Recover Cell Tensor: Diffusion-Equivalent Tensor Completion for Fluorescence Microscopy Imaging

Chenwei Wang, Zhaoke Huang, Zelin Li, Wenqi Zhu · 2026

Read Recover Cell Tensor: Diffusion-Equivalent Tensor Completion for Fluorescence Microscopy Imaging by Chenwei Wang, Zhaoke Huang, Zelin Li, Wenqi Zhu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-19169.txt.

Introduction

Chenwei Wang and three co-authors rebuild missing slices of live cell images with a tensor method. Fluorescence microscopy lets biologists watch cells divide in 3D, a basic life process. But scanning the full volume slowly would poison the cells with light, so scans stay sparse, uneven, and noisy.

Their method, Recover Cell Tensor, fills the gaps with a completion scheme tied to how diffusion spreads. It treats the image stack as a tensor, a multi way grid of numbers. Fuller volumes from gentler scans would let labs watch division longer without harming the very cells they study.

Catalog record

CREATORChenwei Wang, Zhaoke Huang, Zelin Li, Wenqi Zhu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE73 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-19169

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Recover Cell Tensor: Diffusion-Equivalent Tensor Completion for Fluorescence Microscopy Imaging Chenwei Wang, Zhaoke Huang, Zelin Li, Wenqi Zhu 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 completion task under a uniformly random sampling condition. On one hand, we derive the theoretical lower bound for exact cell tensor completion, validating the feasibility of accurately recovering 3D cell tensor. On the other hand, we reformulate the tensor completion problem as a mathematically equivalent score-based generative model. By incorporating structural consistency priors, the generative trajectory is effectively guided toward denoised and geometrically coherent reconstructions. Our method demonstrates state-of-the-art performance on SR-CACO-2 and three real in vivo cellular datasets, showing substantial improvements in both signal-to-noise ratio and structural fidelity. 1 Introduction Cell division is one of the most fundamental processes of life–it drives growth and reproduction, yet also sets the stage for aging and death sheldrake1974ageing ; winter2024shr ; li2025volume . Observing the 3D membrane structure of cells during live cell division is essential for advancing biological research chung2013structural , which requires scanning through cell volume alone Z-axis with the aid of fluorescent dyes. To minimize lethal phototoxicity, the scanning time is strictly limited by XY-plane imaging at at equidistant intervals along the Z-axis. As a result, 3D live-cell volume faces challenges like anisotropic resolution, and spatially varying noise, as illustrated in Fig. 1 . These constraints significantly compromise the spatial-temporal resolution of live-cell volumetric imaging, thereby impeding fundamental insights in a wide range of biomedical fields voleti2019real ; li2025volume ; wang2025limited . Current most of image recovery algorithms are based on the inverse problem wang2020deep , which aims to model the degradation mapping even the the degradation process is unknown. As discussed in prior studies, many of these methods rely on supervised learning with known degradation models, requiring large paired datasets of high- and low-quality images, such as SRCNN 7115171 , FSRCNN 10.1007/978-3-319-46475-6_25 ; wang2023entropy , and DDPM NEURIPS2020_4c5bcfec . These approaches are less suited to 3D fluorescence microscopy, where high-quality references are not available. To mitigate the reliance on paired data, unsupervised methods like CycleGAN Zhu_2017_ICCV , CinCGAN Yuan_2018_CVPR_Workshops , and Deep Image Prior Ulyanov_2018_CVPR have been proposed. Additionally, recent efforts using self-supervised learning Ning2023 ; Qu2024 ; Park2022 and transformer-based architectures Li2023 ; WANG2024227 ; wang2023sar1 ; wang2022global have shown promise. A more detailed discussion of related work is provided in Appendix C . However, FM imaging involves a nonlinear and complex imaging process influenced by excitation, fluorescence emission and sample heterogeneity lichtman2005fluorescence ; xu2024multiphoton ; wang2022semi . This process is further complicated by high noise levels and incomplete signal acquisition due to attenuation and scattering. Traditional inverse problem methods, which rely on linear models and known physical assumptions sol2024experimentally ; mo2023quantitative ; hui2024microdiffusion ; wang2020deep ; wang2022sar , struggle to accurately capture the underlying core pattern of 3D live-cell with such complexity. Under noisy or incomplete conditions with equidistant Z-axis sampling, these methods often become ill-posed and produce unstable or ambiguous reconstructions yu2019single ; genzel2022solving , as shown in Fig. 1 , which limits their reliability for life science applications. The details are presented in Section 2 . In this paper, we first highlight that most image restoration approaches grounded in inverse problem formulations may adversely affect the accuracy and fidelity of 3D cell recovery. Then, we propose a cell tensor completion model, derive the cell-recovery bound and uncover its equivalence to a conditional diffusion framework to counter the evil. Specifically, our contributions are summarized as follows. \bullet From a new perspective, we introduce a tensor completion model targeted for 3D FM imaging. This model not only aligns with the nature of FM imaging, like nonlinearity and incompletion, but also offers a principled framework for capturing the underlying structural patterns of cell volumes, enabling accurate 3D cell recovery. Furthermore, we derive the lower bound for exact 3D cell completion based on the proposed cell tensor completion model. \bullet Through solving the proposed cell tensor completion model, we derive a mathematically equivalent score-based diffusion model by reformulating the low-rank tensor recovery as a conditional generative process. This model is further guided by structural consistency priors to steer the generative trajectory toward accurate and denoised 3D FM reconstruction. \bullet Evaluations on SR-CACO-2 and three real in vivo cellular datasets confirm that our method outperforms existing approaches, with clear gains in both SNR and structural fidelity. 2 Problem Formulations 2.1 3D FM Live Cell Imaging FM imaging is a complex process involving light excitation, fluorescence emission, and diffraction through optical components lichtman2005fluorescence , all of which are nonlinear and influenced by diverse optical and biological interactions xu2024multiphoton . It is further challenged by high noise levels–such as optical noise, background interference, and low SNR–especially under low-light or high-speed conditions li2023real . Additionally, due to sample heterogeneity and signal attenuation or scattering, the captured data are often incomplete pinkard2021learned . An illustration of the FM imaging process is shown in Fig. 1 . Figure 1: Illustration of the FM imaging process and comparison of reconstructed results across different methods. The schematic (left) shows the fluorescence imaging pipeline, where excitation light passes through optical components and interacts with the specimen before being detected. Raw observations are corrupted by noise and optical degradation. Reconstructed images using inverse-problem-based methods (e.g., IPG, CycleGAN) often suffer from hallucinated structures and residual noise. Most of image restoration methods based on inverse problem modeling typically assume linearity and well-defined physical models sol2024experimentally , which fail to capture the nonlinearities and sample complexity inherent in FM imaging hui2024microdiffusion . Under noisy conditions, these models often become unstable antun2020instabilities , where minor input perturbations lead to large reconstruction errors genzel2022solving . Such limitations result in unreliable reconstructions, which are unacceptable in life science applications, as demonstrated in Fig. 1 . \hat{I}_{y}=\mathcal{F}(I_{x};\theta), (1) where I_{x} denotes the low-resolution image, \hat{I}_{y} is the high-resolution estimate, and \mathcal{F}(\cdot) represents the restoration model, approximating the inverse of the degradation process \mathcal{D}(\cdot) . Although \mathcal{D} is typically unknown and influenced by various factors, it is often simplified as a single downsampling operation: \mathcal{D}(I_{y};\delta)=(I_{y})\downarrow_{s} , where s\subset\delta and \downarrow_{s} denotes downsampling with scale factor s , typically via bicubic interpolation with antialiasing wang2020deep . Compared to inverse problem-based methods, tensor completion offers key advantages for FM imaging. In 3D FM cell scenarios with the nature of nonlinearity and incompletion, tensor completion effectively leverages the low-rank structure of multi-dimensional data to recover missing information goldfarb2014robust , with inherent robustness to noise wang2025tensor , where inverse methods often suffer from instability. As

Read the complete text