PathoS2VD: From 2D Histopathology to Pseudo Z-Stack Representations
Organizations: School of Electrical Engineering, Korea University, Seoul 02841, Korea · Department of Pathology, Asan Medical Center, University of Ulsan College of Medicine, Seoul 05505, Korea
Abstract
For decades, diagnostic pathology has relied on thin two-dimensional (2D) tissue sections, despite biological tissues exhibiting inherently three-dimensional (3D) morphological organization. Because a single section captures only one focal plane, important morphological cues along the z-axis are inevitably discarded, limiting tissue representation for both human interpretation and computational analysis. In this work, we propose a generative framework that learns z-axis morphological priors from pathology z-stack data, offering a cost-effective alternative to dedicated volumetric imaging, and transfers this knowledge to conventional single-plane histopathology. Specifically, an image-to-video (I2V) latent diffusion model (LDM) is trained to synthesize a pseudo z-stack from a single 2D image, capturing plausible morphological variations across adjacent focal planes rather than reconstructing physical tissue volumes. To enable deployment on large-scale clinical datasets without native z-stack acquisition, we further introduce a prior-preserving mixture fine-tuning strategy that adapts the model to new domains while retaining the learned z-axis morphological prior. Extensive experiments demonstrate that the synthesized pseudo z-stacks exhibit realistic cross-plane morphological continuity and consistently improve downstream prostate cancer classification compared with conventional 2D representations. Our results suggest that learning transferable z-axis priors from limited z-stack data provides a practical way to enrich routine pathology images with complementary morphological information. Code: https://github.com/QuIIL/PathoS2VD.
Figures & tables
| Method | SSIM | LPIPS | FID | FVD | Orientation Similarity | ST Frobenius | Anisotropy MAE | Vz Wasserstein | ST Evolution corr |
| PathoS2VD | 0.6794 | 0.1797 | 13.34 | 31.71 | 0.9407 | 0.7727 | 0.0306 | 0.0234 | 0.8949 |
| Pix2Pix | 0.6331 | 0.2436 | 14.20 | 140.59 | 0.8612 | 257.5114 | 0.0567 | 0.0914 | 0.7707 |
| OpenSoraPlan | 0.5410 | 0.3065 | 42.96 | 87.99 | 0.9025 | 3.5136 | 0.0433 | 0.0505 | 0.8803 |
| Gleason Grade Class | |||||||
| Metric | Stage | Background | Benign | G3 | G4 | G5 | Total |
| SSIM | Zero-shot SVD | 0.6682 | 0.7075 | 0.6644 | 0.6651 | 0.6693 | 0.6339 |
| PathoS2VD | 0.8115 | 0.7884 | 0.7755 | 0.7803 | 0.7556 | 0.7823 | |
| PathoS2VD w/ MFDA | 0.8249 | 0.7940 | 0.7879 | 0.7902 | 0.7607 | 0.7915 | |
| LPIPS | Zero-shot SVD | 0.2498 | 0.2228 | 0.2598 | 0.2488 | 0.2466 | 0.2712 |
| PathoS2VD | 0.1358 | 0.1475 | 0.1548 | 0.1530 | 0.1606 | 0.1504 | |
| Method | AGGC2022 (Patch-level) | SICAPv2 (Slide-level) | ||||
| Acc | W-F1 | M-F1 | Acc | W-F1 | M-F1 | |
| 2D MLP Baseline | 0.8131 0.0008 | 0.8146 0.0024 | 0.6780 0.0015 | 0.7188 0.0255 | 0.6490 0.0267 | 0.4501 0.0495 |
| Max Pool | 0.8234 0.0048 | 0.8247 0.0042 | 0.6834 0.0013 | 0.7812 0.0000 | 0.7449 0.0139 | 0.6273 0.0199 |
| Self Attention | 0.8194 0.0044 | 0.8209 0.0044 | 0.6806 0.0029 | 0.7604 0.0390 | 0.7191 0.0349 | 0.5873 0.0621 |
| DSMIL | 0.8182 0.0026 | 0.8204 0.0015 | 0.6798 0.0013 | 0.7812 0.0255 | 0.7428 0.0679 | 0.6030 0.1317 |
Medical Image Synthesis
Medical image synthesis creates, translates, reconstructs, and edits clinical images to fill missing data and support analysis.
See all 331 papers