A comparative study of computer vision models for oral cancer detection from oral photographs

Early detection of oral cavity cancers is critical for improving patient survival rates and treatment efficacy. In this context, this study evaluates the potential of computer vision models as diagnostic tools for identifying cancerous lesions from oral cavity photographs.

A comparative study was conducted using modern deep learning-based object detection models to detect lesions, such as squamous cell carcinoma, from a biopsy-proven dataset of oral lesion photography. A comprehensive workflow was developed to evaluate and compare these models, including fine-tuning, hyperparameters optimization, and performances assessment using precision, sensitivity, and specificity metrics. Additionally, the various parameters influencing the model were systematically measured and analyzed, distinguishing it from previous studies and providing a novel and comprehensive assessment of detection methodologies in the field.

The studied models demonstrated high performance in a single-class detection setting, effectively localizing oral lesions with promising precision. However, in a two-class detection setting, distinguishing between malignant and benign lesions proved challenging, indicating a need for further refinement.

This study underscores the potential of artificial intelligence in aiding early detection of oral cancers while identifying areas for improvement (notably: small lesion detection and distinguishing between malignant and benign). These findings provide a foundation for advancing medical computer vision tools to support early diagnosis and improve patient outcomes.