Refined Subset of the Intel & MobileODT Cervical Cancer Screening Dataset: Manual Curation and ROI Preprocessing for Colposcopic Image Classification

This dataset constitutes a manually reviewed and preprocessed subset of the Intel & MobileODT Cervical Cancer Screening dataset, originally released through a Kaggle competition. The original collection was filtered from 1,481 to 766 high-quality colposcopic images through a rigorous curation process that excluded samples with severe glare, occlusions, and poor focus, while performing precise cropping of the Region of Interest (ROI) centered on the cervical transformation zone. The subset preserves the original three-class structure for cervix type classification: Type 1 (n=185), Type 2 (n=408), and Type 3 (n=173). A cryptographic hash analysis using the MD5 algorithm confirmed a 0% duplication rate across partitions, ensuring full sample independence. This resource is intended to support reproducibility of the results reported in the associated publication and to provide the research community with a clean, clinically reliable benchmark for automated cervical screening tasks.