Related Experiment Video
Updated: Jun 23, 2026
08:24
Extended Time-lapse Intravital Imaging of Real-time Multicellular Dynamics in the Tumor Microenvironment
Published on: June 12, 2016
8.7K
Cellflow: Advancing pathological image augmentation from spatial views to temporal trajectories
Zeyu Liu1, Tianyi Zhang2, Yufang He1
1School of Biological Science and Medical Engineering, Beihang University, Beijing, 100191, China.
Medical Image Analysis
|February 18, 2026
Summary
Cellflow, a novel temporal-aware generative framework, enhances pathological image analysis by modeling disease progression. This approach improves diagnostic accuracy and data augmentation for computational pathology tasks.
Area of Science:
- Computational Pathology
- Medical Image Analysis
- Artificial Intelligence
Background:
- Deep learning in pathology is limited by scarce annotated data for fine-grained tasks.
- Existing spatial data augmentation methods lack morphological plausibility and temporal awareness.
Purpose of the Study:
- Introduce Cellflow, the first temporal-aware generative framework for pathological image augmentation.
- Model pathological transitions as biologically plausible temporal trajectories.
Main Methods:
- Cellflow uses a stair-based diffusion bridge with classifier-guided probability-flow ODEs.
- Generates intermediate states capturing cellular and tissue-level details.
Main Results:
- Cellflow outperforms spatial augmentation and generative models across 7 diverse datasets.
- Achieved improved classification performance, image fidelity, and temporal coherence.
- Quantitative analysis validated the biological authenticity of generated transition sequences.
Conclusions:
- Cellflow represents a paradigm shift in pathological data augmentation, moving from spatial to temporal modeling.
- Enables robust model training, rare disease exploration, and educational simulations in computational pathology.
