HyCervix

This dataset contains in-vivo hyperspectral (HS) images of the human cervix acquired during routine clinical colposcopic examinations, designed to support research on non-invasive detection of precancerous and cancerous cervical lesions. The dataset comprises 77 hyperspectral cubes from 77 patients, captured using a clinical colposcope equipped with a hyperspectral imaging system operating in the 400–900 nm spectral range. Each HS image provides both spatial and spectral information, enabling detailed tissue characterization. All images are accompanied by pixel-level annotations corresponding to clinically relevant tissue classes, including: Ectocervix Endocervix Cervical intraepithelial neoplasia (CIN) lesions Invasive carcinoma Annotations were generated by expert colposcopic assessment and validated using cytology and/or histopathological biopsy results, ensuring clinically grounded ground truth. In addition to imaging data, the dataset includes anonymized clinical metadata for each patient, such as demographic information, colposcopic findings, biopsy outcomes, and final clinical diagnoses. This combination of hyperspectral imaging data and clinical annotations makes the dataset suitable for tasks including spectral analysis, tissue segmentation, lesion classification, and development of computer-aided diagnostic algorithms. The code associated with this dataset is publicly available to facilitate further research. The repository includes scripts for processing spectral signatures from the different tissue annotations, enabling straightforward use of this dataset in machine learning and image analysis pipelines. The repository can be accessed at: https://github.com/carlosvegagc/HyCervix