Machine Learning and AI-Powered Diagnostics for Early Detection of Oral Pre-Malignant and Malignant Lesions

Department of Computer Engineering, KSV University, Gandhinagar, India

^* Corresponding author: vbc01@ganpatuniversity.ac.in

Abstract

Oral squamous cell carcinoma (OSCC) accounts for over 90% of oral malignancies, with late detection contributing to high mortality rates (5-year survival rate: ~60%). Traditional diagnostic methods, such as visual examination and biopsy, are subjective, invasive, and lack sensitivity for early-stage lesions. This review evaluates the transformative potential of artificial intelligence (AI) in improving early detection of oral pre-malignant and malignant conditions. A systematic analysis of 120+ studies (2018–2023) reveals that AI models, particularly convolutional neural networks (CNNs), achieve an average accuracy of 92.4% in classifying oral lesions, outperforming conventional methods. Key challenges include dataset heterogeneity, model generalizability, and integration into clinical workflows. This paper synthesizes advancements in AI-driven diagnostics, critiques limitations of existing research, and proposes frameworks for scalable, equitable deployment. Contributions include a meta-analysis of performance metrics, identification of clinical validation gaps, and recommendations for federated learning and explainable AI (XAI) adoption.