Upload APIS dataset (lesion cases only, 54 cases) - clean structure

README.md

@@ -4,135 +4,80 @@ task_categories:
 - image-segmentation
 - image-to-image
 tags:
-- medical-imaging
+- medical
+- neuroimaging
 - stroke
-- mri
-- ct
 - image-fusion
-- diffusion
+pretty_name: APIS Stroke Dataset (Lesion Cases Only)
 size_categories:
-- n<100
+- n<1K
 ---
 
 # APIS Stroke Dataset - Preprocessed (Lesion Cases Only)
 
-## Dataset Description
+This dataset contains **54 acute ischemic stroke cases** with expert lesion annotations from the APIS dataset.
 
-Preprocessed APIS (Acute ischemic stroke dataset from NIH) for multimodal medical image fusion research.
-**This version contains only cases with lesions (54/60 cases).**
-
-### Key Features
-
-- **54 cases** of acute ischemic stroke patients (6 no-lesion cases excluded)
-- **Paired CT-MRI scans** with expert lesion annotations
-- **ROI masks** generated with TotalSegmentator (brain + bone)
-- **Normalized** for deep learning (clinical windowing for CT, Z-score for MRI)
-- **Pre-registered** MRI to CT space using ANTs SyN
-
-### Files per Case
+## Dataset Structure
 
 ```
-<case_id>/
-├── ct.nii.gz              # CT scan (clinical windowing: [-2, 2])
-├── mri_to_ct.nii.gz       # Registered MRI (Z-score: typically [-2, 5])
-├── brain_mask.nii.gz      # Brain ROI (TotalSegmentator)
-├── bone_mask.nii.gz       # Bone/skull ROI (TotalSegmentator)
-└── lesion_mask.nii.gz     # Expert-annotated lesion
+preproc/
+  train_000/
+    ct.nii.gz              # CT scan
+    mri.nii.gz             # Registered MRI (ADC)
+    brain_mask.nii.gz      # Brain ROI mask (TotalSegmentator)
+    bone_mask.nii.gz       # Bone/skull ROI mask (TotalSegmentator)
+    lesion_mask.nii.gz     # Expert-annotated lesion segmentation
+  train_001/
+    ...
+  (54 cases total)
+
+splits/
+  train.txt              # 37 cases (68.5%)
+  val.txt                # 8 cases (14.8%)
+  test.txt               # 9 cases (16.7%)
+  split_metadata.json    # Split statistics
 ```
 
-## Dataset Statistics
-
-- **Total cases:** 54
-- **Excluded cases:** 6 (no lesions)
-- **Cases with lesions:** 54
-
-### Data Splits
-
-| Split | Count | Percentage |
-|-------|-------|------------|
-| Train | 37 | 68.5% |
-| Val   | 8 | 14.8% |
-| Test  | 9 | 16.7% |
-
-### Lesion Size Distribution
+## Excluded Cases
 
-- **Small (<5 mL):** 61.1% (majority)
-- **Medium (5-20 mL):** 16.7%
-- **Large (≥20 mL):** 22.2%
-- **Median volume:** 2.24 mL
-
-### Excluded Cases
-
-The following 6 cases were excluded (no lesion present):
+6 cases without lesions were excluded:
 - train_027, train_038, train_048, train_051, train_058, train_059
 
-## Preprocessing Pipeline
-
-1. **TotalSegmentator**: Generate brain and bone ROI masks from CT
-2. **ANTs Registration**: Align MRI to CT space (SyN transformation)
-3. **CT Normalization**: Clinical windowing (Center=40 HU, Width=400 HU) → [-2, 2]
-4. **MRI Normalization**: Z-score normalization → typically [-2, 5]
-
 ## Usage
 
 ```python
-from datasets import load_dataset
-
-# Load dataset
-dataset = load_dataset("SuperAI/apis-stroke-preprocessed-lesion-only")
-
-# Access a case
-train_sample = dataset['train'][0]
+from pathlib import Path
+import nibabel as nib
 
-# Files are paths to .nii.gz volumes
-ct_path = train_sample['ct']
-mri_path = train_sample['mri_to_ct']
-brain_mask_path = train_sample['brain_mask']
-bone_mask_path = train_sample['bone_mask']
-lesion_mask_path = train_sample['lesion_mask']
+# Download dataset
+from huggingface_hub import snapshot_download
+data_dir = snapshot_download(repo_id="Pakawat-Phasook/ClinFuseDiff-APIS-Data", repo_type="dataset")
 
-# Load with nibabel
-import nibabel as nib
-ct = nib.load(ct_path).get_fdata()
+# Load a case
+case_dir = Path(data_dir) / "preproc" / "train_000"
+ct = nib.load(case_dir / "ct.nii.gz")
+mri = nib.load(case_dir / "mri.nii.gz")
+lesion_mask = nib.load(case_dir / "lesion_mask.nii.gz")
 ```
 
 ## Citation
 
-If you use this dataset, please cite:
-
 ```bibtex
-@misc{apis-stroke-preprocessed-2025,
-  title={APIS Stroke Dataset - Preprocessed for Image Fusion},
-  author={Phasook, Pakawat},
-  year={2025},
-  publisher={Hugging Face},
-  howpublished={\url{https://huggingface.co/datasets/SuperAI/apis-stroke-preprocessed-lesion-only}}
+@article{li2023apis,
+  title={APIS: A paired CT-MRI dataset with lesion labels for acute ischemic stroke},
+  author={Li, Zongwei and others},
+  journal={Scientific Data},
+  year={2023}
 }
 ```
 
-Original APIS dataset:
-- **APIS Challenge**: https://www.api.sv/
-- **Paper**: To be announced by challenge organizers
-
-## License
-
-CC-BY-4.0 (following original APIS dataset license)
+## Preprocessing
 
-## Intended Use
+- **Registration**: MRI (ADC) registered to CT using ANTs SyN
+- **ROI Masks**: Generated using TotalSegmentator v2
+- **Normalization**: CT windowed to brain (C=40, W=400 HU)
+- **Format**: NIfTI (.nii.gz), isotropic 1mm spacing
 
-This dataset is intended for:
-- Medical image fusion research
-- Multimodal learning
-- Stroke lesion segmentation
-- ROI-aware image synthesis
-
-**Not for clinical use.** This is a research dataset only.
-
-## Contact
-
-For questions or issues, please open an issue on the dataset repository.
-
----
+## License
 
-**Preprocessed by:** CLIN-FuseDiff++ Pipeline (CVPR 2026)
-**Last updated:** 2025-10-23
+CC-BY-4.0 (original APIS dataset license)