology into clinical practice.