Hyperspectral Mapping of Oral and Pharyngeal Cancer: Estimation of Tumor-Normal Margin Interface using Machine Learning

As hyperspectral imaging (HSI) becomes a more accessible technology, there is increasing interest in applying the technology to clinical medicine. An HSI method that was recently developed at PARC which relies on polarization interferometry through a liquid-crystal variable retarder, has shown promising results in terms of demonstrating the compactness and low-cost potential necessary to increase the wide adoption of HSI for medical applications. The goal of the current study is to discern differences between clinically significant tissues using HSI; we believe the information within hyperspectral images may aid physicians and surgeons in clinical diagnosis and treatment beyond what is possible with standard color vision and imaging. In this study, we have mapped differences in the soft-tissue of the head and neck by imaging residual surgical specimens with the PARC hyperspectral camera. We correlate hyperspectral images with histopathology and color photography, and we apply machine learning to the hyperspectral images to attempt to make multiple clinically significant distinctions: nerves from blood vessel, tumor from normal tissues, and parathyroid glands from fat. Using HSI in this way to guide diagnostic biopsy and surgical navigation would significantly reduce the time, cost and morbidity of evaluation and treatment of many human diseases.