Live-cell microscopy restoration is constrained by a trade-off between inference latency and texture preservation. While diffusion models provide high textural fidelity, the computational cost of iterative sampling currently limits their use in low-latency instrument feedback loops. Here, we present NAFNet GAN, a restoration framework that couples an activation-free backbone with a perceptual adversarial objective to enable high-throughput analysis. Unlike diffusion architectures, NAFNet GAN achieves an inference latency of ~110 ms for 1024x1024 inputs, potentially suitable for real-time instrument feedback loops. Across eight datasets ranging from STED nanoscopy to histopathology, the method achieves the lowest Learned Perceptual Image Patch Similarity (LPIPS) scores in 7 of 8 benchmarks while preserving structural coherence (e.g., MS-SSIM > 0.968 in Cryo-EM), which facilitates reliable downstream analysis. Supported by performance benchmarks in the AI4Life Denoising Challenge, NAFNet GAN restores structural features from low-photon-budget acquisitions, maintaining the temporal resolution required for dynamic live-cell workflows. ### Competing Interest Statement The authors have declared no competing interest. Ministerio de Ciencia, Innovación y Universidades (Agencia Estatal de Investigación), PID2023-152631OB-I00 Consejería de Educación, Ciencia y Universidades (Comunidad de Madrid), 2023-T1/SAL-GL-29109 European Regional Development Fund (ERDF / FEDER), MCIN/AEI/10.13039/501100011033/

Cells experience time-varying mechanical cues when navigating complex microenvironments, yet whether and how they retain a short-term memory of recent deformations remains unclear. Here, we show that glioblastoma cells encode such memory through transient cytoskeletal anisotropy. Combining magneto-mechanical actuation, nanoindentation, and selective cytoskeletal perturbations, we find that actin architectures drive opposite mechanical responses: stress fibers stiffen cells under stretch, whereas the actin cortex governs softening under compression. Vimentin intermediate filaments stabilize actin organization under load, preserving these deformation-specific responses. Mechanical actuation aligns both networks, more strongly for actin than vimentin, and this anisotropy persists after unloading. Using a two-step actuation protocol, we show that residual alignment biases the response to a second deformation: cells retain information about prior loading, and this bias decays as the cytoskeleton relaxes, defining a memory window of minutes to tens of minutes. To integrate these observations, we develop a multi-network constitutive model that links cytoskeletal architecture and loading history to cell mechanics, reproducing asymmetric mechanical responses, cytoskeletal reorganization dynamics, and memory effect. These findings show how invasive cancer cells could exploit residual cytoskeletal order to adapt to fluctuating solid stresses and confinement, and identify vimentin-actin coupling and remodeling kinetics as levers to limit that adaptability.

Image analysis in the life sciences is constrained by fragmented software ecosystems, heterogeneous data formats, and limited reproducibility. These barriers hinder the reuse of image analysis methods and the sustainability of tools. In this article, we describe how the Galaxy platform enables FAIR (Findable, Accessible, Interoperable, and Reusable) image analysis by providing an integrated environment for data access, workflow execution, provenance tracing, and training. We present Galaxy as a computational workbench that supports diverse image formats and integrates with public, institutional, and private repositories. We describe a reference structure for FAIR image analysis workflows and illustrate how this pattern supports reproducibility, interoperability, and reuse. We also describe community-driven training and sustainability practices that embed FAIR principles directly into executable tutorials and shared workflows. Together, these foundations position Galaxy as a reproducible, scalable, and community-maintained platform for FAIR image analysis across the life sciences and beyond.

Image-based machine learning tools are powerful resources for analyzing medical images, with deep learning-based semantic segmentation commonly utilized to enable the spatial quantification of structures visible in images. However, dataset generation and training of segmentation algorithms requires advanced programming skills and intricate workflows, limiting their accessibility to scientists without prior coding expertise. Here we present the step-by-step instructions to carry out automatic segmentation of medical images guided by a graphical user interface using the CODAvision algorithm. This workflow simplifies the process of semantic segmentation of microanatomical structures by enabling users to train highly customizable deep learning models without extensive coding expertise. The protocol outlines best practices for creating robust training datasets, configuring model parameters and optimizing performance across diverse biomedical image modalities. CODAvision enhances the usability of the CODA algorithm by streamlining parameter configuration, model training and performance evaluation, automatically generating quantitative results and comprehensive reports. We show the use of CODA to serial histology by demonstrating robust performance across numerous medical image modalities and diverse biological questions. We provide sample results in data types, including histology, magnetic resonance imaging and computed tomography. We demonstrate the diverse use of this tool in applications, including quantification of metastatic burden in in vivo models and deconvolution of spot-based spatial transcriptomics datasets. This protocol is designed for researchers with interest in rapid design of highly customizable semantic segmentation algorithms and a basic understanding of programming and anatomy.

Background: Radiomic analysis of preclinical tuberculosis imaging offers quantitative biomarkers of lesion evolution and treatment response; however, feature robustness is highly sensitive to lesion size, segmentation variability, and acquisition parameters. This study aimed to define the methodological limits of radiomics feature extraction in marmosets infected with Mycobacterium tuberculosis and to identify a reproducible subset of features associated with treatment response. Longitudinal PET/CT scans from infected marmosets, either untreated or treated with HRZE, were analysed. Lesions were independently segmented by multiple readers, and radiomic features were evaluated for inter-reader reproducibility using Lin’s concordance correlation coefficient and inter-machine variability using the Mann-Whitney U-test. Linear mixed-effects models were then applied to assess time-dependent differences between treated and untreated lesions. Results: Second-order and shape features were unreliable in lesions < 16 mm 3 , whereas larger lesions showed higher reproducibility. Filtering for inter-reader and inter-machine robustness excluded 58.1% of features. Linear mixed-effects models identified 176 features exhibiting significant time-dependent differences, predominantly PET-derived second-order features. Hierarchical clustering reduced this set to 37 representative features capturing distinct treatment-associated trajectories. The analysis revealed that lesion level variability contributed more to the temporal changes than subject level variability. The resulting feature set was further supported in independent cohorts of animals treated with alternative regimens or agents. Conclusions : This study defines a reproducible subset of predominantly PET-derived features that captures treatment-associated changes, supporting standardized radiomic pipelines for longitudinal assessment of therapeutic response in preclinical tuberculosis models.

Microscopy is fundamental to biological research, yet fluorescence images frequently suffer from acquisition noise, which degrades quality and complicates analysis. In this work, we propose NAFNetGAN, a generative adversarial network for supervised denoising of heterogeneous fluorescence microscopy data. Our approach employs a Nonlinear Activation Free Network (NAFNet) as the generator backbone within an adversarial perceptual framework to enhance both fidelity and perceptual realism. Evaluated on the AI4Life Microscopy Supervised Denoising Challenge 2025, NAFNet-GAN achieved top-ranked performance in three of four datasets, with PSNR-SI scores up to 38.16 dB, outperforming heavier architectures while remaining computationally efficient. These results demonstrate the potential of NAFNet-GAN as a practical, high-quality denoiser for real microscopy workflows.

Pancreatic cancer features a dense, immune-excluded stroma whose origins remain unclear. In this work, we extended CODA, a cellular-resolution three-dimensional (3D) histology pipeline, to map inflammation around more than 1,000 pancreatic precancers in large human pancreas specimens. Bulk analyses reproduce prior associations between overall inflammation and precancer burden, fibrosis, and acinar dropout, implicating ductal obstruction and stromal remodeling in early immune changes. Crucially, 3D mapping reveals that inflammation around individual precancers is highly heterogeneous, with immune hotspots and cold spots interchanging over tens of microns. Hotspots are found around regions of higher-grade dysplasia and ductal obstruction and are enriched for regulatory T cells and macrophages, indicating focal emergence of immunosuppression at the precancer stage. Integration with spatially resolved DNA sequencing implicated mutation in inflammation. These results position 3D mapping as a framework to identify rare sites of active microenvironmental priming and highlight focal immunosuppressive niches as candidate sites for early interception.

We present OREHAS (Optimized Recognition Evaluation of volumetric Hydrops in the Auditory System), the first fully automatic pipeline for volumetric quantification of endolymphatic hydrops (EH) from routine 3D-SPACE-MRC and 3D-REAL-IR MRI. The system integrates three components – slice classification, inner ear localization, and sequence-specific segmentation – into a single workflow that computes per-ear endolymphatic-to-vestibular volume ratios (ELR) directly from whole MRI volumes, eliminating the need for manual intervention. Trained with only 3 to 6 annotated slices per patient, OREHAS generalized effectively to full 3D volumes, achieving Dice scores of 0.90 for SPACE-MRC and 0.75 for REAL-IR. In an external validation cohort with complete manual annotations, OREHAS closely matched expert ground truth (VSI = 74.3 These results show that reliable and reproducible EH quantification can be achieved from standard MRI using limited supervision. By combining efficient deep-learning-based segmentation with a clinically aligned volumetric workflow, OREHAS reduces operator dependence, ensures methodological consistency. Besides, the results are compatible with established imaging protocols. The approach provides a robust foundation for large-scale studies and for recalibrating clinical diagnostic thresholds based on accurate volumetric measurements of the inner ear.

Neoantigens are mutated peptides arising from tumor genomic alterations, which can be recognized and attacked by the immune system, leading to antitumor immune responses. In the last decades, many immunotherapeutic strategies have been developed, which has increased the interest in neoantigens. This led to the development of computational tools that facilitate neoantigen identification and prioritization, prior to their validation using experimental approaches. This chapter aims at explaining the key steps that need to be conducted to identify potential neoantigens in silico, including an example of the most frequently used tools. This is followed by a description and comparison of the cutting-edge tools and pipelines for neoantigen prediction both for human and mouse. The last aim of this chapter is to depict the technical challenges that limit neoantigen prediction using bioinformatics, as well as the expected improvements, given the current revolution of artificial intelligence, which is implemented in most of the neoantigen-related tools. As exposed in this book chapter, we believe that advances in immunomics and computational biology will be key to implement personalized cancer immunotherapy in the clinical practice, to improve outcomes of cancer patients.

We introduce a deep learning pipeline for segmenting apical stem cells in three-dimensional confocal microscopy volumes of Arabidopsis thaliana. By integrating pre-trained 2D and 3D U-Net models from the BioImage Model Zoo, our method combines in-plane boundary sharpness with volumetric continuity. The workflow encompasses parallel preprocessing, dual-model inference, logical fusion, 3D reconstruction, and membrane-aware post-processing to extract key morphometric features, including cell counts and volumes. Evaluated over 22 timepoints, our pipeline achieves a mean Dice similarity coefficient (DSC) of 0.886 ± 0.002 . Passing–Bablok regression on cell counts yields a slope of 0.885 ( p = 0.573 ), surpassing the standalone 2D U-Net (0.539) and 3D U-Net (0.782). Volume estimates exhibit a slope of 0.754 ( p = 0.107 ), with smaller cell volumes but substantially fewer cell-merging artifacts compared to the baselines (0.738 and 0.930, respectively). These results highlight the advantages of model fusion for robust, biologically meaningful segmentation. The complete pipeline and source code are publicly available at: https://github.com/GolpedeRemo37/Arabidopsis_DualDL

Accurate assessment of PD-L1 expression is critical for guiding immunotherapy, yet current immunohistochemistry (IHC) based methods are resource-intensive. We present nnUNet-B: a Bayesian segmentation framework that infers PD-L1 expression directly from H E-stained histology images using Multimodal Posterior Sampling (MPS). Built upon nnUNet-v2, our method samples various model checkpoints during cyclic training to approximate the posterior, enabling accurate segmentation and epistemic uncertainty estimation via entropy and standard deviation. Evaluated on a dataset of lung squamous cell carcinoma, our approach achieves competitive performance against established baselines with a mean Dice Score and a mean IoU of 0.805 and 0.709, respectively, while providing pixel-wise uncertainty maps. Uncertainty estimates show a strong correlation with segmentation error, while calibration can be further optimized. These results suggest that uncertainty-aware H E-based PD-L1 prediction is a promising step toward scalable and interpretable biomarker assessment in clinical workflows.

Abstract Cells can experience time-varying mechanical cues, particularly when navigating through changing and complex microenvironments. Yet whether and how cells retain and use a short-term mechanical memory of recent deformations remains unclear. Here we show that, in glioblastoma cells, this memory is encoded by transient cytoskeletal anisotropy. Using uniaxial magneto-mechanical actuation aligned or perpendicular to the cell long axis, nanoindentation, and selective cytoskeletal perturbations, we find that distinct architectures of the actin cytoskeleton drive opposite mechanical responses: actin stress fibers mediate stiffening under stretch, whereas the actin cortex underlies softening under perpendicular loading. Vimentin intermediate filaments are essential to stabilize actin organization under load, preserving deformation-specific mechanics. Quantitative imaging reveals that mechanical actuation induces network-specific alignment and anisotropy, stronger for actin than vimentin, that persists transiently after unloading and bias subsequent responses, revealing a short-lived, deformation-dependent mechanical memory. To integrate these observations, we develop a multi-network constitutive model that links cytoskeletal architecture and loading history to cell-scale mechanics, reproducing both the asymmetric mechanical responses and the measured reorganization dynamics. These findings provide a structural basis for short-term mechanical memory and suggest how cancer cells could exploit residual anisotropy to adapt to fluctuating solid stresses and confinement, transiently biasing polarization, force transmission, and directional persistence during invasion. They also identify vimentin-actin coupling and the kinetics of cytoskeletal remodeling as potential levers to limit the mechanical adaptability of invasive cancer cells.

Microtubules are dynamic cytoskeletal filaments that have important organizing roles in all eukaryotic cells. Early in vitro studies on the effect of pH on microtubule stability focused on microtubules isolated from whole cell lysates, which can only replicate changes in intracellular pH. However, how extracellular pH can affect microtubule dynamics remains unclear. Here, we report that acidosis activates β1 integrin by increasing its affinity for RGD-containing ligands through the displacement of divalent ions in the metal-ion-binding sites of β1 integrin extracellular domain via protonation of Asp138. This induces the activation of RhoA and its downstream effector ROCK, which, via phosphorylation of Collapsin Response Mediator Protein-2 (CRMP-2), negatively regulates microtubules stability and changes the positioning and architecture of the Golgi apparatus. Thereby, extracellular pH modulates microtubule dynamics, which could have important consequences for intracellular organization, cell polarization, vesicular trafficking, nucleocytoplasmic shuttling, and cell division. There are multiple processes in human physiology associated with acidosis. The presented mechanochemical mechanism that links low extracellular pH and microtubule stability may serve as a blueprint for advancing our knowledge of cellular transport and exploring potential targets for drug development.

Introduction In preclinical Alzheimer’s disease (AD), oxidative stress induces non-enzymatic protein damage—detected as cerebrospinal fluid (CSF) biomarkers—and disrupts sleep-related networks, altering sleep electroencephalographic patterns. Due to the invasiveness of CSF sampling, quantitative electroencephalography (qEEG) is proposed as a non-invasive alternative for predicting oxidatively modified protein levels via Machine Learning (ML). Methods Forty-two mild-to-moderate AD patients underwent polysomnography (PSG). qEEG features were extracted. CSF protein oxidation markers levels —glutamic semialdehyde, aminoadipic semialdehyde, N ε -carboxyethyl-lysine, N ε- carboxymethyl-lysine, and N ε -malondialdehyde-lysine —were assessed by gas chromatography/mass spectrometry, and ML models trained to predict CSF biomarker levels. Results qEEG features from slow-wave sleep (SWS) and rapid eye movement (REM) sleep, particularly over frontal and central regions, yielded R2 > 0.9 and RMSE < 0.1 for biomarker prediction. Conclusion qEEG is a non-invasive, scalable tool for detecting AD-related oxidative stress, with potential implications for early diagnosis and risk stratification. ### Competing Interest Statement The authors have declared no competing interest.

Fluorescence microscopy faces challenges in resolution, phototoxicity, and anisotropic artifacts. The Fuse My Cells challenge, organized by France-BioImaging, aims to develop deep learning models that predict fused 3D volumes from single-view acquisitions, reducing phototoxic exposure while enhancing resolution. In this work, we benchmark state-of-the-art super-resolution models, including DFCAN, RCAN-3D, UNETR, and a 3D-adapted RCAN-it, evaluating their performance on the Fuse My Cells challenge dataset, which encompasses 802 3D light-sheet microscopy images. A novel training strategy prioritizing high-discrepancy regions optimizes efficiency and improves reconstruction accuracy. Our findings suggest that super-resolution models can not fully reconstruct the information on those image areas with minimum signal information. Code and documentation can be found at https://github.com/danifranco/BiaPy.

We present an accessible methodology for the absolute quantification and spatial mapping of hyaluronic acid (HA) in paraffin-embedded tissues. Although HA plays critical roles in tissue hydration, organization, and disease progression, its local concentration and spatial distribution remain poorly defined due to the lack of quantitative measurement tools. By integrating immunofluorescence, ELISA, and image analysis our approach enables the generation of pixel-level HA concentration maps. This strategy bridges the gap between qualitative imaging modalities and bulk biochemical assays, offering both spatial resolution and quantitative data. The method is validated across multiple tissues and is robust, scalable, and readily applicable to other extracellular matrix components, offering a practical tool for studying tissue microenvironments in health, disease, and biomaterial design.

Horizontal focusing, which precisely aligns cells within the focal plane of microscopic imaging, is essential for ensuring the analytical accuracy and reliability of imaging microflow cytometry. Hydrodynamic focusing, dependent on sheath flows, is constrained by parabolic velocity profiles that induce uneven shear forces, limiting lateral alignment precision. Active methods, such as acoustic focusing, enhance particle positioning but often face challenges in complexity and scalability. In contrast, viscoelastic focusing provides a simple and scalable alternative by leveraging the viscoelastic properties of non-Newtonian fluids. To optimize its application in microflow cytometry, we systematically investigate horizontal focusing performance depending on the effects of polymer molecular weight, concentration, and flow rate. Our findings reveal that stable single-plane focusing requires a fine balance between elastic and inertial forces, as higher molecular weight polymers generate stronger viscoelastic forces but may induce multiple equilibrium positions due to significant shear-thinning effects, disrupting single-plane alignment. These results highlight the importance of fine-tuning polymer properties and flow conditions to achieve precise and reliable horizontal focusing for microflow cytometry applications.

Artificial Intelligence methods are powerful tools for biological image analysis and processing. High-quality annotated images are key to training and developing new methods, but access to such data is often hindered by the lack of standards for sharing datasets. We brought together community experts in a workshop to develop guidelines to improve the reuse of bioimages and annotations for AI applications. These include standards on data formats, metadata, data presentation and sharing, and incentives to generate new datasets. We are positive that the MIFA (Metadata, Incentives, Formats, and Accessibility) recommendations will accelerate the development of AI tools for bioimage analysis by facilitating access to high quality training data.

Recent advances in imaging and computation have enabled analysis of large three-dimensional (3D) biological datasets, revealing spatial composition, morphology, cellular interactions and rare events. However, the accuracy of these analyses is limited by image quality, which can be compromised by missing data, tissue damage or low resolution due to mechanical, temporal or financial constraints. Here, we introduce InterpolAI, a method for interpolation of synthetic images between pairs of authentic images in a stack of images, by leveraging frame interpolation for large image motion, an optical flow-based artificial intelligence (AI) model. InterpolAI outperforms both linear interpolation and state-of-the-art optical flow-based method XVFI, preserving microanatomical features and cell counts, and image contrast, variance and luminance. InterpolAI repairs tissue damages and reduces stitching artifacts. We validated InterpolAI across multiple imaging modalities, species, staining techniques and pixel resolutions. This work demonstrates the potential of AI in improving the resolution, throughput and quality of image datasets to enable improved 3D imaging.

BiaPy, a unified open-source bioimage analysis library, offers a comprehensive suite of deep learning-powered workflows. Tailored for users of all levels, BiaPy features an intuitive interface, zero-code notebooks, and Docker integration. With support for 2D and 3D image data, it addresses existing gaps by providing multi-GPU capabilities, memory optimization, and compatibility with large datasets. As a collaborative and accessible solution, BiaPy aims to empower researchers by democratizing the use of sophisticated and efficient bioimage analysis workflows.### Competing Interest StatementThe authors have declared no competing interest.