--- dataset_info: features: - name: image_id dtype: string - name: split dtype: string - name: image dtype: image - name: mask dtype: image - name: label dtype: int32 - name: sequence dtype: string - name: source_file dtype: string splits: - name: train num_bytes: 103678331 num_examples: 1589 download_size: 115728833 dataset_size: 103678331 configs: - config_name: default data_files: - split: train path: data/train-* license: cc-by-4.0 language: - en tags: - mri - imaging - medical - brain - tumor - segmentation - classification - pretraining task_categories: - image-classification - image-segmentation pretty_name: BRISC 2025 (Axial) size_categories: - 1K____t1`). 3. Images were converted to single-channel grayscale PNG at their original resolution. 4. Masks come from BRISC's `segmentation_task` folder and were binarized to {0, 255}. Masks that didn't match the image size were resized with nearest-neighbour interpolation. 5. **No-tumor** slices have no mask in BRISC, so they get an all-zero mask. 6. The original train/test split is kept in the `split` column. ## Data Fields | Field | Type | Description | |---|---|---| | `image_id` | string | `_`, e.g. `train_00001` | | `split` | string | Original BRISC split: `train` or `test` | | `image` | image | Axial T1-weighted contrast-enhanced slice (grayscale PNG) | | `mask` | image | Binary tumor mask (0 = background, 255 = tumor). All zeros for no-tumor slices | | `label` | int32 | Tumor class (see below) | | `sequence` | string | MRI sequence (`t1`) | | `source_file` | string | Original BRISC filename (without extension) | ### Labels The labels follow the same coding as [chehablab/JCBTD](https://huggingface.co/datasets/chehablab/JCBTD), plus a no-tumor class: | Label | Class | |---|---| | 0 | No tumor | | 1 | Meningioma | | 2 | Glioma | | 3 | Pituitary tumor | ## Usage ```python from datasets import load_dataset ds = load_dataset("chehablab/BRISC2025", split="train") train = ds.filter(lambda x: x["split"] == "train") test = ds.filter(lambda x: x["split"] == "test") sample = train[0] sample["image"], sample["mask"], sample["label"] ``` ## Notes and Limitations - **No patient IDs.** BRISC doesn't release patient identifiers, so a patient-level split isn't possible. Several slices may come from the same patient. - **Overlap with JCBTD.** BRISC images were selected from the Kaggle "Brain Tumor MRI Dataset", which includes the Figshare brain tumor dataset by Jun Cheng (JCBTD). Some slices may appear in both datasets. Deduplicate (for example by image hash) before combining them or using one to evaluate a model trained on the other. - Slices are 2D JPEG-derived images, not original DICOM/NIfTI volumes, so they have no voxel spacing or intensity calibration. ## License Released under the license of the original BRISC dataset (CC BY 4.0). Please check the [original dataset page](https://www.kaggle.com/datasets/briscdataset/brisc2025) for its terms. ## Citation If you use this dataset, please cite the original BRISC paper: ```bibtex @article{Fateh_2026, title = {BRISC: Annotated Dataset for Brain Tumor Segmentation and Classification}, author = {Fateh, Amirreza and Rezvani, Yasin and Moayedi, Sara and Rezvani, Sadjad and Fateh, Fatemeh and Fateh, Mansoor and Abolghasemi, Vahid}, journal = {Scientific Data}, volume = {13}, number = {1}, year = {2026}, publisher = {Springer Science and Business Media LLC}, doi = {10.1038/s41597-026-06753-y}, url = {http://dx.doi.org/10.1038/s41597-026-06753-y} } ``` --- [Chehab Lab](https://chehablab.com) @ 2026