Create README_LGG_MedThinkSeg.md

lgg/README_LGG_MedThinkSeg.md

@@ -0,0 +1,107 @@
+# LGG MRI Segmentation (TCGA-LGG / Kaggle “lgg-mri-segmentation”) — packaged in `saiteja33/MedThink-Seg/lgg`
+
+This README documents:
+1) **The original LGG MRI Segmentation dataset** (what it contains, size, and original file structure), and  
+2) **How it is organized inside this Hugging Face repo** (`saiteja33/MedThink-Seg`, under the `lgg/` directory).
+
+---
+
+## 1) Original dataset (source)
+
+**Original Kaggle dataset:** “Brain MRI segmentation” (`mateuszbuda/lgg-mri-segmentation`)  
+**Link:** https://www.kaggle.com/datasets/mateuszbuda/lgg-mri-segmentation
+
+**Underlying cohort:** TCGA lower-grade glioma (**TCGA-LGG**) cases from **The Cancer Imaging Archive (TCIA)**. 
+
+**Primary references (as listed with the Kaggle dataset):**
+- Buda, Saha, Mazurowski (2019), *Computers in Biology and Medicine* — DOI: **10.1016/j.compbiomed.2019.05.002**. 
+- Mazurowski et al. (2017), *Journal of Neuro-Oncology* — DOI: **10.1007/s11060-017-2420-1**. 
+
+### What images are in the dataset?
+- **Modality:** brain **MRI** (axial 2D slices) with segmentation masks. 
+- **Resolution:** **256 × 256** pixels per slice. 
+- **Channels / sequences:** images are stored as **3-channel** slices corresponding to **pre-contrast**, **FLAIR**, **post-contrast** (in that order). 
+  - Some cases have a missing sequence; missing channels are replaced with the available FLAIR sequence so all images remain 3-channel.  
+- **Masks:** **binary, 1-channel** masks that segment the **FLAIR abnormality**.  
+
+### Dataset size and counts
+- **Patients:** **110**.  
+- **Total slice-level pairs:** **3,929** axial slices with corresponding masks.  
+- **Slices per patient:** varies by patient (reported range **20–88** slices per volume).  
+
+### Labels and annotations
+- **Segmentation label:** tumor-related **FLAIR abnormality region** (binary mask).  
+
+**Naming convention (original):**  
+- Image: `TCGA_<institution-code>_<patient-id>_<slice-number>.tif`  
+- Mask: same name with **`_mask`** suffix, e.g. `..._<slice-number>_mask.tif`  
+
+### License
+The Kaggle dataset is commonly redistributed under **CC BY-NC-SA 4.0**.  
+
+---
+
+## 2) Organization inside this Hugging Face repo (`saiteja33/MedThink-Seg`)
+
+This repo packages the dataset under the `lgg/` directory with a flat image/mask layout plus a separate **normal** split (as shown in the HF file browser).
+
+### Directory structure (this repo)
+```
+lgg/
+├─ images/                 # MRI slices (tumor / abnormality present)
+├─ masks/                  # corresponding binary masks (paired with images/)
+└─ normal/
+   ├─ images/              # MRI slices with no abnormality (normal)
+   └─ masks/               # corresponding masks (typically empty/all-black)
+```
+
+### Pairing rule
+- Each image in `images/` has a corresponding mask in `masks/` with the **same base filename** (or the original `_mask` suffix pattern).  
+- Similarly, each file in `normal/images/` pairs with a file in `normal/masks/`.
+
+> Note: In the original Kaggle distribution, every slice typically has an associated mask file (some masks may be empty). In this HF packaging, those “empty-mask” slices are separated into `normal/`.
+
+---
+
+## How to download / use (recommended)
+
+### A) Download the raw files for `lgg/` from this HF repo
+```python
+from huggingface_hub import snapshot_download
+
+local_dir = snapshot_download(
+    repo_id="saiteja33/MedThink-Seg",
+    repo_type="dataset",
+    allow_patterns=["lgg/**"],   # only pull the lgg subdir
+)
+print(local_dir)
+```
+
+### B) Load with 🤗 Datasets (try these options)
+
+**Option 1 — load from the repo (works if the repo has a compatible dataset layout/metadata):**
+```python
+from datasets import load_dataset
+
+ds = load_dataset("saiteja33/MedThink-Seg", data_dir="lgg")
+print(ds)
+```
+
+**Option 2 — treat it as an image-folder dataset (useful for quick inspection):**
+```python
+from datasets import load_dataset
+
+ds = load_dataset("imagefolder", data_dir="lgg")
+print(ds)
+```
+
+---
+
+## Citation
+If you use this dataset, please cite the papers associated with the Kaggle release:
+
+- Mateusz Buda, Ashirbani Saha, Maciej A. Mazurowski. *Association of Genomic Subtypes of Lower-Grade Gliomas with Shape Features Automatically Extracted by a Deep Learning Algorithm.* Computers in Biology and Medicine (2019). DOI: **10.1016/j.compbiomed.2019.05.002**. 
+- Maciej A. Mazurowski et al. *Radiogenomics of lower-grade glioma: algorithmically-assessed tumor shape is associated with tumor genomic subtypes and patient outcomes in a multi-institutional study with The Cancer Genome Atlas data.* Journal of Neuro-Oncology (2017). DOI: **10.1007/s11060-017-2420-1**.   
+
+And cite the Kaggle dataset record:
+- **Kaggle:** `mateuszbuda/lgg-mri-segmentation` (Brain MRI segmentation).