HMI-LUSC: A Histological Hyperspectral Imaging Dataset for Lung Squamous Cell Carcinoma

The dataset comprises 62 hyperspectral images collected from 10 patients, each accompanied by image-level tumor annotations based on pathological gold standards and refined into cell-level labels via a semi-automated process. The images cover a spectral range of 450–750 nm across 61 bands, captured at 10X magnification with an image resolution of 3088 × 2064 pixels. The dataset provides substantial utility for both research and clinical applications.Folders are grouped by patient, with each folder named Pi, where i denotes the patient index. Within each patient folder, there are subfolders containing multiple hyperspectral image regions of interest (ROIs) as well as a whole-slide scan. Each ROI subfolder is named LUSC_ROI_j, where j denotes the ROI index within the corresponding slide. The full-slide image is provided as Pi_Slide_Scan_40×.tif, where 40× indicates the scanning magnification. Each ROI folder contains the hyperspectral image cube of the selected tissue region, the corresponding white and dark reference images required for calibration, image-level annotation masks, cell-level annotation masks, and an RGB image synthesized from the hyperspectral cube using the CIE 1931 color-matching functions. The hyperspectral image cube and calibration references are stored in ENVI format, which is the standard format for hyperspectral data storage. This format consists of a binary raster file paired with an ASCII header file (with the .hdr extension).