Machine Learning Algorithms for Early Detection of Oral Squamous Cell Carcinoma from Clinical Photographs: A Diagnostic Accuracy Study

Abstract

Oral squamous cell carcinoma (OSCC) represents approximately 90% of all oral malignancies and continues to pose significant diagnostic challenges in clinical practice, particularly in resource-limited settings where access to specialized histopathological services remains restricted. This diagnostic accuracy study investigates the application of machine learning algorithms for early detection of OSCC using clinical photographs, addressing the critical need for accessible, non-invasive screening tools. We systematically evaluated multiple machine learning architectures, including convolutional neural networks, support vector machines, and ensemble methods, using a comprehensive dataset of clinical photographs from patients with confirmed OSCC diagnoses and healthy controls. The study demonstrates that deep learning approaches, particularly transfer learning models based on ResNet and DenseNet architectures, achieved diagnostic accuracy exceeding 92% with sensitivity values ranging from 88% to 94% and specificity values between 89% and 95%. These findings suggest that machine learning-based diagnostic systems can potentially serve as effective screening tools in primary healthcare settings, enabling earlier detection and improved patient outcomes. The implementation of such systems could significantly reduce the diagnostic burden on specialized healthcare facilities while maintaining high diagnostic accuracy standards. This research contributes to the growing body of evidence supporting the integration of artificial intelligence technologies in oral cancer detection pathways, offering practical solutions for addressing the global challenge of late-stage OSCC diagnoses.