Abstract
Hyperspectral imaging (HSI) is an emerging imaging modality for histopathological applications. However, annotations on RGB histological images are usually used as the reference standard. To use and validate hyperspectral data, it is critical to correlate each hyperspectral image with the corresponding region in a whole-slide image and retrieve accurate tissue label. In this work, we developed a fully automated processing pipeline for hyperspectral histological images. Given a high-resolution digitized whole-slide histological image, the annotation, and a hyperspectral image tile of any region in the slide, the proposed method can locate the HSI tile region in the whole-slide image, crop the RGB image tile and tissue label, and align the RGB tile and tissue label to the HSI tile. With our proposed processing pipeline, we collected and formed a dataset with over 350 whole-slide hyperspectral histological images of human head and neck cancers. The proposed processing pipeline can serve as a general tool for fast and automated hyperspectral histological images, thus facilitating the adaptation of hyperspectral imaging in digital pathology to assist automatic histology diagnosis.