A machine-learning algorithm for the reliable identification of oral lichen planus.

Area of Science:

• Oral pathology
• Digital pathology
• Artificial intelligence in medicine

Background:

• Oral lichen planus (OLP) is a common oral disorder with features overlapping other lichenoid lesions, causing diagnostic challenges.
• Inter-observer disagreement in OLP diagnosis hinders understanding of its pathogenesis and malignant potential.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for accurate OLP diagnosis.
• To differentiate OLP from oral lichenoid lesions (OLLs) and oral epithelial dysplasia (OED) with lichenoid host response using AI.

Main Methods:

• An artificial neural network was trained using digitized hematoxylin and eosin slides to quantify inflammatory cells.
• The AI model was trained on 24 regions of interest and validated on 130 patient cases.
• The model analyzed mononuclear cells and granulocytes within inflammatory infiltrates.

Main Results:

• AI demonstrated a statistically significant difference in inflammatory cell counts between OLP and other lichenoid conditions (p < 0.0005).
• The AI model achieved high accuracy in detecting OLP, with an area under the curve of 0.982 for inflammatory cells and 0.988 for mononuclear cells.
• A cut-off for mononuclear cells achieved 100% sensitivity and 94.62% accuracy in distinguishing OLP from other conditions.

Conclusions:

• AI offers a robust method to enhance diagnostic accuracy for oral lichen planus.
• This AI approach aids anatomical pathologists in diagnosing OLP based on disease pathogenesis features.