melisklc0/stroke-classification

1---2license: cc-by-4.03task_categories:4- image-classification5tags:6- medical7- stroke8- ct-scan9- data-augmentation10language:11- en12pretty_name: Stroke Classification Dataset (CT Scans)13---14 15# Stroke Classification: Pre-processed Brain CT Dataset16 17### Related resources18 19- [Live demo](https://huggingface.co/spaces/melisklc0/stroke-classification)20- [Training code](https://github.com/melisklc0/Stroke-Classification)21- [Distilled model](https://huggingface.co/melisklc0/efficientnet-b0-stroke-distilled)22 23## Summary24 25- **Task:** Binary CT image classification (`No-Stroke` = 0, `Stroke` = 1)26- **Dataset Size:** ~16.5k images per fold, ~49.5k files across folds on disk. ~3 GB total 27- **Structure:** 3-fold cross-validation (`Fold1`, `Fold2`, `Fold3`), each containing `train/` and `test/` splits28- **Labeling Strategy:** Original classes (`No Stroke`, `Bleeding`, `Ischemia`) were converted into a binary setup by merging `Bleeding` and `Ischemia` into a single `Stroke` class for emergency triage29 30## Preprocessing Pipeline31 321. Merged `Bleeding` and `Ischemia` into a unified `Stroke` class.332. Applied stratified 3-fold cross-validation with separate `train/` and `test/` sets.343. Balanced test sets to ~750 images per class per fold.354. Applied data augmentation on training sets:36   - Rotation (±10°)37   - Zooming38   - Translation39   - Horizontal flipping40   resulting in ~7,500 images per class per fold.41  42## Repository Structure43 44The dataset is organized to support cross-validation training workflows and external generalization testing seamlessly:45 46* **`dataset/`**: Contains all pre-processed, augmented, and validation CT scan images.47  * `External_Dataset/`: An independent dataset (sourced from Kaggle) used purely for external validation and testing model generalization across different scanner calibrations.48  * `Fold1/`: Training and validation split 1.49  * `Fold2/`: Training and validation split 2.50  * `Fold3/`: Training and validation split 3.51 52*Inside each fold and the external dataset, the images are categorized into two sub-folders: `Stroke` and `No-Stroke`.*   53*Train on `FoldN/train`, validate on `FoldN/test`. Rotate folds for 3-fold cross-validation.*54 55## Source Data & Attribution56 57The data utilized to build this augmented dataset originates from two primary sources. In compliance with open data policies, please find the attributions below:58 591. **Primary Dataset (Turkish Ministry of Health):** The raw CT scans utilized to build the core augmented training folds were obtained from the Open Data Portal of the Republic of Turkey Ministry of Health. 60   * *Original Source:* [Open Data Portal of the Turkish Ministry of Health](https://acikveri.saglik.gov.tr/Home/DataSetDetail/1)612. **External Validation Dataset (Kaggle):** An external dataset was used purely to test the robust generalization of the trained models across different scanner calibrations.62   * *Original Source:* [Head CT Hemorrhage Dataset by Felipe Kitamura](https://www.kaggle.com/datasets/felipekitamura/head-ct-hemorrhage)63 64## Licensing & Usage65 66This processed dataset is distributed under the **Creative Commons Attribution 4.0 International (CC BY 4.0)** license. 67 68You are free to share and adapt this material for any purpose, even commercially, under the following terms:69* **Attribution:** You must give appropriate credit to the original data sources (mentioned above) and indicate that this is an augmented, pre-processed version of the original files.70 71*Disclaimer: This dataset is provided for research and educational purposes. It is not intended to replace professional medical diagnosis or serve as a standalone clinical tool.*72 73## Authors74* **Melis Kılıç**75* **Esra Koç**76 77**Advisor:** Assoc. Prof. Dr. Kali Gürkahran