INTELKAMAT: An Anonymised Cervical Colposcopy Image Dataset with Per-Patient Clinical Annotations from a Bulgarian Tertiary Centre
Open Access
Abstract
Background: Public colposcopy image datasets are scarce, geographically narrow, and rarely accompanied by structured clinical annotations. We present INTELKAMAT, an anonymised cervical colposcopy image dataset of 934 RGB JPEG images at 1280 × 960 pixels from 30 women examined at the Medical University Hospital of Pleven, Bulgaria (2022–2024). Methods: Images were prospectively acquired during standard colposcopic examination and retrospectively curated for release. All images were anonymised via an automated, reproducible pipeline (Python 3.10, Pillow, piexif), with patient folders renamed with opaque IDs (MUP_001..MUP_030), filenames rewritten to MUP_xxx_IMG_yy.jpg, and EXIF metadata stripped. The reverse mapping is retained by the Medical University of Pleven and never released, so the data are pseudonymised at the source and the distributed bundle carries no direct identifiers. Clinical annotations were provided by the clinician co-authors at the patient level and translated from the original Bulgarian shorthand into an English controlled vocabulary. Results: All 30 patients (100%) are annotated with transformation-zone type per the 2011 IFCPC nomenclature (Type 1: 14, Type 2: 7, Type 3: 9), non-neoplastic findings (Ovula Nabothi, zona glandularia, Ectopia, Polyp, Vessels, Leukoplakia), and view/examination modifiers. Twenty-four doctor-annotated reference photos accompany the images, each resolved to the exact frame it annotates and flagged per image. Conclusions: The dataset broadens the geographic representation of open cervical colposcopy data. Transformation-zone type is a morphological descriptor rather than a diagnosis, and the release contains no histopathological, cytological or HPV results; INTELKAMAT therefore supports cervical-region segmentation, view and image-quality research, and transformation-zone work on suitably filtered views. Given the 30-patient cohort, it is best used as an external validation cohort or within patient-stratified cross-validation rather than as a standalone training set for precancer detection.Keywords
Funding Information
- European Commission (BG16RFPR002-1.014-0002-С001)
- Bulgarian National Bank (BG-175467353-2022-04-0257-C01)