cs.CVOct 9, 2026

BigCell: Generating Gigapixel Whole-Slide Images

Authors: Srikar Yellapragada, Alexandros Graikos, Zilinghan Li, Kostas Triaridis, Tarak Nath Nandi, Ji Dong K Bai, Beatrice Knudsen, Tahsin Kurc, 

Organizations: Stony Brook University · Argonne National Laboratory · University of Chicago · The University of Utah · 

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Whole-slide images (WSIs) are gigapixel-scale digital scans of histopathology tissue and serve as the primary basis for cancer diagnosis. Existing histopathology generative models operate at the patch scale, producing fixed-size tiles of at most 1024×\times1024 pixels which encompass only a small fraction of the WSI. In contrast, text reports for clinical diagnoses and slide-level diagnostic labels -- the supervision that exists at scale -- operate at the whole-slide level. We present BigCell, the first approach to generate coherent whole-slide histopathology images at native gigapixel resolution. BigCell trains a flow-matching transformer at 2048×\times2048 pixels on over 120,000 publicly available WSIs spanning multiple organ types. Conditioning on embeddings from pathology foundation models and using an inference-time sliding-window pipeline, BigCell generates coherent images of up to 32,768×\times32,768 pixels, outperforming prior tile generators on both tile-level and slide-level fidelity metrics. To make gigapixel synthesis practical, we introduce an adaptive scheduling strategy that prioritizes important slide regions, speeding up image generation by over 3\ttimes with minimal quality degradation. Furthermore, we propose the first text-to-image synthesis at 8k resolution by training a separate model that produces dense conditioning grids directly from pathology text reports. We demonstrate that augmenting with synthetic samples improves slide-level few-shot classification performance by up to 18%18 \% over baselines trained on real data.

Figures & tables

Figure 1: We present synthetic images generated using BigCell. Left: 32768 ×\times 32768 images along with 2048 ×\times 2048 crops, highlighting the realism achieved both at the macro-scale and cellular detail level. Right: Two 8192 ×\times 8192 images per report generated from text. Best viewed zoomed in.
Figure 2: Overview of the BigCell pipeline. (a) We obtain a dense grid of conditioning embeddings either from a reference WSI or sampled using a model conditioned on text (CGM). This serves as the controllable blueprint that defines the large-scale structures of the whole-slide image. (b) The tile generator uses the local conditioning to predict how to transform the slide locally. (c) Using the tile generator on overlapping windows, we aggregate the per-window velocities into a global velocity field and iteratively denoise the noisy slide into a gigapixel image, allowing us to efficiently synthesize WSIs without explicitly training on their scale.
 Slide-level SimilarityPatch-level SimilarityFréchet Distance
GeneratorTITAN↑MOOZY↑V2 mean ↑V2 5th ↑H1 mean ↑H1 5th ↑V2 FD ↓H1 FD ↓
PixCell. 941. 900. 778. 632. 667. 5091023696
CytoSyn.929.864.753.502.640.457895604
MuPaD.646.755.462.258.342.196981641
LRDM.531.693.513.334.395.247731502
ZoomLDM.708.801.646.464.477.3258 155 53
Table 1: Comparison at 8192 ×\times 8192 resolution. V2: Virchow-2; H1: H-Optimus-1; FD: Fréchet distance (no test). ∗ : p<0.001p<0.001 , two-sided paired Wilcoxon vs. best baseline ( N=100N=100 ). Resolution- and data-matched comparisons are in Appendix E.2 .
GeneratorSame organ↑Same diagnosis↑All three fields↑
PixCell6 3%4 2%3 8%
CytoSyn54%26%25%
MuPaD41%17%17%
ZoomLDM61%40%36%
LRDM59%31%30%
BigCell69%46%41%
Table 2: Report-level agreement at 8192 ×\times 8192 resolution ( N=100N=100 , same ROIs as Table 1 ): fraction of generated/reference pairs whose MedGemma-4B reports agree on the tissue of origin, the diagnosis, and all three fields (organ, malignancy, diagnosis).
Figure 3: Adaptive window scheduling compared with simple, fewer-step sampling and the full 38440 NFE baseline at 32k resolution (N=256), evaluated at r∈{0.05,0.09,0.25,0.41}r\in\{0.05,0.09,0.25,0.41\} . At r=0.25r=0.25 , we obtain a 3.1 ×\times speedup (15.6 h →\rightarrow 4.97 h on a single Intel Data Center GPU Max 1550).
 TITAN PRISM
MethodSimRecall@1Recall@5 SimRecall@1Recall@5
Real data (paired)0.5510.2110.655 0.4990.1480.430
BigCell0.4870.1690.655 0.4540.1060.387
Table 3: TITAN and PRISM CLIP-style similarity and report-to-image retrieval recall on 150 held-out reports. CGM-generated images approach the alignment of real data under both encoders.
Figure 7
Figure 5: Synthetic 8192 ×\times 8192 image generated by BigCell from a pathology text report, showing extensive pancreatic ductal adenocarcinoma.
Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

StepTensor2048 ×\times 2048 tile32,768 ×\times 32,768 canvas
Input / output imageimage xx2048×2048×32048\times 2048\times 332768×32768×332768\times 32768\times 3
VAE encode / decodelatent zz256×256×32256\times 256\times 324096×4096×324096\times 4096\times 32
UNI-2 on 256 ×\times 256 patchesconditioning grid yy8×8×15368\times 8\times 1536128×128×1536128\times 128\times 1536
Denoising (per window)velocity vθ(zt,t,y)v_{\theta}(z_{t},t,y)256×256×32256\times 256\times 32256×256×32256\times 256\times 32
Appendix
Table 5: Tensor shapes in BigCell for a single 2048 ×\times 2048 tile and a 32,768 ×\times 32,768 canvas. At 32k, each sliding window operates on a 256×256×32256\times 256\times 32 latent region with the corresponding 8×8×15368\times 8\times 1536 slice of the conditioning grid.
ResolutionBatch sizeConditioninglrt schedule
25648001×15361\times 15362e-4Uniform
51224004×15364\times 15361e-4Uniform
102472016×153616\times 15361e-4Shifted, α=3\alpha=3
204836064×153664\times 15361e-4Shifted, α=3\alpha=3
Appendix
Table 6: Progressive training schedule for BigCell tile . We train for one epoch at each resolution, and initialize the model from the previous stage weights.
ComponentBigCell tileCGM
OptimizerAdamWAdamW
β1,β2\beta_{1},\beta_{2}0.9, 0.9990.9, 0.95
Weight decay00
EMA decay0.99990.9995
Gradient clipping1.01.0
Warmup5,000 steps (linear)40 epochs (linear)
Appendix
Table 7: Training configuration for BigCell tile and CGM
ResolutionBatch sizeSec/IterCompute/Sample
10247201.351.0 ×\times
20483605.598.3 ×\times
40963606596 ×\times
Appendix
Table 8: Training cost across native resolutions, measured on the Intel GPU cluster. Cost scales quadratically, making training at 4096 impractical. The 4096 row is a single-iteration benchmark; we did not train a model at this resolution.
Figure 6: Effect of tile generator’s native resolution on 8192 ×\times 8192 generation quality. Both slide-level and tile-level metrics improve with resolution, with diminishing returns beyond 1024.
 Slide-level SimilarityPatch-level SimilarityFréchet Distance
GeneratorTITAN↑MOOZY↑V2 mean ↑H1 mean ↑V2 FD ↓H1 FD ↓
Matched native resolution
PixCell-1024.941.900.778.6671023696
BigCell-1024.952.971.849.723912617
Matched native resolution, data, and compute
BigCell-summarizer-2048.926.868.627.5051229813
Appendix
Table 9: Resolution-matched comparisons at 8192 ×\times 8192 resolution ( N=100N=100 , same ROIs as Table 1 ). Top: BigCell and PixCell at the same 1024 native resolution. Bottom: BigCell against a model with identical backbone, data, sampler, and training budget that uses ZoomLDM’s summarizer-based conditioning.
Figure 7: Effect of window selection strategy on 8192 ×\times 8192 generation quality. Our CFG importance-based strategy outperforms random selection across all metrics.
Field agreementBigCell vs. referenceChance
Same tissue of origin85.9% [81.1, 89.6]9.1% [7.7, 10.7]
Same diagnosis72.2% [66.4, 77.3]9.2% [7.4, 11.1]
All three fields identical71.0% [65.1, 76.2]6.0% [4.6, 7.7]
Appendix
Table 10: Report-level agreement at 32,768 ×\times 32,768 resolution ( N=256N=256 , full compute). Chance is the rate at which a real report matches the synthetic report of a different slide, averaged over all such pairings. 95% Wilson intervals in brackets.
Figure 8: Pathologist evaluation interface. Each trial shows an 8192 ×\times 8192 image (real or synthetic) and four candidate reports (one of which is true). Pathologists indicate matching report(s) and classify the image as real or synthetic.
Figure 9: Qualitative comparison of 8192 ×\times 8192 generation between BigCell tile and baselines. Top row: full 8k image. Bottom row: 2k crop from the boxed region.
Figure 10: Visualization of CFG-magnitude importance scores on two synthetic 8192 ×\times 8192 images. Tissue regions receive high importance, while background regions receive low importance.
Figure 11: 32,768 ×\times 32,768 pixel breast tissue image generated by BigCell (top), with progressive zoom-in crops from the marked regions (bottom).
Figure 12: 32,768 ×\times 32,768 pixel lung tissue image generated by BigCell (top), with progressive zoom-in crops from the marked regions (bottom).
Figure 13: 32,768 ×\times 32,768 pixel mesothelial tissue image generated by BigCell (top), with progressive zoom-in crops from the marked regions (bottom).
More in this topic

Image Generation

Image generation studies models that create or edit images from text, examples, layouts, and other structured controls.

See all 461 papers

Explore similar work

CardsList
Aug 4, 2026

From Multi-Resolution Cells to Gigapixel Whole Slide Images Foundation Model for Computational Pathology

Vision Transformers (ViTs) and their hierarchical variants have achieved strong performance in Computational Pathology (CPath). However, most are pre-trained on single-resolution Whole Slide Images (WSIs), limiting their generalization across arbitrary resolutions. Gigapixel WSIs inherently contain diagnostic patterns at multiple scales, including cellular morphologies, tissue architectures, and global context, mirroring how expert pathologists examine WSIs. We introduce Multi-Resolution Pyramid Transformer (MRPT), a model that hierarchically aggregates multi-resolution information from cellular to tissue and WSI levels. MRPT employs a biologically meaningful Consecutive Cross-Resolution Attention (CCRA) mechanism to capture scale-independent interactions and enforces multi-resolution semantic consistency by aligning embeddings across resolutions, yielding robust and generalizable WSI representations. Pre-trained in a multi-resolution self-supervised manner on 624M patches, 2.4M regions, and 36K WSIs, MRPT learns rich coarse-to-fine histopathology features. Extensive experiments on 34 diverse datasets show that MRPT surpasses recent foundation models and Multimodal Large Language Models (MLLMs) in cancer subtype classification, tissue phenotyping, and Visual Question Answering (VQA) for WSI understanding.

Basit Alawode, Moshira Ali Abdalla, Dwarikanath Mahapatra +2

Khalifa University of Science and Technology, UAE · University of Western Australia, Australia

Representation LearningVision TransformerPathology Foundation ModelsHistopathology Image ClassificationWhole-Slide Image Analysis
AskPDF
Mar 24, 2026

MLLM-HWSI: A Multimodal Large Language Model for Hierarchical Whole Slide Image Understanding

Whole Slide Images (WSIs) exhibit hierarchical structure, where diagnostic information emerges from cellular morphology, regional tissue organization, and global context. Existing Computational Pathology (CPath) Multimodal Large Language Models (MLLMs) typically compress an entire WSI into a single embedding, which hinders fine-grained grounding and ignores how pathologists synthesize evidence across different scales. We introduce MLLM-HWSI, a Hierarchical WSI-level MLLM that aligns visual features with pathology language at four distinct scales, cell as word, patch as phrase, region as sentence, and WSI as paragraph to support interpretable evidence-grounded reasoning. MLLM-HWSI decomposes each WSI into multi-scale embeddings with scale-specific projectors and jointly enforces (i) a hierarchical contrastive objective and (ii) a cross-scale consistency loss, preserving semantic coherence from cells to the WSI. We compute diagnostically relevant patches and aggregate segmented cell embeddings into a compact cellular token per-patch using a lightweight Cell-Cell Attention Fusion (CCAF) transformer. The projected multi-scale tokens are fused with text tokens and fed to an instruction-tuned LLM for open-ended reasoning, VQA, report, and caption generation tasks. Trained in three stages, MLLM-HWSI achieves new SOTA results on 13 WSI-level benchmarks across six CPath tasks. By aligning language with multi-scale visual evidence, MLLM-HWSI provides accurate, interpretable outputs that mirror diagnostic workflows and advance holistic WSI understanding. Code is available at: GitHub.

Basit Alawode, Arif Mahmood, Muaz Khalifa Al-Radi +6

Department of Computer Science, Khalifa University of Science and Technology, UAE. · Information Technology University, Pakistan. · KAU, KSA. +1

Whole-Slide Image AnalysisMedical VLMs
AskPDF
Aug 31, 2026

SlideMix: Enhancing Whole Slide Image Analysis via Multimodal Shuffling

Histopathological whole slide images (WSIs) are central to cancer diagnosis, but their gigapixel scale, tissue heterogeneity, weak slide-level supervision, sparse diagnostic regions, and multi-scale evidence make robust automated analysis challenging. Multiple instance learning (MIL) is widely used to aggregate tile-level features into slide-level predictions, yet existing augmentation strategies often perturb tissue regions without preserving diagnostic relevance, slide context, or cross-scale structure. We propose SlideMix, a model-agnostic multimodal augmentation framework for MIL-based WSI analysis. SlideMix uses a retrieval-augmented vision-language model (VLM)-based Visual-Language Adaptive Region selector to identify diagnostically relevant regions and reduce weak-label noise. It then performs In-place Tile Shuffling within meaningful tissue regions to mix feature embeddings while preserving slide-level context. A VLM-based soft-labeling module supervises mixed samples, while a multi-factor, loss-driven online Curriculum-Learning Feedback scheme adaptively controls shuffle granularity, feature similarity, and shuffle ratio to promote cross-scale representation learning. Across 11 WSI datasets comprising 20,523 slides, 8 diagnostic tasks, and 10 WSI backbones, SlideMix improves accuracy and generalization in most settings and compares favorably with established augmentation baselines, providing a simple plug-and-play approach for more robust and scalable digital pathology models. Source code: https://github.com/Xia-Research-Lab/SlideMix

Chad Wong, Sicheng Chen, Tianyi Zhang +4

1Nhu Department of Electrical Engineering and Computer Science, University of California, Irvine · Department of Electrical & Computer Engineering, National University of Singapore · 3PuzzleLogic Pte Ltd, Singapore 229594, Singapore

Data AugmentationHistopathology Image ClassificationMultiple Instance LearningWhole-Slide Image Analysis
AskPDF