# T-SYNTH: synthetic digital breast tomosynthesis and mammography with lesion masks (FDA)

## Facts

- Provider: US FDA Division of Imaging, Diagnostics, and Software Reliability (Wiedeman, Sarmakeeva, Sizikova, Filienko, Lago, Delfino, Badano)
- License: CC0-1.0 (https://huggingface.co/datasets/didsr/tsynth)
- Size: 601 files
- Format: CSV (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/didsr/tsynth
- Access: public, readable without an account
- Category: Health & Biomedical
- Source: https://arxiv.org/abs/2507.04038
- Page: https://scigantic.com/archive/13e209f5-3987-58f9-a877-be70b2eb0a52
- Keywords: Health & Biomedical, image segmentation, CSV, Hugging Face

## About

T-SYNTH is a large open dataset of synthetic breast images from the US FDA's Division of Imaging, Diagnostics, and Software Reliability, generated with physics simulation using the VICTRE (Virtual Imaging Clinical Trial for Regulatory Evaluation) toolkit and MC-GPU Monte Carlo x-ray transport, replicating a Siemens detector. It pairs 2D digital mammography with 3D digital breast tomosynthesis (DBT), with DBT also projected to synthetic C-View images. The card describes 9,000 samples, half lesion-free and half with masses, across four breast density categories (fatty, scattered, heterogeneously dense, dense), mass radii of 5, 7 and 9 mm and three relative mass densities. Lesion masks and bounding boxes are generated automatically from the phantoms, so labels are exact. It complements M-SYNTH, the companion synthetic mammography release. As measured 2026-11-14 the repository uses about 64 GB, with images stored as .raw arrays under data/ (including cview and metadata folders); read them with numpy using the documented dimensions. Used for lesion detection and segmentation in DM and DBT, regulatory-science studies and augmentation of small patient datasets. Released by the FDA authors under CC0 1.0; cite Wiedeman et al. 2025 (arXiv:2507.04038).

## Load the data

```python
import pandas as pd
df = pd.read_csv("hf://datasets/didsr/tsynth/data/embed_metadata/EMBED_split_dbt.csv")
df.head()
```

## Data structure

File types in sampled files: CSV

Example file: `data/embed_metadata/EMBED_split_dbt.csv`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/13e209f5-3987-58f9-a877-be70b2eb0a52
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/13e209f5-3987-58f9-a877-be70b2eb0a52
- Over MCP: `get_schema_card` with id `13e209f5-3987-58f9-a877-be70b2eb0a52` 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)
