Update README.md
Browse files
README.md
CHANGED
|
@@ -30,7 +30,7 @@ arxiv: 2509.17847
|
|
| 30 |
|
| 31 |
[Saghir Alfasly](https://saghiralfasly.github.io/) · [Wataru Uegami](https://www.linkedin.com/in/wataru-uegami-8b106920a/) · [MD Enamul Hoq](https://www.linkedin.com/in/mhoq89/) · [Ghazal Alabtah](https://www.linkedin.com/in/ghazal-alabtah-00/) · [H.R. Tizhoosh](https://tizhoosh.com/)
|
| 32 |
|
| 33 |
-
KIMIA Lab, Mayo Clinic, Rochester, MN, USA
|
| 34 |
|
| 35 |
[](https://arxiv.org/abs/2509.17847)
|
| 36 |
[](https://kimialabmayo.github.io/hetero_tissue_diffuse_page/)
|
|
@@ -38,10 +38,6 @@ KIMIA Lab, Mayo Clinic, Rochester, MN, USA
|
|
| 38 |
|
| 39 |
---
|
| 40 |
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
---
|
| 44 |
-
|
| 45 |
## Model Description
|
| 46 |
|
| 47 |
**HeteroTissueDiffuse** is a latent diffusion model (LDM) that synthesizes heterogeneous histopathology images by conditioning on both a **binary semantic map** and **raw tissue crop exemplars**. Unlike text- or embedding-guided approaches, it injects actual tissue appearance directly into the diffusion process, preserving staining characteristics, nuclear morphology, and cellular texture.
|
|
@@ -152,13 +148,6 @@ python inference_heteroTissueDiffuse_camelyon.py \
|
|
| 152 |
|
| 153 |
## Performance
|
| 154 |
|
| 155 |
-
### Fréchet Distance (lower is better)
|
| 156 |
-
|
| 157 |
-
| Method | Camelyon16 | PANDA | TCGA |
|
| 158 |
-
|--------|-----------|-------|------|
|
| 159 |
-
| Unconditional LDM | 430.1 | — | — |
|
| 160 |
-
| **HeteroTissueDiffuse (ours)** | **72.0** | ↓ 2–3× | ↓ 2–3× |
|
| 161 |
-
|
| 162 |
### Downstream Segmentation (IoU)
|
| 163 |
|
| 164 |
| Training data | Camelyon16 | PANDA |
|
|
@@ -182,11 +171,6 @@ A certified pathologist evaluated 120 images in a blinded study. Synthetic image
|
|
| 182 |
- **Privacy-preserving data sharing**: synthetic data as a substitute for patient slides
|
| 183 |
- **Education**: illustrating tissue morphology variations
|
| 184 |
|
| 185 |
-
### Out-of-scope use
|
| 186 |
-
|
| 187 |
-
- Clinical diagnosis or patient care — this is a research model
|
| 188 |
-
- Generating images to deceive or misrepresent clinical findings
|
| 189 |
-
|
| 190 |
---
|
| 191 |
|
| 192 |
## Training Details
|
|
|
|
| 30 |
|
| 31 |
[Saghir Alfasly](https://saghiralfasly.github.io/) · [Wataru Uegami](https://www.linkedin.com/in/wataru-uegami-8b106920a/) · [MD Enamul Hoq](https://www.linkedin.com/in/mhoq89/) · [Ghazal Alabtah](https://www.linkedin.com/in/ghazal-alabtah-00/) · [H.R. Tizhoosh](https://tizhoosh.com/)
|
| 32 |
|
| 33 |
+
KIMIA Lab, Department of AI & Informatics, Mayo Clinic, Rochester, MN, USA
|
| 34 |
|
| 35 |
[](https://arxiv.org/abs/2509.17847)
|
| 36 |
[](https://kimialabmayo.github.io/hetero_tissue_diffuse_page/)
|
|
|
|
| 38 |
|
| 39 |
---
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
## Model Description
|
| 42 |
|
| 43 |
**HeteroTissueDiffuse** is a latent diffusion model (LDM) that synthesizes heterogeneous histopathology images by conditioning on both a **binary semantic map** and **raw tissue crop exemplars**. Unlike text- or embedding-guided approaches, it injects actual tissue appearance directly into the diffusion process, preserving staining characteristics, nuclear morphology, and cellular texture.
|
|
|
|
| 148 |
|
| 149 |
## Performance
|
| 150 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
### Downstream Segmentation (IoU)
|
| 152 |
|
| 153 |
| Training data | Camelyon16 | PANDA |
|
|
|
|
| 171 |
- **Privacy-preserving data sharing**: synthetic data as a substitute for patient slides
|
| 172 |
- **Education**: illustrating tissue morphology variations
|
| 173 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 174 |
---
|
| 175 |
|
| 176 |
## Training Details
|