A multi-class, image-level labeled dataset of baseline non-contrast head computed tomography for stroke triage

This dataset consists of 7,965 baseline non-contrast head computed tomography (NCCT) images retrospectively collected from 99 patients at the Complejo Asistencial Universitario de León (Spain). Each axial image was independently reviewed and assigned one of four image-level labels: anterior ischemic stroke, posterior ischemic stroke, hemorrhagic stroke, or no stroke. The images were exported directly from the hospital imaging system without preprocessing, preserving the variability of routine clinical practice, including differences in spatial resolution. The dataset is designed to support the development, validation, and benchmarking of artificial intelligence methods for automated stroke detection and classification under real-world clinical conditions.