Brain-Dwi-Adc: An Anonymized Vision-Language Segmentation Dataset Adapted from ISLES-2022

Brain-Dwi-Adc is a processed vision-language medical image segmentation dataset adapted from the public ISLES-2022 training dataset. It contains paired DWI/ADC slice-level samples, binary ischemic stroke lesion masks, anonymized physician-written textual reports, and patient-level train/validation/test splits. The textual reports follow a fixed pathology-number-location template and do not contain pixel-level mask coordinates. The dataset is intended to support reproducible evaluation of medical image segmentation and quantitative measurement reliability.