AI-Driven Texture Analysis for Differentiation of Benign and Malignant Oral Lesions

Abstract

Background: Delayed or inaccurate clinical differentiation of benign from malignant oral mucosal lesions remains a critical diagnostic challenge, contributing to late-stage detection and poor survival outcomes in oral squamous cell carcinoma.

Objective: To develop and validate an artificial intelligence (AI)-driven texture analysis model for differentiating benign and malignant oral lesions using digitized clinical photographic images.

Methodology: One hundred and eighty histopathologically confirmed oral lesions (87 benign, 93 malignant) were photographed under standardized lighting conditions; grey-level co-occurrence matrix (GLCM), grey-level run-length matrix (GLRLM), and fractal dimension features were extracted and evaluated across five machine learning and deep learning classifiers.

Results: The convolutional neural network (CNNs) and GLCM fusion model achieved the highest diagnostic accuracy of 94.4%, sensitivity of 95.7%, specificity of 93.1%, and area under the curve of 0.978, significantly outperforming both individual algorithms and expert clinician assessment (accuracy 79.4%).

Conclusion: AI-driven texture analysis demonstrates high discriminatory power for distinguishing malignant from benign oral lesions and holds promise as a non-invasive, adjunctive screening tool in resource-limited clinical settings