Research map / Image generation / Text-to-image diffusion
PathDiff: Histopathology Image Synthesis with Unpaired Text and Mask Conditions
Mahesh Bhosale, Abdul Wasi, Yuanhao Zhai, Yunjie Tian, Samuel Border, Nan Xi, Pinaki Sarder, Junsong Yuan, David Doermann, Xuan Gong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| jointly train on unpaired text and mask datasets (PathDiff) versus ControlNet for text-conditioned generation | PathDiff · PathCap train (M2I PanNuke) | CLIP-FID: 45.39 -> 18.52 | Improved |
| add PathDiff synthetic images to real training data for nucleus segmentation | CellViT · PanNuke segmentation | Dice: 0.7834 -> 0.8164 | Improved |
| add PathDiff synthetic images to real training data for nucleus segmentation | CellViT · CoNIC segmentation | Dice: 0.7927 -> 0.8356 | Improved |
| add SDM synthetic images to real training data | CellViT · MoNuSAC segmentation | Dice: 0.7089 -> 0.6783 | Hurt |
| train segmentation on PathDiff synthetic images only (ablation) | CellViT · PanNuke segmentation | Dice: 0.7834 -> 0.7311 | Hurt |
| scale synthetic augmentation set beyond 2.5x real train size (ablation) | PathDiff augmentation · PanNuke segmentation | Dice, F1 | No clear effect |
| use silver-standard masks instead of random text-mask pairing (ablation) | PathDiff · PathCap (unified text+mask sampling) | PLIP similarity: 22.09 -> 24.02 | Improved |
| set p_split to 0.5 rather than 0.2 or 0.8 for joint training (ablation) | PathDiff · PanNuke test (mask-only) | CLIP-FID: 7.88 / 8.97 -> 7.28 | Improved |
| set p_split to 0.5 rather than 0.2 for text-only generation (ablation) | PathDiff · PathCap test (text-only) | CLIP-FID: 15.74 -> 19.60 | Hurt |
| add instance edge mask channels to the conditioning mask (depth 3 -> 6) (ablation) | PathDiff · PanNuke test | CONCH FID: 102.37 -> 7.21 | Improved |
| fine-tune VAE on the target datasets (ablation) | PathDiff VAE · VAE reconstruction | FID: 6.94 -> 6.31 | Improved |
| use PathCap-fine-tuned U-Net checkpoint instead of TCGA-BRCA (ablation) | PathDiff · PanNuke mask-to-image | CLIP-FID: 14.44 -> 7.21 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Semantic and Visual Crop-Guided Diffusion Models for Heterogeneous Tissue Synthesis in Histopathology (NeurIPS 2025 · similarity 0.90)
- Dynamic Entity-Masked Graph Diffusion Model for Histopathology Image Representation Learning (AAAI 2025 · similarity 0.89)
- STREAM: Stochastic Riemannian Flow Matching with Anisotropic Decoder for Digital Histopathology Image Generation (NeurIPS 2026 · similarity 0.89)
- Beyond ViT Tokens: Masked-Diffusion Pretrained Convolutional Pathology Foundation Model for Cell-Level Dense Prediction (NeurIPS 2026 · similarity 0.87)
- PathGen-1.6M: 1.6 Million Pathology Image-text Pairs Generation through Multi-agent Collaboration (ICLR 2025 · similarity 0.87)
- Learned Representation-Guided Diffusion Models for Large-Image Generation (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
