npj Imaging

Quantitative imaging of extrachromosomal DNA (ecDNA) is increasingly important for studying cancer heterogeneity and adaptation, yet automated analysis has been limited by the absence of accessible imaging data, gold standard annotations and adaptable computational tools. Here we establish an open resource for computational ecDNA imaging, integrating 2,986 native-resolution metaphase FISH image sets with manual annotations, standardized benchmarks and open-source quantification frameworks. We use this resource to systematically compare existing and newly developed approaches spanning rule-based computer vision, deep-learning-based segmentation and probabilistic localization. This comparison reveals count-dependent underestimation that distorts ecDNA copy-number distributions and motivates ecCount, a probabilistic localization method developed here to preserve individual ecDNA signals and quantitative burden. ecCount achieves an object-level F1 score of 0.939 on held-out images with minimal count bias. Together, the images, annotations, retrainable models, evaluation tools and guided workflows provide community infrastructure for applying, adapting and improving automated ecDNA quantification across experimental systems.

Fluorescence microscopy is a central tool in biomedical image analysis, but its practical use is often constrained by incomplete and heterogeneous data. This is especially true in rare disease research, where patient cohorts are small, biological material is limited, and complex study designs frequently yield images with missing, degraded, or low-quality channels. We introduce Reliability-Mask-Conditioned DDPM (RM-DDPM), a conditional diffusion model that formulates multichannel image generation, missing-channel imputation, and degraded-channel restoration as a single reliability-guided inference problem. The model is trained on a curated subset of complete, high-quality multichannel images using a self-supervised masking objective to learn rich cross-channel dependencies from the available channels. At inference time, reliable observations condition the reverse diffusion process, allowing the same trained model to solve multiple reconstruction tasks without task-specific retraining. We evaluate RM-DDPM against task-specific baselines on one public and two in-house fluorescence microscopy datasets, comprising a rare-disease preclinical model study and a clinical patient cohort. Across these settings, RM-DDPM matches the specialized baselines in reconstruction quality, hallucinates fewer false structures, and remains more robust as degradation severity increases, while repeated sampling provides per-pixel uncertainty maps that flag unreliable reconstructions.

Scientific imaging is undergoing a fundamental transformation. Advances in multiplex immunofluorescence, computational pathology, and electron microscopy now enable observation of biological and physical systems at unprecedented spatial resolution. Rather than simply classifying known objects, these data offer the opportunity to discover previously unknown organizational principles. However, scientific image datasets are typically highly curated, information-rich, and limited in size, making reliable representation learning particularly challenging. We show that standard continuous self-supervised adaptation can become unstable in this regime, allowing teacher representations to drift and degrading downstream transfer. We introduce Kodiak, a discrete representation learning architecture that discovers latent scientific concepts through Sinkhorn-balanced codebook assignments. Instead of learning unconstrained continuous embeddings, Kodiak constructs a reusable vocabulary of discrete concepts that provides stable supervision for both masked-patch prediction and cross-view representation learning within a single-stage framework. Across five scientific imaging domains, including multiplex immunofluorescence for pancreatic cancer pathology and scanning electron microscopy, Kodiak consistently improves representation quality under both continued foundation-model adaptation and training from scratch. Beyond improved classification and transfer,the learned concepts are interpretable, reusable, and produce spatially coherent concept maps that support applications such as cell phenotyping, segmentation-free patch phenotyping, spatial niche discovery, tissue organization, and materials characterization. We view Kodiak as the representation-learning engine of our foundation model for spatial biology. More broadly, our results suggest that discrete representation learning provides a principled foundation for discovering reusable latent scientific concepts that can accelerate scientific interpretation across imaging domains.

Nuclear segmentation in 3D microscopy remains challenging, particularly for low signal-to-noise ratio (SNR) data from timelapse imaging, in vivo experiments, or dense tissue samples. Yet accurate 3D nuclear segmentation is a prerequisite for quantitative single-cell analysis, making this a key bottleneck for high-throughput study of challenging datasets. We present NucLogic, a fast and robust 3D nuclear segmentation algorithm that combines the maturity of 2D segmentation with geometric stitching rules based on nuclear shape, orientation, and intensity, to reconstruct accurate 3D volumes. NucLogic reliably segments challenging datasets including long-term timelapse imaging of organoids, and large dense structures such as the zebrafish brain. Benchmarked against other state-of-the-art segmentation tools, NucLogic matches F1 scores at high SNR, with increasing outperformance as SNR decreases. Importantly, segmentation completes at 20-50% of the computation time, with greater speed gains on larger images. Beyond segmentation, NucLogic includes an intuitive graphical user interface allowing non-experts to segment, quantify, and visualize 3D and 4D data. This enables easy quantification of cell numbers and integration of a variety of both cell-intrinsic properties, such as phenotype, morphology, and local environment metrics, including density and spatial interactions. Together, NucLogic makes quantitative 3D nuclear analysis of previously unmanageable datasets accessible to a broad research community.

Coarse to fine segmentation is a natural strategy for three dimensional lesion delineation becausefull resolution processing of an entire scan is expensive while only a small fraction of the volume contains pathology. Its main failure mode, however, is structural: a lesion missed by the coarse stage is usually absent from the region of interest and therefore cannot be recovered by the fine stage. This problem is most severe for small, low contrast, and multifocal lesions. We introduce RECALLREFINE, a coverage constrained coarse to fine framework that treats candidate-region selection as a recall critical learning problem rather than a deterministic crop around the coarse mask.A full-volume network first predicts coarse lesion probabilities and multi-scale features. A candidate score network then combines coarse probability, predictive uncertainty, feature novelty, and spatial context to score overlapping 3D blocks. During training, a differentiable selector is optimized with a component-wise coverage objective that requires every annotated lesion instance to remain covered by at least one selected block under a fixed refinement budget. Selected blocks are processed at native resolution by a local refiner with a context-ring branch, and overlapping predictions are merged back into the full volume. Across four public 3D lesion benchmarks, RECALLREFINE improves mean tumor Dice from 76.5% for the strongest matched coarse-to-fine baseline to 78.4%, while increasing sensitivity for lesions below 10 mm from 60.9% to 70.8%.At a 20% candidate budget, the selector covers 94.4% of lesion instances, substantially reducing irreversible coarse-stage misses. The results suggest that the central problem in cascaded lesion segmentation is not only how to refine a detected lesion, but how to guarantee that difficult lesions are still represented in the refinement set.

Drosophila melanogaster is a powerful model for studying systemic metabolism. Although metabolism is compartmentalized across tissues, prior work in adult Drosophila has mostly measured it in dissected tissues rather than in situ, limiting the spatial information that can be gained. Here, we combine whole-fly mass-spectrometry imaging with stable-isotope tracing and flux modeling to quantify metabolic activities across the adult fruit fly in a spatially resolved manner. Across 68 tracer-fed Oregon-R female flies and 187,709 pixels, we map 81 metabolites onto eight annotated regions, including muscle, fat body, gut, and nervous system. Glucose labeling and fitted relative fluxes indicate higher glycolytic and tricarboxylic acid cycle turnover in thoracic muscle than in abdominal fat body; proline labels the nervous system and eye; glutamine the abdominal reproductive system and gut. Alanine carries glucose-derived carbon through the hemolymph and transfers nitrogen into glutamate. In flies bearing yorkie-driven gut tumors, alanine contribution to malate rises in gut and abdomen, and reducing gut alanine uptake lowers tumor burden. Together, this work establishes a framework for quantitative spatial metabolic analysis, showing compartmentalized nutrient use across the adult fruit fly.

Counterfactual medical images are useful only when a requested clinical change is introduced without silently altering patient characteristics that should remain fixed. Existing diffusion editors can produce convincing images, but realism or target-classifier flips do not establish that an edit is localized, diagnosis-specific, anatomically faithful, or useful on real clinical data. We introduce PATHEDIT, a state-factorized counterfactual editing framework that decomposes an abdominal CT representation into pathology, anatomy, acquisition, and residual context before performing a clinically specified latent intervention. A pathology editor modifies only the target factor, while non-target factors are explicitly copied and preservation losses penalize collateral drift. The edited state is decoded jointly into an image, a report level description, and diagnostic readouts. Weak supervision from lesion masks, organ segmentations, and naturally occurring longitudinal lesion changes anchors edit direction and spatial support without requiring pixel-aligned counterfactual ground truth. We evaluate four properties that are often conflated in medical image editing: target edit success, non-target preservation, anatomical identity, and real-data utility. In our evaluation protocol, PATHEDIT achieves 94.2% target success while reducing mean off-target diagnostic change to 3.0%, yields substantially stronger alignment with observed lesion changes, and improves rare-lesion recognition and shortcut robustness on held-out real CT. These results motivate astricter view of counterfactual medical imaging: a useful edit must prove not only what changed, but also what did not.

The Transport-of-Intensity Equation (TIE) and polarization differential phase contrast (pDPC) exhibit transfer characteristics that limit their use as standalone quantitative phase imaging modalities. TIE provides stable recovery of low spatial frequencies but attenuates fine structure, while pDPC enhances phase gradients but is intrinsically insensitive near zero frequency. We present a hybrid reconstruction framework that combines a single-shot pDPC measurement with an in-focus-defocused TIE image pair and fuses the resulting phase estimates in the Fourier domain using frequency-selective weighting. The illumination geometry is co-designed to enforce angular separation between modalities, enabling high-frequency phase-gradient acquisition in a single exposure while preserving low-frequency sensitivity through low-numerical-aperture symmetric illumination. The proposed method requires three intensity measurements without requiring interferometric stability or sequential asymmetric illumination. Experimental results demonstrate improved phase fidelity across spatial frequencies, reduced halo artifacts, and enhanced recovery of both smooth and fine structural features. This approach provides a compact and robust method that moves towards extended spatial-frequency phase imaging for dynamic and resource-constrained applications.

Three-dimensional medical image segmentation is unusually heterogeneous in its computational needs: large homogeneous organ interiors are often resolved at coarse scale, whereas small organs, lesions, thin vessels, and ambiguous boundaries benefit from native-resolution processing. Yet modern 3D segmentation networks allocate nearly uniform computation across the entire volume. Existing dynamic-resolution or coarse-to fine methods reduce cost, but typically route regions using hand-crafted image complexity, prediction confidence, or a fixed crop, none of which directly estimates whether spending additional computation on a region will improve the segmentation. We introduce ERRORROUTE, a budgeted inference framework that learns the marginal refinement utility of candidate 3D blocks. A coarse branch first predicts global low-resolution semantics. During training, we estimate for each candidate block the loss reduction achievable by native resolution refinement and supervise a utility estimator to predict this gain before refinement. A differentiable budgeted selector then allocates a fixed compute budget to blocks with the highest expected utility per unit cost, while unselected regions remain on the efficient global path. Selected blocks are refined locally and merged through residual cross-scale fusion. Across six 3D CT/MRI benchmarks, ERRORROUTE achieves a mean Dice of 90.5% at 128 GFLOPs, improving the accuracy-compute frontier over dense 3D baselines and fixed coarse-to-fine systems. At a 30% routing budget, the learned utility estimator recovers 83% of the blocks selected by an oracle router, with the largest allocation concentrated on lesions, small structures, and complex boundaries. These results suggest that efficient medical segmentation benefits from learning where additional computation is useful, rather than merely where the model is uncertain.

Vasculogenic mimicry (VM) describes the ability for tumor cells to mimic endothelial cells and form tumor-lined blood vessels. Although VM is associated with poor prognosis, resistance to therapeutics and metastasis, its presence, relevance and functionality have been debated, largely due to the techniques employed to determine its presence. Dual CD31 and periodic acid-Schiff (PAS) histological staining, the method widely used to detect VM within tumor tissue, is hampered by issues with artifacts of 2D sections and does not assess active blood flow through vessels and hence cannot confirm vessel functionality. Here, we present a 3D imaging pipeline to detect functionally perfused VM and host vessels within mouse tumor models. Mice tumor models are intravenously injected with fluorescently-labelled tomato lectin which binds to the lumen of functional blood vessels. These tumors are thickly sliced, optically-cleared, stained with endothelial markers and imaged using light-sheet microscopy. We apply this imaging modality to image tumors from six breast cancer patient-derived tissue xenograft models, an orthotopic mammary cancer model and two small cell lung cancer circulating-tumor cell explant models, where VM vessels, mosaic vessels and host-VM vessel junctions can be detected. To our knowledge, this is the largest collection of 3D VM vessel images to date, providing convincing evidence of the presence and functionality of VM within tumors and defining the level of evidence necessary to detect VM in vivo.

Image scanning microscopy (ISM) is emerging as the natural evolution of confocal laser-scanning microscopy, combining optical sectioning and a two-fold improvement in lateral resolution with high signal-to-noise ratio. Single-photon avalanche diode (SPAD) array detectors enable an effective implementation of ISM by recording a micro-image at each scan position with single-photon sensitivity, negligible readout noise, and without the fixed frame-readout constraints of conventional cameras. Nevertheless, SPAD array-based ISM remains largely confined to laboratories with expertise in microscope development. Here, we present BrightEyes-Scope, a reference implementation of a SPAD array-based ISM system. We provide step-by-step guidelines for optical design, system construction, alignment, calibration, data acquisition, and image reconstruction. Presented as a complete ISM microscope, BrightEyes-Scope also serves as a reference for upgrading existing laser-scanning microscopes. We anticipate that it will accelerate the adoption of ISM for quantitative super-resolution imaging.

In this paper, we introduce Nagini3D, a 3D biological image segmentation method that hybridizes two high-performing yet highly complementary segmentation paradigms: deep learning-based approaches and snake-based methods. Our approach leverages both the high efficiency of data-driven techniques and the convenient shape representation provided by active surfaces, which proves highly practical for incorporating a priori knowledge as well as for the post-analysis of segmented objects. By considering a complete 3D volume, our method ensures segmentation consistency across all three dimensions, unlike many approaches that handle volumes in a slice-by-slice manner. Additionally, it theoretically enables the representation of any object with spherical topology, overcoming the limitations of common 3D methods that rely on object convexity assumptions. Furthermore, the continuous representation of objects allows for a resolution-independent description, facilitating the extraction of various geometric features of interest (e.g., local curvature, principal deformations). We demonstrate both the quality of the segmentation produced by our method and the richness of the accessible geometric features through an extensive set of experiments conducted on multiple real and simulated datasets, including objects with challenging geometries.

Capturing fast hemodynamics in deep organs is essential for understanding microvascular regulation of organ responses to physiological demands and pathological stress. However, non-invasive three-dimensional (3D) imaging of these microscale processes remains challenging due to trade-offs between spatial resolution, imaging speed, and penetration depth. Here, we present fluctuation-based high-order super-resolution acoustic microscope (FLAME), a tracking-free 3D ultrasound imaging technology capable of fast microvascular angiography and flow measurement. Using as few as 30 volumes, FLAME improves 3D ultrasound resolution by 8-fold (~50 m) and shortens data acquisition time by over two orders of magnitude, from tens of seconds to tens of milliseconds, compared with conventional tracking-based super-resolution approaches1-6. This high-throughput capability supports high-fidelity imaging of tissue functions under flexible experimental settings. FLAME achieves an unprecedented 3D super-resolution frame rate of ~40 Hz, capturing transient hemodynamic responses to various physiological and pathological stimuli in different mouse organs, from the microvascular level to the whole-body scale. With open-sourced implementation, FLAME provides an accessible platform for real-time, in-depth hemodynamic monitoring across diverse biomedical applications and clinical translation.

Focal adhesions (FAs) and actin stress fibers (SFs) are mechanically coupled but often analyzed separately. We present FilAdh, a unified framework for FA instance segmentation, SF instance reconstruction, morphometry, and FA-anchorage-based SF classification. Across 57 held-out HeLa, SUIT-2, and MRC-5 cell line images, FilAdh achieved the highest FA Dice, intersection over union (IoU), and recall and the highest SF clDice and skeleton precision among benchmarked methods; paired t tests with Holm correction indicated significant gains over all compared methods for FA Dice, IoU, and recall, and over classical methods for SF centerline metrics, with no significant differences detected relative to the deep-learning comparators. Conditional on expert-curated FA and SF masks, SFs were assigned to dorsal, ventral, and peripheral classes with 81.5%-87.3% accuracy. Controlled Poisson-Gaussian noise, Gaussian defocus, and point-spread-function blur revealed distinct, structure-specific errors in overlap and object-level measurements. By linking segmentation and reconstruction to count, area, length, apparent width, and structural class, FilAdh offers an interpretable view of the FA-actin structural system and makes degradation-dependent measurement biases explicit. BIGGER PICTURECells sense and respond to their physical environment through focal adhesions and actin stress fibers. The two form a mechanically coupled system, yet they are usually analyzed separately, with workflows re-tuned for each new set of images. FilAdh analyzes both within a single framework, tailoring segmentation, reconstruction, and classification to each structures geometry: compact adhesion patches versus thin, extended fiber networks. Testing on expert-annotated references and under controlled degradations that mimic low signal, focus errors, and optical blur shows that these common imaging imperfections distort adhesion and fiber measurements in different ways. The same design principle--matching the analysis to each structures shape and judging methods by how errors propagate into the final biological measurements--may benefit other bioimage-analysis problems. Released code, trained weights, and fixed data splits make all comparisons reproducible and extensible. HIGHLIGHTSO_LIFilAdh jointly quantifies focal adhesions and actin stress fibers C_LIO_LIFilAdh achieves the highest benchmarked FA overlap and comparable SF topology C_LIO_LIMatched synthetic degradations reveal geometry-specific measurement errors C_LIO_LIFA-SF spatial context enables dorsal, ventral, and peripheral SF classification C_LI

Functional ultrasound (fUS) imaging provides deep, wide-field access to brain function, but instrument cost and complex acquisition and analysis workflows restrict its wider adoption. Here we introduce @fUS, an agentic transcranial imaging platform with an open architecture that integrates purpose-built hardware with executable experimental skills and persistent structured memory. The system combines 128-channel, 16-bit acquisition at 125 MHz with GPU-based processing and enables functional imaging through the intact scalp and skull of mice, with approximately 100-m spatial resolution and 10-Hz temporal sampling, providing a basis for longitudinal studies without cranial-window surgery. Its agent connects scientific objectives to experimental design, direct instrument control and data analysis, retaining context across interactions and supporting inspectable workflows through text and hands-free voice control. Neuroscience trainees with limited engineering experience independently configured the system within 30 min. Agent-assisted exploration of whisker-stimulation datasets revealed low-frequency vascular changes beyond conventional response mapping, supported by complementary two-photon measurements of single-vessel diameter. By combining transcranial imaging with accessible instrument control and analysis, @fUS provides a foundation for wider adoption and larger, more diverse neuroscience datasets. More broadly, it offers a framework for scientific instruments in which measurement, computation and experimental reasoning are developed as parts of the same system.

Small numbers of cells can determine the fate of an entire organism. For example, metastatic cancer cells seed distant organs, therapy-resistant clones repopulate tumors, and T cells infiltrate and expand in tissues with matching antigens. Visualizing the activity of such cells in living organisms requires highly sensitive, specific and repeatable imaging in optically opaque tissues. Here we introduce cell chirp, an ultrasound imaging method for real-time imaging of single cells in living organs inaccessible to light. Cell chirp exploits the harmonic response of genetically encoded gas vesicles to frequency-coded ultrasound to sensitively and specifically visualize single cells in scattering background. We first develop this method in tissue-mimicking phantoms, demonstrating the ability to quantitatively detect single cells as validated by optical cell counts. We then establish the ability of cell chirp to image thousands of genetically labeled cells within a million-cell background tissue, track reporter gene expression over days, and suppress background signals by more than two orders of magnitude. Finally, we use cell chirp to track the fate of antigen-positive or antigen-negative tumor-cell minorities during weeks-long T-cell engager treatments. Cell chirp reveals the dynamics of sparsely distributed, genetically defined cells in opaque tissues.

The clonogenic assay is the foundational standard for evaluating cellular radiosensitivity, yet its manual quantification is limited by labor-intensive processing, high inter-observer variability, and a reductive binary scoring system. Here, we present and validate a deep learning-based segmentation pipeline for the automated, high-throughput quantification of mammalian cell survival across diverse radiation qualities (photons, protons, and carbon ions). By benchmarking against expert consensus using Bland-Altman analysis, we demonstrate that this AI framework consistently operates within natural human error margins. Crucially, transitioning to automated whole-well image-based analysis enables the extraction of continuous multidimensional data, including colony area, density, and cellularity. This phenotypic profiling successfully identified and quantified high-cellularity colonies formed by radioresistant BxPC-3 cells under particle irradiation. Coupled with rapid image acquisition platforms, this methodology significantly reduces processing bottlenecks while standardizing the derivation of relative biological effectiveness (RBE). By advancing beyond binary survival metrics to continuous morphological profiling, this framework provides a vital new analytical dimension for investigating structural mechanisms of radioresistance.

Antibiotic resistance is an increasingly serious problem worldwide. Identifying the bacterial species can help guide antibiotic selection and improve patient outcomes. In sepsis, where every minute counts, rapid identification is particularly important, and new approaches are needed. In this study, deep learning models were trained to identify bacterial species from phase-contrast time-lapse images of cells growing in microfluidic traps, without fluorescent labels. We demonstrated this approach in two previous studies using laboratory strains. Here, the previous work is extended to clinical bacterial isolates, evaluating performance on isolates not seen during training. Furthermore, it was tested whether autofluorescence could aid species identification. Two validation sets were acquired in separate experiments from the training data, the first containing clinical-isolate traps with five species (Escherichia coli, Enterococcus faecalis, Klebsiella pneumoniae, Staphylococcus aureus and Enterobacter cloacae), imaged every two minutes, and the second containing clinical isolates of two species (E. coli and K. pneumoniae) recovered from spiked donor blood, imaged every minute. Training with the best previous CNN method achieved unsatisfactory results. Therefore, vision transformers were explored, and a self-supervised adaptation procedure was developed for pretrained DINOv3 using a frozen teacher, followed by supervised classification using time-lapse montages. Mean macro-F1 was 93.6 {+/-} 1.3% for isolates precultured in Mueller-Hinton medium using 60 frames (118 minutes) and 87.2 {+/-} 2.7% for isolates precultured in blood using 55 frames (54 minutes) (mean {+/-} sample standard deviation across training seeds). Performance was evaluated over time by increasing the number of frames available to the model during testing, and Fourier-domain filtering approximated reduced numerical aperture. The results are promising and suggest that the aperture can be decreased while retaining good results. However, further evaluation on independent isolates, experiments and patient samples is needed to establish generalization. Together with previous findings, these results bring us closer to rapid image-based diagnostics that could transform the initial treatment of acute bacterial infections by enabling better-informed antibiotic selection.

Machine learning benchmarks often treat image files as independent observations, although multiple files may represent repeated views or other related samples. Prior work has shown that non-independent partitioning can inflate model performance, but the sensitivity of a biomedical benchmark to progressively increasing train-test relatedness is less often quantified. Here, we quantified how apparent classification performance changes when recoverable sample groups are progressively allowed to cross training, validation, and test partitions. Using source structure and filenames, we reconstructed analysis groups in a microscopy corpus compiled from multiple sources and generated five class stratified partitioning conditions, ranging from an intact group reference to random assignment of individual files. We applied the same training recipe within each model family to 129,173 images spanning 14 classes of blood cells and hematopoietic cell states. For a pretrained Vision Transformer, random assignment of individual files increased the mean test macro F1 from 0.853 to 0.977 and accuracy from 0.851 to 0.990 relative to the intact group reference. Apparent performance increased progressively across the intermediate conditions. The endpoint pattern also appeared with a ResNet-50 backbone, and the largest apparent gains occurred in immature granulocytic and blast classes. A descriptive analysis within the random assignment condition further showed higher scores among test images that had another member of the same group in the training set, although the exposed and unexposed subsets were not exchangeable. The grouping workflow, exact split definitions, code, and compact result files are released as HSD-Bench to support reproducible evaluation of hidden sample dependence.

Microscopy-based cell profiling has broad applications in biological discovery, disease characterization, and phenotypic drug screening. Modern microscopy continues to push the limits of resolution, speed, depth and throughput, but better imaging does not automatically lead to better biomedical discovery and translation. A key bottleneck is feature representation: existing features are either handcrafted or learned as black-box embeddings and often lack explicit biological meaning. Here we propose biological grounding as a first principle for cell profiling and implement it through MorphAgent, an AI agent framework in which each quantitative feature originates from a biological hypothesis and is anchored to a cellular structure or process. By integrating biological knowledge, multimodal reasoning and automated validation, MorphAgent makes biologically grounded feature design systematic and scalable. We show biological grounding fundamentally changes the properties of cell profiling features. Across three microscopy modalities, MorphAgent produces compact, expressive and transferable features for drug screening, cell state characterization and disease morphology analysis. In wide-field Cell Painting, a common assay for phenotypic drug screening, it improves perturbation-retrieval mean average precision by 48% over CellProfiler and 20% over DeepProfiler while using substantially fewer dimensions. In confocal mitochondrial imaging, biologically grounded features transfer across independently acquired datasets and imaging resolutions, supporting accurate aging-state classification. In structured illumination microscopy (SIM) imaging of Tau-labeled samples, biologically grounded super resolution features reveal nanoscale morphologies inaccessible at conventional resolution, enhancing the discrimination of disease-associated mutations. Moreover, biologically grounded features provide a hierarchical organization of cellular morphology, enabling multilevel alignment with transcriptomics data and facilitating mechanistic understanding. Biologically grounded cell profiling thus brings AI reasoning closer to biological mechanisms, advancing biomedical discovery and translation.