Recover Cell Tensor: Diffusion-Equivalent Tensor Completion for Fluorescence Microscopy Imaging
Chenwei Wang, Zhaoke Huang, Zelin Li, Wenqi Zhu
TL;DR
This paper introduces a novel tensor completion framework for fluorescence microscopy imaging that effectively reconstructs 3D cell volumes from sparse, noisy data, leveraging theoretical bounds and a score-based generative model.
Contribution
It proposes a new tensor completion method tailored for FM imaging, incorporating structural priors and a generative model, with theoretical validation and state-of-the-art results.
Findings
Achieves significant improvements in signal-to-noise ratio.
Demonstrates high structural fidelity in reconstructions.
Validates the theoretical lower bound for tensor completion.
Abstract
Fluorescence microscopy (FM) imaging is a fundamental technique for observing live cell division, one of the most essential processes in the cycle of life and death. Observing 3D live cells requires scanning through the cell volume while minimizing lethal phototoxicity. That limits acquisition time and results in sparsely sampled volumes with anisotropic resolution and high noise. Existing image restoration methods, primarily based on inverse problem modeling, assume known and stable degradation processes and struggle under such conditions, especially in the absence of high-quality reference volumes. In this paper, from a new perspective, we propose a novel tensor completion framework tailored to the nature of FM imaging, which inherently involves nonlinear signal degradation and incomplete observations. Specifically, FM imaging with equidistant Z-axis sampling is essentially a tensor…
Peer Reviews
Decision·ICLR 2026 Poster
1. The paper establishes a novel mathematical equivalence between Tucker-based tensor completion and conditional diffusion models, providing rigorous theoretical guarantees with clear derivations for exact recovery under sparse sampling conditions inherent to fluorescence microscopy. 2.The method achieves robust 3D cell reconstruction without requiring paired high-resolution ground truth data, instead leveraging low-rank structure and sparse noise decomposition directly from incomplete noisy obs
1. Since the paper focuses on biological image reconstruction, relying solely on visual quality metrics lacks sufficient persuasiveness—high PSNR does not guarantee biological correctness, as reconstructions may appear visually plausible yet contain biologically inaccurate structures. Based on Figure 3 and other renderings, the cycle+IPG method gives me a better overall impression, and it performs better in reproducing details compared to the method proposed in this paper. 2. The paper lacks cr
- The paper reframes 3D fluorescence microscopy reconstruction as cell-tensor completion rather than a traditional inverse problem, and establishes a provable connection between low-rank tensor recovery (via ADMM) and conditional score-based diffusion dynamics. This theoretical bridge is genuinely novel and goes beyond heuristic model design, giving a principled interpretation of diffusion priors in biological imaging. - The paper provides clear recovery guarantees under missing-data and sparse
- The paper highlights the diffusion-equivalence result but does not compare against recent unsupervised or self-supervised generative FM restoration methods (e.g., score-based microscopy denoising, inverse-consistent diffusion models, diffusion-based deconvolution pipelines). Without such comparisons, it is hard to quantify whether the proposed tensor-based approach benefits primarily from low-rank structure or from the implicit generative prior interpretation. Adding baselines like self-superv
1. Well-written and well-organized, with comprehensive supplementary material. 2. Introduces a tensor completion approach specifically tailored to fluorescence microscopy imaging, addressing a novel research problem that has not been extensively explored in the literature. 3. Derives the theoretical lower bound for exact 3D cell tensor recovery and reformulates tensor completion into a score-based generative modeling framework.
### Mismatch Between Noise Model and Physical Imaging Process In the manuscript, the observation is formulated as Y_\Omega = X_\Omega + E_\Omega (an additive sparse noise model), which is a convenient and commonly used formulation. However, in fluorescence microscopy, the primary degradations typically involve **Poisson noise** (due to photon counting), **system PSF** (blur or convolution), **signal attenuation and scattering**, and even **multiplicative effects**. Therefore, such a simp
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsCell Image Analysis Techniques · Advanced Fluorescence Microscopy Techniques · Digital Holography and Microscopy