Add comprehensive MM-OphBench Dataset Card

README.md

@@ -0,0 +1,219 @@
+---
+annotations_creators:
+- expert-annotated
+- crowdsourced
+- machine-generated
+language_creators:
+- found
+language:
+- en
+- zh
+license: cc-by-nc-sa-4.0
+multilinguality:
+- bilingual
+size_categories:
+- 1M<n<10M
+source_datasets:
+- multi-center-clinical
+task_categories:
+- image-classification
+- image-segmentation
+- object-detection
+- visual-question-answering
+task_ids:
+- multi-label-image-classification
+- semantic-segmentation
+- medical-imaging
+pretty_name: "MM-OphBench: Multi-Center Multimodal Clinical Ophthalmic Benchmark"
+tags:
+- medical
+- ophthalmology
+- multimodal
+- oct
+- fundus
+- uwf
+- ffa
+- ubm
+- b-scan
+- retina
+- glaucoma
+---
+
+# MM-OphBench: Multi-Center Multimodal Clinical Ophthalmic Benchmark Dataset
+
+<div align="center">
+
+**A Large-Scale, Standardized Multi-Center Benchmark Covering 7 Imaging Modalities & 4.3M+ Clinical Records**
+
+[![Hugging Face Dataset](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Dataset-blue)](https://huggingface.co/datasets/Kaphathy/Dataset)
+[![License: CC BY-NC-SA 4.0](https://img.shields.io/badge/License-CC_BY--NC--SA_4.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
+[![Modalities](https://img.shields.io/badge/Modalities-CFP%20|%20UWF%20|%20OCT%20|%20FFA%20|%20UBM%20|%20B--Scan-brightgreen)]()
+[![Images](https://img.shields.io/badge/Images-4%2C307%2C415-orange)]()
+
+</div>
+
+---
+
+## 1. Executive Summary & Repository Overview
+
+The **MM-OphBench** repository hosts a petabyte-scale, clinically harmonized ophthalmic image archive compiled from leading ophthalmic hospitals and benchmark cohorts. It spans **4,307,415 high-resolution diagnostic images** and multimodal pairs across **7 distinct imaging modalities**, with comprehensive clinical metadata, expert lesion segmentations, and hierarchical diagnostic taxonomies.
+
+All archives are compressed into high-speed Zstandard chunks (`.tar.zst`) with pre-indexed SHA-256 manifests. A centralized, fully de-identified master index is provided in `metadata/adapter_manifest_reviewed.parquet` (39.9 MB), allowing instant cross-cohort querying across imaging modalities, device models, scan protocols, and clinical diagnoses without requiring massive raw downloads.
+
+---
+
+## 2. Multi-Center Clinical Provenance & Ethical Governance
+
+All patient identifiable information (PII) including patient names, hospital record numbers (MRNs), national identity codes, and date-of-birth timestamps have been removed or replaced with deterministic SHA-256 cryptographic hashes (`patient_hash`, `study_id`), strictly complying with HIPAA Safe Harbor and GDPR pseudonymization principles.
+
+### Key Contributing Clinical Centers & Cohorts
+1. **The Second Affiliated Hospital of Zhejiang University School of Medicine (浙二眼科)**
+   - *Contributions*: Longitudinal FFA/ICGA dynamic angiography sequences (`Final_eryuan_FFA`), 17-class retinal disease benchmark (`eryuanFundusT17RDCls`), anterior segment photography.
+   - *Clinical Leads*: Certified ophthalmologists (Prof. Wang Guoping and retinal angiography specialist team).
+2. **Beijing Tongren Hospital, Capital Medical University (北京同仁医院)**
+   - *Contributions*: Longitudinal multi-condition clinical cohorts (`tongren_archives`), 66-class fine-grained disease screening benchmark (`trhd3_FRD_Cls66_Train`), and the standard 95-disease clinical taxonomy (`tongren95_taxonomy.csv`).
+3. **The Second Xiangya Hospital of Central South University (中南大学湘雅二医院)**
+   - *Contributions*: High-frequency ocular B-scan ultrasonography with detailed diagnostic text reports (`BUltrasound`).
+4. **InEye Hospital of Chengdu University of TCM (成都中医药大学附属银海眼科医院)**
+   - *Contributions*: Comprehensive Ultra-Widefield (UWF) Optos color imaging (`yinhai_uwf`, 274.6 GB), FFA dynamic sequences, and ocular B-scans.
+5. **Chengdu First People's Hospital (成都市第一人民医院)**
+   - *Contributions*: 50MHz Ultrasound Biomicroscopy (UBM) ciliary body and anterior chamber angle quantitative morphometry (`ChengduShiyiUBM`).
+6. **Handan Eye Hospital (河北邯郸眼科医院)**
+   - *Contributions*: Fluorescein angiography paired cohorts (`handan_ffa`, `handan_ffa2`) and epidemiological diabetic retinopathy screenings.
+7. **Zhongshan Hospital, Fudan University (复旦大学附属中山医院)**
+   - *Contributions*: Multimodal cohort comprising 419,014 clinical visits with cross-system validation.
+
+---
+
+## 3. Optical Modalities & Hardware Instrumentation
+
+The dataset captures full optical, acoustic, and angiography representations across the anterior and posterior segments of the human eye:
+
+| Modality Subtype | Physical Imaging Principle | Primary Acquisition Hardware | Typical Clinical Protocol & Resolution | Sample Count |
+| :--- | :--- | :--- | :--- | :--- |
+| **Standard Color Fundus (CFP)** | 45° Broadband White Flash Photography | Canon CR6-45NM, CR-DGi, Topcon TRC-NW8 | 45° macular/disc centered; 2048x1536 to 4096x3072 px | **2,332,938** |
+| **OCT B-scan** | Swept-Source (1050nm) & Spectral-Domain (870nm) OCT | Topcon DRI OCT Triton, Optovue RTVue XR Avanti, Heidelberg Spectralis | 3D Macula (512x256), 6x6mm / 12mm radial lines; axial res 2.6-7.0 μm | **939,312** |
+| **Ultrasound Biomicroscopy (UBM)**| 50 MHz High-Frequency Acoustic Ultrasound | Tianjin Suowei SW-3200, MEDA MD-300L | Anterior chamber angle, ciliary body, ACD depth, 50 MHz probe | **449,588** |
+| **Ultra-Widefield Fundus (UWF)** | Dual-wavelength (532/635nm) Confocal Scanning Laser | Optos 200Tx / California (200° Field of View) | Non-mydriatic 200° peripheral retina; 3900x3072 px | **206,215** |
+| **Ocular B-Scan Ultrasound** | 10–20 MHz Pulse-Echo Acoustic Transducer | Sonomed Escalon, Quantel Medical, Mada | Vitreous opacity, retinal detachment, axial length | **112,612** |
+| **External Anterior Photography**| Diffuse Visible Light Illumination | Topcon / Slit Lamp Imaging Camera | Cornea, iris, lens, cataract grading, external eye | **207,001** |
+| **Reflectance SLO (En-face)** | 820 nm Infrared Scanning Laser Ophthalmoscopy | Heidelberg Spectralis HRA, Optos | En-face confocal infrared reflectance (IR-SLO) | **36,024** |
+| **Fluorescein Angiography (FFA/ICGA)** | Dynamic Fluorescein Sodium (490/520nm) & ICG (790/830nm) | Heidelberg Spectralis HRA2, Topcon TRC-50DX | Dynamic perfusion sequence: early (0-30s), mid (1-3m), late (>5m) | **19,406** |
+| **OCT / OCTA En-Face C-scans** | Motion-contrast decorrelation & slab projection | Optovue AngioVue, Topcon DRI Triton | Superficial/deep capillary plexus, outer retina slab | **3,000** |
+
+---
+
+## 4. Archive Directory & Shard Inventory
+
+Archives are organized into self-contained sub-directories chunked at ~1024 MiB boundaries. Each shard is compressed using `zstd -1` for fast decompressibility:
+
+| Dataset Directory | Modality Focus | Shards | Total Size | Primary Source Center | Key Files Included |
+| :--- | :--- | :--- | :--- | :--- | :--- |
+| `archives/yinhai_uwf/` | Ultra-Widefield (UWF) | 55 | 274.6 GB | Chengdu Yinhai Eye Hospital | 200° Optos color scans (55 tar.zst shards) |
+| `archives/data0410/` | CFP, OCT Buckets, Comorbidity | 7 | 273.1 GB | Multi-center | `fundus_hcw.zip`, `octBuckets.zip`, clinical diagnosis Excel |
+| `archives/tongren_archives/`| Longitudinal Retinal Cohort | 2 | 133.0 GB | Beijing Tongren Hospital | Raw longitudinal image archives |
+| `archives/zhanghuan_selected_public/` | Public Benchmark Suites | 40 | 106.8 GB | Curated international public sets | 40 deduplicated archives (Messidor, ODIR, DDR, etc.) |
+| `archives/Final_eryuan_FFA/` | Dynamic FFA/ICGA Sequences | 19 | 76.4 GB | ZJU-2 Eye Center | Full dynamic FFA frames + `output2.jsonl`, `visionoph4.jsonl` |
+| `archives/eryuanFundusT17RDCls/` | 17-Class Retinal Disease CFP | 74 | 47.4 GB | ZJU-2 Eye Center | 17-category labeled fundus images for benchmark training |
+| `archives/ChengduShiyiUBM/` | 50MHz Anterior UBM | 255 | 39.3 GB | Chengdu 1st People's Hospital | 50MHz ultrasound sweeps + `ChengduShiyiUBM.json` (ACD mm) |
+| `archives/OCTA-500/` | 3D OCTA + Projection Slabs | 37 | 37.1 GB | NJUST / Public benchmark | 6mm & 3mm OCTA volumes with vascular masks |
+| `archives/OCTA500_OCT/` | Structural OCT Volumes | 30 | 29.3 GB | NJUST / Public benchmark | Structural OCT volumes paired with OCTA-500 |
+| `archives/WLOA/` | Wide-Field OCT Angiography | 18 | 17.4 GB | Benchmark suite | Wide-field OCTA scans |
+| `archives/BUltrasound/` | Ocular B-Scan Ultrasound | 38 | 11.3 GB | Xiangya-2 & Yinhai | B-scan ultrasounds + `XiangyaBscan.json`, `YinhaiBscan.json` |
+| `archives/trhd3_FRD_Cls66_Train/` | 66-Class Fine-Grained CFP | 8 | 7.8 GB | Tongren & Handan Eye Hospital | 66-disease multiclass training set with splits |
+| `archives/handan_ffa/` | FFA Dynamic Series | 5 | 4.5 GB | Handan Eye Hospital | Handan clinical FFA angiograms |
+| `archives/handan_ffa2/` | FFA Dynamic Series 2 | 3 | 2.7 GB | Handan Eye Hospital | Longitudinal follow-up FFA |
+| `archives/chengdu_ffa/` | FFA Dynamic Series | 2 | 1.2 GB | Chengdu Yinhai Eye Hospital | Chengdu clinical FFA angiograms |
+| `archives/yinhai_ffa/` | FFA Dynamic Series | 2 | 0.5 GB | Chengdu Yinhai Eye Hospital | High-resolution FFA sequences |
+| `archives/DGOCF/` | Optic Disc & Glaucoma CFP | 1 | 0.8 GB | Benchmark suite | Optic nerve head center annotations |
+
+---
+
+## 5. Master Metadata (`metadata/`) & Clinical Crosswalks
+
+Instead of downloading multi-hundred-gigabyte archives to inspect sample attributes, download the lightweight parquet files in `metadata/`:
+
+### 1. `metadata/adapter_manifest_reviewed.parquet` (39.9 MB)
+Contains **4,307,415 rows** mapping every asset with standard schema:
+```python
+import duckdb
+
+con = duckdb.connect()
+# Query device distribution across modalities
+con.execute("""
+    SELECT modality_subtype, device_vendor, device_model, count(*) AS count
+    FROM 'metadata/adapter_manifest_reviewed.parquet'
+    GROUP BY 1, 2, 3
+    ORDER BY count DESC
+    LIMIT 10
+""").df()
+```
+
+### 2. `metadata/tongren95_taxonomy.csv` (12 KB)
+Comprehensive 95-class disease taxonomy mapping local clinical diagnoses (`raw_zh`) to standardized English terms (`canonical_en`), abbreviations (`AMD`, `PDR`, `RVO`, `CSC`), disease families (`family_en`), and specificity levels.
+
+### 3. `metadata/labeled_assets_summary.json` (2.0 KB)
+Complete breakdown of 1,817,159 instruction-tuned samples and 1,459,879 downstream high-confidence diagnostic splits.
+
+---
+
+## 6. Benchmark Evaluation Tasks
+
+1. **17-Class Retinal Disease Classification (`eryuanFundusT17RDCls`)**:
+   Standardized 17-class classification for common and sight-threatening retinal conditions (Normal, DR, AMD, RVO, Pathological Myopia, Macular Hole, Epiretinal Membrane, Retinal Detachment, Retinitis Pigmentosa, etc.).
+2. **66-Class Multi-Center Fine-Grained Diagnosis (`trhd3_FRD_Cls66_Train`)**:
+   Challenging 66-class classification covering rare and subtle ophthalmic pathologies validated across Beijing Tongren and Handan populations.
+3. **95-Class Clinical Semantic Classification (`tongren95_taxonomy`)**:
+   Full-spectrum hospital diagnostic classification aligning free-text clinical reports to standardized ICD/canonical labels.
+4. **Dynamic FFA Vessel & Microvascular Leakage Segmentation**:
+   Temporal sequence segmentation across early, arteriovenous, and late phases.
+5. **Quantitative Anterior Chamber Morphometry (UBM)**:
+   Automated measurement of Anterior Chamber Depth (ACD in mm), Trabecular-Iris Angle (TIA), and angle-closure risk assessment.
+
+---
+
+## 7. Quick Start: Extraction & Usage Guide
+
+### A. Download & Extract Archives (Linux / macOS)
+To extract multi-part `.tar.zst` archives with maximum parallelism:
+
+```bash
+# Install zstd if not present
+sudo apt-get install -y zstd  # Ubuntu/Debian
+brew install zstd             # macOS
+
+# Extract all shards for a specific dataset (e.g., BUltrasound)
+cat archives/BUltrasound/BUltrasound-*.tar.zst | tar -I zstd -xvf - -C /path/to/destination/
+```
+
+### B. Python Fast Metadata Query
+```python
+import pandas as pd
+
+# Read metadata directly from Hugging Face or local path
+df = pd.read_parquet("metadata/adapter_manifest_reviewed.parquet", 
+                     columns=["image_id", "modality_subtype", "device_model", "hospital_domain"])
+
+# Filter for SS-OCT B-scans from Topcon Triton
+topcon_oct = df[(df["modality_subtype"] == "oct_bscan") & (df["device_model"] == "DRI OCT Triton")]
+print(f"Loaded {len(topcon_oct)} Topcon SS-OCT B-scans")
+```
+
+---
+
+## 8. Citation & Acknowledgments
+
+If you use this benchmark or any subsets in your research, please cite:
+```bibtex
+@dataset{mm_ophbench2026,
+  title={MM-OphBench: A Multi-Center Multimodal Clinical Ophthalmic Benchmark Dataset},
+  author={Zhejiang University, Beijing Tongren Hospital, InEye Hospital, Xiangya Hospital, and Contributors},
+  year={2026},
+  publisher={Hugging Face},
+  howpublished={\url{https://huggingface.co/datasets/Kaphathy/Dataset}}
+}
+```
+
+## 9. License & Terms of Use
+- The aggregated metadata and clinical annotations are made available under **Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)**.
+- Research use only. Not for direct diagnostic or medical intervention without regulatory clearance.