{"success":true,"data":{"author":"Oskar Maier, Mauricio Reyes, Bjoern Menze et al. (ISLES organizers); mirrored by MedOtter","bucketAccessType":"public","bucketName":"MedOtter/isles2015","bucketPrefix":"","category":"Neuroscience","cloudProvider":"huggingface","createdAt":"2026-11-21T22:16:41.322Z","description":"Training data of the MICCAI 2015 ISLES (Ischemic Stroke Lesion Segmentation) challenge, in two sub-challenges. SISS (sub-acute ischemic stroke lesion segmentation) has 28 cases from UKSH Luebeck with T1, T2, diffusion-weighted (DWI) and FLAIR MRI at 1 mm isotropic, each with a binary lesion mask traced on FLAIR by an experienced physician. SPES (stroke perfusion estimation) has 30 acute cases from University Hospital Bern with T1c, T2, DWI and perfusion maps (CBF, CBV, TTP, Tmax) at 2 mm, each with an infarct core label, a penumbra label (the hypoperfused tissue at risk, Tmax > 6 s with manual correction) and a merged map. All volumes are skull-stripped and co-registered. Uncompressed NIfTI (.nii) under ISLES2015_SISS/training/ and ISLES2015_SPES/training/ in the original SMIR VSD.Brain folder naming, plus train.jsonl and a preview Parquet, about 4.6 GB (measured 2026-11-21); read with nibabel. Test cases are not included. The organizers re-deposited the data on Zenodo (DOI 10.5281/zenodo.17736412) under CC BY 4.0; the original SMIR release carried ODbL. Cite Maier et al., Medical Image Analysis 35:250-269 (2017).","displayTitle":"ISLES 2015: Ischemic Stroke Lesion Segmentation, sub-acute stroke MRI (SISS) and acute penumbra estimation (SPES), 58 training cases","doi":"10.5281/zenodo.17736412","huggingFaceUrl":"https://huggingface.co/datasets/MedOtter/isles2015","id":"aeb5502d-8493-5815-bd8c-e8310392f026","license":"CC-BY-4.0","licenseEvidenceUrl":"https://zenodo.org/records/17736412","objectCountEstimate":884,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1016/j.media.2016.07.009","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/isles2015\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"MedOtter/isles2015\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=MedOtter, license:odbl, size_categories:n<1K, format:parquet, downloads=766"],"format":"mixed","generatedAt":"2026-11-21T22:32:03.070Z","layout":{"example":"data/train-00000-of-00001.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"50415231150415901215c6044c1574150012000088094c10000000736973735f747261696e696e675f30314a140000324a140000334a140000344a140000354a140000364a140000374a140000384a140000394614000431304a14004ac80000314ac80000314ac80000314ac80000314ac80000314ac80000314ac80000314ac80000314ac80000324ac80000324ac80000324ac80000324ac80000324ac80000324ac80000324ac80000324ac800003209c804706546300236140009a036140009a036140009a036140009a036140009a036140009a03614004ea00029683228004d3036140029683614004ac8004d303628004ac8004d303628004ac8004d303628004ac8004d303228004d303614004ac8004d303628004a90014d303628004a90014d303628004a90014e300200324a90010433301500157015742c1574151015061506000038dc020000007401061140200c44611c48a22c4ce33c50244d54655d58a66d5ce77d60288e64699e68aaae6cebbe702ccf746ddf780e000000001504152015244c15041500120000103c0400000053495353040000005350455315001516151a2c157415101506150600000b280200000074010138003c01150415f00115f6014c153c150012000078f0770100000002000000030000000400000005000000060000000700000008000000090000000a0000000b0000000c","encoding":"hex","key":"data/train-00000-of-00001.parquet","truncatedAt":512}],"sampleObjects":["ISLES2015_SISS/training/1/VSD.Brain.XX.O.MR_DWI.70613/License_ODC_ODBL.txt","ISLES2015_SISS/training/1/VSD.Brain.XX.O.MR_DWI.70613/VSD.Brain.XX.O.MR_DWI.70613.nii","ISLES2015_SISS/training/1/VSD.Brain.XX.O.MR_Flair.70614/License_ODC_ODBL.txt","ISLES2015_SISS/training/1/VSD.Brain.XX.O.MR_Flair.70614/VSD.Brain.XX.O.MR_Flair.70614.nii","ISLES2015_SISS/training/1/VSD.Brain.XX.O.MR_T1.70615/License_ODC_ODBL.txt"],"sidecars":[{"bytes":"---\nlicense: odbl\npretty_name: ISLES 2015 (SISS + SPES) — training splits\ntags:\n- medical-imaging\n- mri\n- brain\n- stroke\n- lesion-segmentation\n- 3d\ntask_categories:\n- image-segmentation\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: data/train-*\ndataset_info:\n  features:\n  - name: sample_id\n    dtype: string\n  - name: subchallenge\n    dtype: string\n  - name: case_id\n    dtype: int32\n  - name: split\n    dtype: string\n  - name: image\n    dtype: image\n  - name: mask\n    dtype: image\n  - name: overlay\n    dtype: image\n  - name: base_modality\n    dtype: string\n  - name: primary_label\n    dtype: string\n  - name: modalities\n    dtype: string\n  - name: shape\n    dtype: string\n  - name: spacing_mm\n    dtype: string\n  - name: num_slices\n    dtype: int32\n  - name: slice_index\n    dtype: int32\n  - name: lesion_voxels\n    dtype: int64\n  splits:\n  - name: train\n    num_bytes: 1493396\n    num_examples: 58\n  download_size: 1500249\n  dataset_size: 1493396\n---\n\n# ISLES 2015 (SISS + SPES) — training splits\n\nRaw mirror of the MICCAI **ISLES 2015** (Ischemic Stroke Lesion Segmentation) challenge\n**training data**, from the challenge organizers' Zenodo re-archive\n([record 19135955](https://zenodo.org/records/19135955), v3 of concept DOI\n[10.5281/zenodo.17736411](https://doi.org/10.5281/zenodo.17736411); Reyes / de la Rosa / Menze).\n\n**Scope (read this):** this mirror contains the **58 training cases with public ground truth**.\nThe 56 test cases (SISS 36, SPES 20) are **not included**: their ground truth was never\nreleased and the original SMIR online evaluation service is dead, so they cannot be scored.\n\n## Composition — two sub-challenges, not mere splits\n\n| Sub-challenge | Task | Cases here | Modalities | Voxels | Center |\n|---|---|---|---|---|---|\n| **SISS** | Sub-acute ischemic stroke lesion segmentation | 28 train | T1, T2, DWI (b=1000), FLAIR | 1×1×1 mm (~230×230×154) | UMC Schleswig-Holstein, Lübeck |\n| **SPES** | Acute stroke penumbra estimation | 30 train | T1c, T2, DWI, CBF, CBV, TTP, Tmax | 2×2×2 mm (96×110×71-72) | University Hospital Bern (2005–2013) |\n\nPer case, all volumes are skull-stripped and rigidly co-registered to a common grid\n(SISS: to FLAIR, resampled 1 mm³; SPES: to T1c). Uncompressed NIfTI (`.nii`).\n\n## Ground truth\n\n- **SISS** — `OT` folder per case: binary lesion mask by one experienced MD, delineated on\n  FLAIR (lesion counted only if pathologic in both FLAIR and DWI). Single public tier (GT01).\n  A second test-set rater existed (inter-observer Dice 0.70±0.20 — the human ceiling reported\n  by the paper) but test GT is withheld.\n- **SPES** — three label tiers per case:\n  - `corelabel` — binary infarct core (ADC < 600×10⁻⁶ mm²/s threshold).\n  - **`penumbralabel` — the designated gold standard**: complete hypoperfused lesion\n    (semi-manual, Tmax > 6 s + manual correction). This is the target the 2015 challenge\n    actually evaluated.\n  - `mergedlabels` — disjoint decomposition: **1 = penumbra − core, 2 = core**. Verified\n    voxel-exact: `merged==2` ≡ core and `merged>0` ≡ core ∪ penumbra.\n  - ⚠️ Core ⊄ penumbra exactly: a few core voxels (~14–205 per case) fall outside\n    `penumbralabel`. Treat core and penumbra as two overlapping binary targets (fan-out),\n    or use `mergedlabels` for a clean 3-class map — do not assume strict nesting.\n\nNo empty masks: every training case has non-empty GT (all tiers).\n\n## Structure\n\n```\nISLES2015_SISS/training/<1..28>/VSD.Brain.XX.O.<MR_DWI|MR_Flair|MR_T1|MR_T2|OT>.<id>/<same>.nii\nISLES2015_SPES/training/<1..30>/VSD.Brain.XX.O.<MR_CBF|MR_CBV|MR_DWI|MR_T1c|MR_T2|MR_Tmax|MR_TTP>.<id>/<same>.nii\nISLES2015_SPES/training/<n>/{corelabel,penumbralabel,mergedlabels}/VSD.Brain*.XX.O.OT.<id>/<same>.nii\ntrain.jsonl          # one row per case: relative paths per modality/label tier + shape/spacing\n```\n\n- The numeric `<id>` suffixes are SMIR-internal object IDs, unique per file and\n  **release-specific** — match files by modality token (`.MR_Flair.`, `.OT.`), never by ID.\n- `mergedlabels` files use a `VSD.Bra","encoding":"utf8","key":"README.md","name":"README.md","size":5714,"truncated":true}],"sourceObjectKey":"data/train-00000-of-00001.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/isles2015\")\nprint(ds)","starterSamplePath":"ISLES2015_SISS/training/1/VSD.Brain.XX.O.MR_DWI.70613/VSD.Brain.XX.O.MR_DWI.70613.nii","totalObjects":884},"summary":"Training data of the MICCAI 2015 ISLES (Ischemic Stroke Lesion Segmentation) challenge, in two sub-challenges.","title":"ISLES 2015: Ischemic Stroke Lesion Segmentation, sub-acute stroke MRI (SISS) and acute penumbra estimation (SPES), 58 training cases","updatedAt":"2026-11-22T00:53:09.638Z","visibility":"public","websiteUrl":"https://www.isles-challenge.org/"}}