Abstract. The pulmonary capillary network (PCN) is a highly complex and dynamic structure essential for gas exchange and systemic homeostasis. Beyond its primary role in oxygen transport, the PCN also mediates a range of transport processes relevant to both health and disease. Scientific interest in the PCN has evolved considerably over time, from early efforts characterizing its anatomy and architecture to modern investigations of its in situ microcirculatory dynamics using advanced imaging technologies. More recently, research has shifted toward developing biomimetic models - such as in vitro microfluidic systems and tissue-engineered constructs—that replicate the PCN's intricate structure and cellular function. These platforms now play a crucial role in disease modeling, drug delivery research, and the study of pulmonary microvascular biology under both physiological and pathological conditions. In this review, we first outline the anatomical and physiological characteristics of the PCN and discuss its roles in homeostasis and disease. We then examine classical biomechanical models of blood flow in the PCN, followed by an overview of recent advances in in vitro modeling approaches, with an emphasis on microfluidic platforms. Finally, we highlight emerging next-generation models designed to better replicate PCN complexity and discuss how they can accelerate the development of new therapeutic strategies.

Extracting structured clinical information from radiology reports is challenging, especially in low-resource languages. This is pronounced in Crohn's disease, with sparsely represented multi-organ findings. We developed Hierarchical Structured Matching Prediction BERT (HSMP-BERT), a prompt-based model for extraction from Hebrew radiology text. In an administrative database study, we analyzed 9,683 reports from Crohn's patients imaged 2010-2023 across Israeli providers. A subset of 512 reports was radiologist-annotated for findings across six gastrointestinal organs and 15 pathologies, yielding 90 structured labels per subject. Multilabel-stratified split (66% train+validation; 33% test), preserving label prevalence. Performance was evaluated with accuracy, F1, Cohen's $\kappa$, AUC, PPV, NPV, and recall. On 24 organ-finding combinations with $>$15 positives, HSMP-BERT achieved mean F1 0.83$\pm$0.08 and $\kappa$ 0.65$\pm$0.17, outperforming the SMP zero-shot baseline (F1 0.49$\pm$0.07, $\kappa$ 0.06$\pm$0.07) and standard fine-tuning (F1 0.30$\pm$0.27, $\kappa$ 0.27$\pm$0.34; paired t-test $p < 10^{-7}$). Hierarchical inference cuts runtime 5.1$\times$ vs. traditional inference. Applied to all reports, it revealed associations among ileal wall thickening, stenosis, and pre-stenotic dilatation, plus age- and sex-specific trends in inflammatory findings. HSMP-BERT offers a scalable solution for structured extraction in radiology, enabling population-level analysis of Crohn's disease and demonstrating AI's potential in low-resource settings.

DNA repair is critical for cellular function and genomic stability across organisms. Yeast mating-type switching serves as an established model for studying DNA break repair and chromosome dynamics. However, real-time tracking of mating-type switching in live cells remains challenging due to resolution limitations of existing techniques. Here, we use high-throughput methods, including three-dimensional imaging, to follow the dynamics of DNA damage and repair and to quantify mating-type switching occurrences at the single live cell level, with unprecedented resolution. We reveal chromosome reconfiguration for both single- and double-strand breaks following switching induction. Our findings provide new observation of the correlation between chromosome folding and single-strand breaks.

Obstructive pulmonary diseases, including asthma and chronic obstructive pulmonary disease (COPD), are widespread and represent a major global health burden. Despite their impact, effective therapeutic delivery to the small airways using inhaled aerosols remains suboptimal. In this study, we present a novel in vitro airway-on-chip platform that mimics both normal and constricted small bronchial geometries to quantify the deposition charged and neutral polystyrene latex (PSL) aerosol particles ranging from 0.2 to 2 µm. Analytical and numerical solutions were derived from dimensionless scaling laws to further support the experiments and predict deposition location. Our experiments showcase how electrostatic forces significantly alter deposition patterns across particle sizes in these small airways. For submicron particles, we observe the enhancement of proximal airway deposition due to the coupling of electrostatic-diffusive screening effects. For larger particles, which typically deposit only in the direction of gravity, the inclusion of electrostatic forces significantly extends their deposition footprint, enabling deposition even in orientations where gravitational sedimentation is not feasible. Constricted regions consistently exhibit lower deposition across all cases, the presence of electrostatic forces enhanced overall deposition, offering a potential strategy for targeting bronchioles. Together, these findings suggest that electrostatic attraction may be strategically leveraged to enhance aerosol targeting in the small airways, providing new opportunities for optimizing inhaled drug delivery in obstructive lung diseases.

The influence of mechanical signals on sprouting angiogenesis has been of interest in the field of tissue engineering and biomechanics. Here, a unique experimental methodology is developed to apply mechanical loading on an engineered macro-vessel model to study the influence on angiogenic sprouting. The polydimethylsiloxane (PDMS) stretchable device contains an engineered macro-vessel embedded within a collagen matrix. The model is loaded either parallel or perpendicular to the macro-vessel (longitudinal or lateral, respectively). A finite element analysis is performed to characterize the strain maps of the PDMS-collagen setup. The results indicate high uniform strain around the macro-vessel perimeter under longitudinal loading, while lateral loading results in low strain in the horizontal direction and high strain along the vertical direction. Experimental results for lateral loading show increased sprouting events and capillary orientation in the stretch direction following the organization of matrix fibers, while longitudinal loading results in sprouting inhibition. These findings allow prediction of angiogenic sprouting under specified mechanical loading profiles and may serve as a tool to rationally design and control vascular network architecture by physical means. Finally, the presented approach can serve as a platform for studying cell behavior under mechanical loading for any physiological tubular duct or vessel model.

Artificial intelligence (AI) is being integrated into nearly every aspect of modern life, and machine learning (ML)–a subfield of AI–has the potential to accelerate the development of nanomedicines. Here, a machine learning workflow is presented to optimize the microfluidic-based formulation of nanomedicines. A database of ≈200 unique nanomedicine formulations with over 550 total measurements is curated by producing liposomes, lipid nanoparticles, and poly(lactic-co-glycolic acid) nanoparticles, either empty or loaded with the model therapeutic agent, curcumin. Nanoparticle production parameters are systematically varied, and the resulting particles are characterized for their diameter, polydispersity index, and encapsulation efficiency. These data are used to train and validate 13 different ML models using open-source libraries, with the task of returning the most accurate prediction of nanomedicine attributes. The most accurate models, based on random forest regression, are implemented to yield particles with user-specified attributes. Finally, the proposed ML workflow, MicrofluidicML, is compared against generative large language models–OpenAI ChatGPT, Google's Gemini, and DeepSeek. MicrofluidicML provides a workflow where the researcher has complete governance and control of the input data with a relatively low computational overhead, and represents a step toward implementing a computationally lightweight ML framework to accelerate nanomedicine development.

In patients with ischaemic stroke, retrograde perfusion of the penumbra by leptomeningeal collateral vessels (LMCs) strongly predicts clinical outcome, suggesting that enhancing LMC flow can offer a novel therapeutic approach. Using in vivo measurements and computational modelling it is shown that LMCs experience elevated fluid shear stress that is significantly higher than that in other blood vessels during ischaemic stroke in rats and humans. We exploit this to selectively enhance flow in LMCs using shear-activated nanoparticle aggregates carrying the vasodilator nitroglycerin (NG-NPAs) that specifically release drug in regions of vessels with high wall shear stress (≥100 dyne cm⁻²). NG-NPAs significantly increased LMC-mediated penumbral perfusion, decreased infarct volume, and reduced neurological deficit without altering systemic blood pressure in a rat ischaemic stroke model. NG-NPAs also avoided common side effects of systemic nitrate administration, such as systemic hypotension, cerebral vascular steal, cortical vein dilation, or intracranial pressure elevation. Systemic administration of free NG at the maximal tolerated dose, which is ten times higher than the dose of NG used in the NG-NPAs, do not enhance LMC perfusion and dropped blood pressure. Thus, packaging NG within shear-activated NPAs can potentially enable this widely available vasodilator to become a highly effective therapeutic for ischaemic stroke.

PolyHIPEs are porous polymer monoliths synthesized within high internal phase emulsions (HIPEs) with >74% dispersed phase. Biodegradable polyester-based poly(urethane urea) (PUU) monoliths have been synthesized within water-in-oil emulsions via simultaneous diisocyanate–polyol urethane reactions (predominant) and diisocyanate–water urea reactions (CO₂ byproduct) and then used for cell culture and soft tissue engineering. Here, in contrast, an innovative sequential reaction route was used. The polycaprolactone and polylactide polyols were end-capped with diisocyanates, generating urethane groups, prior to emulsion formation. The novel monolith-forming reaction used here was the isocyanate–water urea reaction. Different diisocyanates, from relatively flexible to relatively stiff, highlighted the effects of diisocyanate stiffness on macromolecular stiffness, crystallinity and modulus. Similarly, different polyesters highlighted the effects of polyester stiffness. In addition, the novel isocyanate–water monolith formation reaction was investigated using a lysine-derived diisocyanate. The hierarchical porosity of the resulting CO₂-foamed polyHIPEs contained millimeter-scale bubbles and micrometer-scale emulsion-templated porous structures. The crystalline polycaprolactone-based PUUs exhibited melting points between 50 and 80 °C, reflecting the unique macromolecular uniformity generated by the sequential urethane–urea reactions compared to the simultaneous urethane–urea reactions that produced amorphous PUUs. Cell growth within the resulting elastomeric polyHIPEs demonstrated potential for soft tissue engineering, with mouse skeletal muscle cells adhering, spreading, proliferating, organizing and filling the entire space. © 2025 The Author(s). Polymer International published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.

Background and Objective:Optical coherence tomography (OCT) enables high-resolution, both spatial and temporal, imaging of cerebrospinal fluid (CSF) motion in the subarachnoid space (SAS), but measurements are restricted to planes perpendicular to the imaging catheter. This study introduces a Physics-Informed Neural Network (PINN) framework for reconstructing 3D flow fields using planar velocity projections from arbitrarily oriented planes.Methods:The method targets steady or time-averaged low-Reynolds-number flows typical of CSF. It incorporates a projection-based loss function that enforces consistency between measured planar velocities and network predictions. To improve the model’s ability to deal with limited information, we implement gradient exponential moving average (Grad EMA) smoothing with amplification. CFD simulations provide ground truth velocity fields, and the network is trained using sparse projections mimicking realistic imaging constraints. Network performance is assessed by comparing reconstructed velocity fields against full 3D CFD solutions. The framework is first evaluated on synthetic CFD-generated flow over a confined fence featuring sharp gradients and multiscale structures. Next, the method is applied to a realistic canine SAS geometry reconstructed from intravascular OCT imaging.Results:We show that projection planes with a perturbation over 0.05 relative to the main flow axis contain sufficient axial information for accurate 3D reconstruction with or without Grad EMA. Grad EMA smoothing is critical, particularly for perturbations below 0.05, by stabilizing convergence and preventing collapse of low-gradient flow features into zero velocity. In the fence flow case, the network successfully resolves sharp shear layers and vortex structures but tends to predict zero flow in axial regions without Grad EMA. In the realistic SAS geometry, the PINN accurately recovers major flow pathways and complex features, with localized discrepancies near trabecular boundaries. The model shows strong agreement with CFD across the domain, even with limited and anisotropic projection sampling.Conclusion:The proposed PINN framework enables accurate 3D reconstruction of CSF flows from arbitrarily oriented planar projections. It performs robustly with minimal axial information and recovers multiscale features under imaging constraints. These findings support the feasibility of in-vivo 3D CSF flow quantification using OCT-based projection data.