Add Section 7 to root README linking to data_processing toolkit
Browse files
README.md
CHANGED
|
@@ -188,7 +188,22 @@ Complete breakdown of 1,817,159 instruction-tuned samples and 1,459,879 downstre
|
|
| 188 |
|
| 189 |
---
|
| 190 |
|
| 191 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 192 |
|
| 193 |
### A. Download & Extract Archives (Linux / macOS)
|
| 194 |
To extract multi-part `.tar.zst` archives with maximum parallelism:
|
|
@@ -217,7 +232,7 @@ print(f"Loaded {len(topcon_oct)} Topcon SS-OCT B-scans")
|
|
| 217 |
|
| 218 |
---
|
| 219 |
|
| 220 |
-
##
|
| 221 |
|
| 222 |
If you use this benchmark or any subsets in your research, please cite:
|
| 223 |
```bibtex
|
|
@@ -230,6 +245,6 @@ If you use this benchmark or any subsets in your research, please cite:
|
|
| 230 |
}
|
| 231 |
```
|
| 232 |
|
| 233 |
-
##
|
| 234 |
- The aggregated metadata and clinical annotations are made available under **Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)**.
|
| 235 |
- Research use only. Not for direct diagnostic or medical intervention without regulatory clearance.
|
|
|
|
| 188 |
|
| 189 |
---
|
| 190 |
|
| 191 |
+
---
|
| 192 |
+
|
| 193 |
+
## 7. Data Processing & Curation Toolkit (`data_processing/`)
|
| 194 |
+
|
| 195 |
+
All clinical data preprocessing, de-identification, quality filtering, black-border cropping, report parsing, modality decoupling, and taxonomy alignment scripts are provided in the standalone [`data_processing/`](https://huggingface.co/datasets/Kaphathy/Dataset/tree/main/data_processing) directory at the repository root.
|
| 196 |
+
|
| 197 |
+
- **Modular Organization**:
|
| 198 |
+
- `01_deidentification/`: High-throughput circular FOV masking, OCR-based PII redaction, and border trimming (`crop_fundus_images.py`, `deidentify_batch.py`, `audit_fundus_corner_pii.py`, etc.).
|
| 199 |
+
- `02_quality_control/`: ASRMNet-based quality grading (`asrm_preprocess.py`), EyePACS frozen gray-edge artifact scanning (`scan_eyepacs_edge_artifact.py`), and clinical diagnosis Excel verification (`clean_excel.py`, `build_hcw_semantic_patch.py`).
|
| 200 |
+
- `03_report_parsing/`: 41-column master manifest materialization (`materialize_newdata_prompts.py`), clinical note merging (`attach_labels.py`), and RoBERTa-SigLIP zero-shot OCT lesion cross-modal retrieval (`batch_retrieve_reports.py`).
|
| 201 |
+
- `04_modality_split/`: Multi-feature heuristic classifier for decoupling mixed OCT/OCTA/SLO archives (`split_octbuckets_octa_review.py`).
|
| 202 |
+
- `05_taxonomy_and_prompts/`: 95-disease ICD/SNOMED bilingual taxonomy builder (`build_mixed_v6_taxonomy.py`) and prompt de-biasing (`build_clean_captions.py`).
|
| 203 |
+
|
| 204 |
+
Every script contains structured file headers detailing its original purpose, target dataset, authors, inputs, outputs, and dependencies. See [`data_processing/README.md`](https://huggingface.co/datasets/Kaphathy/Dataset/blob/main/data_processing/README.md) for the master script-to-dataset crosswalk matrix.
|
| 205 |
+
|
| 206 |
+
## 8. Quick Start: Extraction & Usage Guide
|
| 207 |
|
| 208 |
### A. Download & Extract Archives (Linux / macOS)
|
| 209 |
To extract multi-part `.tar.zst` archives with maximum parallelism:
|
|
|
|
| 232 |
|
| 233 |
---
|
| 234 |
|
| 235 |
+
## 9. Citation & Acknowledgments
|
| 236 |
|
| 237 |
If you use this benchmark or any subsets in your research, please cite:
|
| 238 |
```bibtex
|
|
|
|
| 245 |
}
|
| 246 |
```
|
| 247 |
|
| 248 |
+
## 10. License & Terms of Use
|
| 249 |
- The aggregated metadata and clinical annotations are made available under **Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)**.
|
| 250 |
- Research use only. Not for direct diagnostic or medical intervention without regulatory clearance.
|