Unpaired Multi-Domain Histopathology Virtual Staining Using Dual Path Prompted Inversion
Authors: Bing Xiong, Yue Peng, Ranran Zhang, Fuqiang Chen, Jiaye He, Wenjian Qin
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive evaluations on publicly available multi-domain unpaired staining datasets demonstrate high structural consistency and accurate style transfer results. |
| Researcher Affiliation | Academia | 1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, She Zhen 518000,China 2University of Chinese Academy of Sciences, Beijing 100000,China 3National innovation center for advanced medical devices, She Zhen 518000,China EMAIL |
| Pseudocode | Yes | Algorithm 1: Dual Path Prompting Optimization Algorithm |
| Open Source Code | Yes | Code github.com/Diana Nerual Network/Stain Prompt Inversion |
| Open Datasets | Yes | In the ANHIR dataset (Borovec et al. 2020), we used kidney tissue slices, each set containing four consecutive slices stained with H&E, MAS, PAS, or PASM. |
| Dataset Splits | Yes | We used whole slide images of kidney tissue from patients 1-4 at 40x magnification for training, and patient 5 for testing. ... The training set includes 42,167 images (7,688 H&E, 12,132 MAS, 11,458 PAS, and 10,889 PASM stained). The test set contains 1,989 H&E, 2,062 MAS, 1,900 PAS, and 2,119 PASM stained images. |
| Hardware Specification | Yes | Our model was implemented in Python using Py Torch on an Ubuntu server with 48 GB memory and an NVIDIA A6000 GPU. |
| Software Dependencies | No | The paper mentions 'Python using Py Torch' but does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | The diffusion model was trained for 80k iterations with a batch size of 8 and a learning rate of 1 10 4 using the ADAMW optimizer. For validation, the batch size was set to 1, and DDIM inversion timesteps were set to 100. We set the structural loss function constants c1 and c2 to 1 10 8 and ISTinit to 50. ... α was set to 0.75 for H&E to MAS staining and 0.55 for H&E to PAS staining. |