# S-SYNTH: synthetic dermoscopy-style skin lesion images with segmentation masks (FDA)

## Facts

- Provider: US FDA Division of Imaging, Diagnostics, and Software Reliability (Kim, Saharkhiz, Sizikova, Lago, Sahiner, Delfino, Badano)
- License: CC0-1.0 (https://huggingface.co/datasets/didsr/ssynth_data)
- Size: 1,029 files
- Format: CSV (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/didsr/ssynth_data
- Access: public, readable without an account
- Category: Health & Biomedical
- Source: https://arxiv.org/abs/2408.00191
- Page: https://scigantic.com/archive/5bddb473-a761-5cf3-b406-a49ab7e151d0
- Keywords: Health & Biomedical, image segmentation, CSV, Hugging Face

## About

S-SYNTH is a knowledge-based, open-source skin simulation framework and pre-generated dataset from the US Food and Drug Administration's Division of Imaging, Diagnostics, and Software Reliability. It builds anatomically inspired multi-layer, multi-component 3D skin models with a growing lesion and digitally renders them, giving synthetic skin images with pixel-level lesion segmentation masks under controlled variation of skin colour (melanosome fraction), hair, blood fraction, lesion shape, lesion material, lighting and rotation. The authors show that comparative AI performance trends on S-SYNTH follow those on real dermatologic images, so it can be used to study bias across skin tones and to augment small patient datasets. Folders encode the generating parameters (skin model, hair model, melanosome and blood fractions, lesion, timepoint, light), and each leaf holds image.png, mask.png, cropped versions and crop_size.txt. As measured 2026-11-14 the repository uses about 327 GB under data/, with dataset_splits and outputs folders; load with the datasets library (load_dataset('didsr/ssynth_data')) or read PNGs directly. The FDA authors release it under CC0 1.0; cite Kim et al. 2024 (arXiv:2408.00191).

## Load the data

```python
import pandas as pd
df = pd.read_csv("hf://datasets/didsr/ssynth_data/data/outputs/segmentation_results/artifacts/MODEL-dermosegdiff--all_tones_real_ISIC_1.0_add_synth_0.0_None_10k--00--dsd_i01--n-dsd_i01_s-128_b-32_t-250_sc-linear_best.pth-DATASET-all_tones_real_ISIC_1.0_add_synth_0.0_None_10k--test_images.csv")
df.head()
```

## Data structure

File types in sampled files: CSV

Example file: `data/outputs/segmentation_results/artifacts/MODEL-dermosegdiff--all_tones_real_ISIC_1.0_add_synth_0.0_None_10k--00--dsd_i01--n-dsd_i01_s-128_b-32_t-250_sc-linear_best.pth-DATASET-all_tones_real_ISIC_1.0_add_synth_0.0_None_10k--test_images.csv`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/5bddb473-a761-5cf3-b406-a49ab7e151d0
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/5bddb473-a761-5cf3-b406-a49ab7e151d0
- Over MCP: `get_schema_card` with id `5bddb473-a761-5cf3-b406-a49ab7e151d0` at https://api.scigantic.com/mcp

## Related topics

- [CSV datasets](https://scigantic.com/topics/csv)
- [Image segmentation datasets](https://scigantic.com/topics/image-segmentation)
- [Health & Biomedical datasets](https://scigantic.com/topics/health-and-biomedical)
