Non-Invasive Oral Cancer Detection Using Hyperspectral Imaging and Advanced Spectral Unmixing Models

Oral cancer is a significant global health concern, often leading to high mortality rates due to late-stage diagnosis and the lack of effective early detection methods. Despite advances in medical science, the absence of reliable early diagnostic tools remains a critical challenge. Hyperspectral imaging (HSI) has emerged as a powerful noninvasive technology, capturing detailed spectral information across a wide range of wavelengths. This allows for accurate differentiation between cancerous and healthy tissues, improving early detection and enhancing treatment outcomes. In this study, we propose the use of HSI for early oral cancer diagnosis. To address the scarcity of labeled data, we developed a synthetic hyperspectral dataset that includes spectral signatures of both normal and cancerous tissues. The dataset was generated using a bilinear mixing model, with key spectral features extracted through Vertex Component Analysis (VCA) and abundances computed using Non-Negative Least Squares (NNLS). The model's performance was evaluated using Spectral angle distance (SAD) and Root mean square error (RMSE) metrics. Our findings demonstrate that HSI significantly improves the accuracy of early oral cancer detection, outperforming traditional methods. This work highlights the potential of advanced imaging technologies in revolutionizing cancer diagnosis, offering a robust framework for non-invasive detection and showcasing the effectiveness of synthetic datasets in medical imaging research.