{"success":true,"data":{"author":"Mahdi Bashiri Bawil et al.; repackaged by Chehab Lab","bucketAccessType":"public","bucketName":"chehablab/MS3SEG_t1w_t2w","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-21T22:16:41.726Z","description":"MS3SEG is a multiple sclerosis (MS) MRI dataset whose labels separate three things that look alike on scans: the brain ventricles, normal white matter hyperintensities (WMH) caused by aging, and abnormal WMH that are focal MS demyelinating plaques. Separating them matters because only the plaques count toward MS lesion burden for diagnosis and treatment monitoring. This repository is a 2D slice version of T1-weighted and T2-weighted images: each row is one brain-extracted 256x256 slice with its sequence type, volume and slice IDs, participant age, sex, weight and height, and three binary masks (vent_mask, nwmh_mask, abwmh_mask). The original release also includes raw DICOM, preprocessed NIfTI volumes, overlays, patient metadata and cross-validation splits. One Parquet file under data/, about 90 MB (measured 2026-11-21); read with Hugging Face datasets. Use for MS lesion segmentation that distinguishes pathological from age-related white matter changes. The data descriptor states the dataset is available under CC BY 4.0 (this mirror's CC0 tag is not the original license). Cite Bashiri Bawil et al., Scientific Data (2026), doi 10.1038/s41597-026-07184-5.","displayTitle":"MS3SEG: multiple sclerosis brain MRI slices with tri-mask annotations (ventricles, normal and MS white matter hyperintensities)","huggingFaceUrl":"https://huggingface.co/datasets/chehablab/MS3SEG_t1w_t2w","id":"c9202808-e93e-5d1a-b148-7148cc066848","license":"CC-BY-4.0","licenseEvidenceUrl":"https://www.nature.com/articles/s41597-026-07184-5","objectCountEstimate":3,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1038/s41597-026-07184-5","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"chehablab/MS3SEG_t1w_t2w\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"chehablab/MS3SEG_t1w_t2w\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=chehablab, license:cc0-1.0, size_categories:1K<n<10K, format:parquet, downloads=108"],"format":"mixed","generatedAt":"2026-11-21T22:32:16.928Z","layout":{"example":"data/train-00000-of-00001.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"5041523115041518151c4c150615001200000c2c0100000002000000030000001500152415282c15c8011510150615061c1804030000001804010000001600280403000000180401000000000000124403000000c80101024800240124022401140215041590011596014c1524150012000048f047000000000100000002000000030000000400000005000000060000000700000008000000090000000a0000000b0000000c0000000d0000000e0000000f000000100000001100000015001594011580012c15c8011510150615061c18041100000018040000000016002804110000001804000000000000004ad803000000c80101051b2088418a3928a9c59a7b3082200629e6a0a4166beec1088218a49883925aacb90723086290620e4a6ab1e61e8c203e2d00080400001504151c15204c150415001200000e3403000000743177030000007432771500152015242c15c8011510150615061c360028037432771803743177000000103c03000000c8010101240024014800380115041518151c4c150615001200000c2c2c0000001c0000001d0000001500152415282c15c8011510150615061c18042c00000018041c000000160028042c00000018041c000000000000124403000000c8010102480024012402240114021504151415184c150415001200000a240100000046010000004d1500151a151e2c15c80115","encoding":"hex","key":"data/train-00000-of-00001.parquet","truncatedAt":512}],"sampleObjects":["data/train-00000-of-00001.parquet"],"sidecars":[{"bytes":"---\nlicense: cc0-1.0\ndataset_info:\n  features:\n  - name: volume_id\n    dtype: int32\n  - name: slice_id\n    dtype: int32\n  - name: type\n    dtype: string\n  - name: age\n    dtype: int32\n  - name: sex\n    dtype: string\n  - name: weight\n    dtype: float32\n  - name: height\n    dtype: float32\n  - name: image\n    dtype: image\n  - name: vent_mask\n    dtype: image\n  - name: nwmh_mask\n    dtype: image\n  - name: abwmh_mask\n    dtype: image\n  splits:\n  - name: train\n    num_bytes: 95451164.224\n    num_examples: 3636\n  download_size: 93588878\n  dataset_size: 95451164.224\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: data/train-*\nlanguage:\n- en\ntask_categories: \n- image-segmentation\ntags:\n- mri\n- medical\n- multiple-sclerosis\n- lesion-segmentation\n- white-matter\npretty_name: MS3SEG — Multiple Sclerosis MRI Dataset with Tri-Mask Annotations\nsize_categories:\n- 1K<n<10K\n---\n# MS3SEG: Multiple Sclerosis MRI Dataset with Tri-Mask Annotations for Lesion Segmentation\n\nThis dataset contains materials from the study *\"A Multiple Sclerosis MRI Dataset with Tri-Mask Annotations for Lesion Segmentation\"* by Bashiri Bawil et al. (2026), aimed at developing and benchmarking segmentation algorithms that distinguish true MS demyelinating lesions from normal age-related white matter hyperintensities (WMH) and ventricular CSF artifacts.\n\n\n## 🧠 Dataset Description\n\nThis repository provides a **2D slice-based version** of the original MS3SEG dataset, designed for efficient deep learning and segmentation workflows.\n\nEach entry corresponds to a **single 2D slice** extracted  from a volume and includes the normalized MRI image, three corresponding segmentation masks, and participant-level clinical/demographic metadata.\n\nThe dataset addresses a critical clinical challenge: not all white matter hyperintensities visible on images represent pathological MS lesions. The **tri-mask annotation framework** explicitly distinguishes:\n\n1. **Ventricles** (`vent_mask`) — Fluid-filled cavities in the brain center. Explicitly masking ventricles gives the model anatomical context to avoid misclassifying these boundaries as pathology.\n\n2. **Normal WMH** (`nwmh_mask`) — Diffuse white matter signal changes from natural aging: gradual myelin degradation, axonal loss, and increased extracellular fluid. Not pathological MS activity — must be excluded from lesion burden calculations to avoid inflating disease scores.\n\n3. **Abnormal WMH** (`abwmh_mask`) — Focal MS demyelinating plaques. These are the primary clinical target — the basis for MS diagnosis and treatment response monitoring.\n\n\n## 📦 Dataset Structure\n\nEach dataset entry contains:\n\n- `volume_id` → Unique identifier for the MRI volume / patient\n- `slice_id` → Index of the 2D slice within the volume\n- `type` → MRI sequence type (e.g., `T1WI`, `T2WI`)\n- `age` → Participant age\n- `sex` → Participant sex\n- `weight` → Participant weight (kg)\n- `height` → Participant height (m)\n- `image` → 2D preprocessed slice (256×256, brain-extracted)\n- `vent_mask` → Binary segmentation mask for ventricles\n- `nwmh_mask` → Binary segmentation mask for normal white matter hyperintensities\n- `abwmh_mask` → Binary segmentation mask for abnormal (MS) white matter hyperintensities\n\n\n## Use\n\n```python\nfrom datasets import load_dataset\nimport matplotlib.pyplot as plt\nimport numpy as np\n\ndataset = load_dataset(\"chehablab/ms3seg_t1w_t2w\", split=\"train\")\n\nsample_id = 42\n\nimage    = dataset[sample_id][\"image\"]\nvent     = dataset[sample_id][\"vent_mask\"]\nnwmh     = dataset[sample_id][\"nwmh_mask\"]\nabwmh    = dataset[sample_id][\"abwmh_mask\"]\nscan_type = dataset[sample_id][\"type\"]\n\nfig, axes = plt.subplots(1, 4, figsize=(16, 4))\naxes[0].imshow(image,  cmap=\"gray\");  axes[0].set_title(f\"Image ({scan_type})\")\naxes[1].imshow(vent,   cmap=\"Blues\"); axes[1].set_title(\"Ventricles\")\naxes[2].imshow(nwmh,   cmap=\"Greens\"); axes[2].set_title(\"Normal WMH\")\naxes[3].imshow(abwmh,  cmap=\"Reds\");  axes[3].set_title(\"Abnormal WMH\")\nfor ax in axes: ax.axis(\"off\")\nplt.tight_layout()\nplt.show()\n```\n\n\n## 📚 Ci","encoding":"utf8","key":"README.md","name":"README.md","size":6777,"truncated":true}],"sourceObjectKey":"data/train-00000-of-00001.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"chehablab/MS3SEG_t1w_t2w\")\nprint(ds)","starterSamplePath":"data/train-00000-of-00001.parquet","totalObjects":3},"summary":"MS3SEG is a multiple sclerosis (MS) MRI dataset whose labels separate three things that look alike on scans: the brain ventricles, normal white matter hyperintensities (WMH)…","title":"MS3SEG: multiple sclerosis brain MRI slices with tri-mask annotations (ventricles, normal and MS white matter hyperintensities)","updatedAt":"2026-11-22T00:50:00.419Z","visibility":"public","websiteUrl":"https://github.com/Mahdi-Bashiri/MS3SEG"}}