Physics-informed neural network modeling of the thermo-mechanical behavior of liquid crystal elastomer artificial muscle

In-situ monitoring methods and deep learning models are increasingly being used for the quality assessment of parts fabricated using laser powder bed fusion to overcome the limitations of poor process repeatability. However, the massive data collection required for part-quality monitoring results in high transmission loads and storage costs. To address this problem, this study utilized the compressed sensing theory to acquire compressed photodiode signals. These signals were then used to train and test convolutional neural networks (CNN) to identify the lack-of-fusion, normal, and keyhole modes. At a compressive-sampling rate of 25%, the classification accuracy decreased from 93.1% (raw signals) to 79.3%. However, increasing the compression rate from 25% to 90% did not significantly decrease the classification accuracy. The linear mapping of the raw signal via a Gaussian measurement matrix causes coordinate information folding, thereby impairing the representation of latent features. Therefore, Gaussian process modeling was adopted for the features extracted using a pretrained CNN to mitigate the temporal information collapse and allow the compressed signals to achieve an accuracy comparable to that of the raw data. Furthermore, the sparsity and rank complexity of the melt-pool radiation signals were evaluated using sparse representation and principal component analysis.

Detection and classification of honeybee castes using thermal imaging and deep learning

GCN combined with snake convolution for enhanced topological perception in thrombotic hepatic portal vein segmentation.

Memory-efficient neural network training via gradient compression through continuous basis tracking

Paediatric mechanical circulatory support with Berlin Heart-EXCOR® Paediatric is predominantly used as a bridge to transplant or recovery, specifically in children up to 30 kg. While survival with ventricular assist devices has improved, insights into morbidity and quality of life remain limited. Safely discharging children, particularly with the new driving unit EXCOR® Active (BH-EA), is now of clinical interest. Multidisciplinary and caregiver perspectives are needed to inform practice. Through semi-structured interviews with 22 professionals (physicians, nurses, psychologists, engineers, physiotherapists, social workers, child education specialists, chaplains) and three caregivers of hospitalised children on BH-EA, we explored: (1) device safety and daily care; (2) hospital environmental factors; (3) requirements for transitioning home with EXCOR® Active. Qualitative analysis yielded three main themes; of which two are explored in this publication: alarm management and home-discharge requirements for paediatric BH-EA patients. Participants described frequent low-priority alarms contributing to alarm fatigue. They called for clearer procedures, shared responsibilities, and enhanced caregiver training and identified prerequisites for safe discharge, including a 24/7 emergency hotline, remote monitoring, comprehensive system-wide support, caregiver training, and strong healthcare networks. The interviews highlight that the BH-EA alarm management is conceptualised for in-hospital care, which leads to reservations concerning reliable home monitoring during medical events, such as blood clot formation. Multidisciplinary efforts are essential to enhance device safety, empower caregivers, and develop effective discharge programmes for children on BH-EA. Furthermore, organ allocation systems should be adjusted to avoid disadvantages in organ waiting times following home discharge.

Hepatitis B virus (HBV) infection can lead to hepatocellular carcinoma, and HBV integration into the host genome is regularly observed in the liver of chronic HBV carriers and is speculated to trigger carcinogenesis. To detect HBV integration, PCR-based methods are sensitive but may not be effective depending on the HBV integration site. High-throughput sequencing reveals not only HBV integration but also the site. However, it is still expensive for clinical applications. In situ hybridization is efficient and allows detection of DNA and RNA in single cells and has been applied to study HBV infection and dynamics. The technique, however, has not been applied to detect HBV integration. The task is challenging because of the small copy number of integrated HBV genomes, which results in a small signal-to-noise ratio. Here, we developed a fluorescence in situ hybridization approach for examining HBV integration in each liver cell. The obtained images of cells were analyzed using a deep learning model. Using several hepatoma cell lines with and without integrated HBV DNA, we showed that our trained neural network achieved an over 90% accuracy for majority of positive and negative control cells. This is the first proof-of-concept study showing that fluorescence imaging and deep learning are useful in detecting HBV integration at the single-cell level.

Tagged magnetic resonance imaging (tMRI) is a valuable tool for visualizing and quantifying tissue deformation in vivo. Its use is often hampered, however, by tag fading, long computation times, and the challenge of ensuring diffeomorphic, incompressible motion fields. In this paper, we describe a novel integration of the harmonic phase (HARP) approach to tMRI analysis with an unsupervised deep learning-based registration framework to estimate 2D and 3D motion fields that are diffeomorphic and nearly incompressible. The resulting method, called deep sinusoidally transformed HARP, or DSHARP, enables end-to-end network training by implementing a transformation of the harmonic phase to remove phase-wrapping discontinuities. It produces diffeomorphic motion by estimating a stationary velocity field from which motion is computed using the scaling and squaring technique. Finally, it encourages incompressibility using a novel Jacobian determinant loss term during network training. We evaluated DSHARP on 2D and 3D phantom data with simulated incompressible motions, real 3D human tongue data acquired during speech from both healthy and glossectomy subjects, and cardiac tagged MRI from the public STACOM 2011 benchmark. Our approach outperforms HARP, SinMod, SyN, PVIRA, VoxelMorph, and DeepTag in tracking accuracy, computation speed, and preservation of incompressibility.

Humanoid robots can be seamlessly integrated into human-robot interaction scenarios due to their human-like appearances. Pneumatic artificial muscles (PAM) are promising actuators for such robots due to their similarity to biological muscles, but their limited contraction ratio constrains both appearance and motion range of the robot. This work presents a 7-degrees-of-freedom (DoF) pneumatic humanoid robotic arm that mimics the human arm in both appearance and movement capabilities. A hybrid actuation scheme, combining direct PAM actuation at shoulder joints and PAM actuation with Bowden cable transmission at the distal joints, is adopted to enable anthropomorphic scaling with a lightweight and compliant structure. To address the control challenges posed by the nonlinear dynamics of PAMs and Bowden cables, Pneumatic Bowden cable Optimized Soft Actor-Critic (PBO-SAC), a model-free reinforcement learning framework, is proposed to enable efficient on-hardware control policy learning for the robotic arm. PBO-SAC incorporates posture-perturbed decoupled training and local recurrent fusion networks to ensure safe and smooth policy learning. Simulation results verify improvements in PBO-SAC, while hardware experiments on trajectory tracking and teleoperated stacking tasks further demonstrate the multi-DoF coordination control performance.

Under the influence of unsafe emotions, miners’ ability to perceive risks is hindered, which can easily lead to decision-making errors and safety accidents. To recognize unsafe emotions exhibited by miners during operations, this study proposes a deep learning-based bimodal framework that integrates speech and facial expression features. A convolutional neural network (CNN) combined with a bidirectional long short-term memory (Bi-LSTM) network is employed to model local spectral patterns and temporal dependencies in speech signals, and ShuffleNet-V2 is used to capture deep facial features. In addition, three feature enhancement strategies are proposed to improve the generalization ability of the model. By constructing a dataset containing five categories of miners’ unsafe emotions for network training, the model achieves a mean recognition accuracy of 85.56%. Furthermore, we conducted a preliminary field test of the bimodal model in a real mining environment. The results provide preliminary evidence of its potential applicability in real-world mining conditions.

The structural dynamics of DNA underpin essential biological processes, yet conventional structural biology methods often obscure conformational heterogeneity through ensemble averaging. Atomic force microscopy (AFM) provides single-molecule topographical maps capable of capturing both local and global variation, but extracting quantitative insight from these images remains challenging. Here, we introduce an automated framework that reduces AFM data to spline representations of the DNA backbone and applies cyclic Procrustes analysis to quantify shape similarity across ensembles. Using purified topoisomers of 339bp DNA minicircles ranging from relaxed to highly negatively supercoiled, we resolved and measured the relative abundance of conformational states across the different topoisomers, capturing gradual transitions among open circles, compact conformations, and self-crossing structures that are invisible to techniques such as gel electrophoresis or cryoelectron microscopy (cryo-EM). We show that beyond quantification, Procrustes distances provide supervisory signals for neural network training, enabling feature extraction tuned to conformational geometry and supporting robust conformation classification of AFM images. Extending the same spline representation to molecular dynamics simulations allows experimental and computational ensembles to be directly compared, establishing a common shape-based framework for probing conformational variability. Together, these advances transform AFM from a descriptive imaging tool into a quantitative platform for mapping conformational continua, with broad applicability to DNA and other dynamic biomolecular systems.

ABSTRACT The digital revolution has reshaped the landscape of education, offering new opportunities for learning. However, these benefits are not equally accessible to all educators, particularly those working with students with disabilities. Using the convergent parallel research design within the mixed study methods approach, this study examined the challenges and opportunities encountered by special needs educators in Kenya and the Czech Republic regarding digital literacy. It sought to enrich our understanding of the context-specific factors that shape the integration of digital skills into the pedagogical practises of educators in diverse settings. Data were collected from 456 educators of learners with disabilities (n = 127 from the Czech Republic and n = 329 from Kenya) using a self-administered questionnaire. The findings underscore the contextual difference shaping digital integration, with material resource and administrative limitations posing challenges in Kenya, while administrative support and resource use complexities were faced in the Czech Republic. Despite these challenges, special needs educators in both countries employed resourceful strategies for mitigation. This study emphasises the significance of tailored training, resource mobilisation and peer learning networks as recommendations to enhance digital integration. These insights could contribute to the discourse on promoting digital integration in education for students with disabilities on a global scale.

Rapid multi-parametric quantitative MRI via deep learning-based synthetic-to-real reconstruction and 3D SSFP-MOLED imaging.

IDH mutation & 1p/19q codeletion are critical biomarkers for glioma diagnosis & therapy. 1p/19q codeletion occurs exclusively in IDH-mutated gliomas. In this study, we developed a 2-stage, non-invasive, MRI-based deep learning method that leverages IDH status to enhance 1p/19q predictions. Multi-contrast brain tumor MRI & genomic information were obtained from five publicly available (TCIA, UCSF, EGD, UPenn & LGG), and three in-house/collaborator institutions (UTSW, NYU, UWM). Subjects were screened for the availability of IDH & 1p/19q status as well as T1, T1CE, T2, FLAIR MR images. For training purposes, missing T1 and FLAIR contrasts for the LGG database were generated using an in-house multi-contrast simulator. Two separate U-Nets were developed for 1p/19q-classification: a multi-contrast network (MC-Net) and a T2w-only network (T2-Net). A separate U-Net was developed for IDH classification (IDH-net). A total of 2044 subjects were used in training and testing IDH-Net, and 1426 subjects were used in training and testing the 1p/19q models. The IDH-Net was trained using subjects from TCIA, UTSW, and UPenn. The 1p/19q networks were trained using subjects from TCIA, UTSW, and LGG. The trained networks were tested on true held-out cases from NYU, UWM, EGD, and UCSF. In the 2-stage approach, subjects were initially classified for IDH status using IDH-Net. Predicted IDH-wildtype cases default to 1p/19q non-codeleted. Then the IDH-mutated cases were further classified for 1p/19q status using the 1p/19q-networks. IDH-Net achieved a classification accuracy of 93.7%. 1p/19q MC-Net & T2-Net achieved classification accuracies of 86.5% & 86.0%, respectively. In the 2-stage approach, 1p/19q MC-Net and T2-Net achieved accuracies of 91.5% & 91.2% respectively, improving the classification accuracy by ∼5%. This study demonstrates the effectiveness of leveraging IDH status to enhance 1p/19q classification. A ∼5% increase in classification accuracy was achieved when using the 2-stage approach, using IDH-Net to gate 1p/19q predictions. The developed method offers a reliable, non-invasive approach to determine important biomarkers for glioma diagnosis.

The observation that activation sparsity emerges in MLP blocks of standardly trained Transformers offers an opportunity to drastically reduce computation costs without sacrificing performance. To theoretically explain this phenomenon, existing works have shown that activation sparsity does not result from the data properties or data fitting but from the implicit bias of the training process. However, these connections are obtained with strong assumptions (e.g., shallow networks, a small number of training steps, and special training techniques), which cannot be applied to deep models standardly trained with a large number of steps. Different from these works, we find that the flatness of loss landscapes is also closely related to the MLP activation sparsity and can serve as a weaker assumption because it naturally emerges in the standard training of deep networks without the above strong assumptions. Specifically, we find that 1) the MLP activation sparsity equals a ratio between "augmented f latness" (a weighted sum of flatness measures) and the product of the input norm and activation gradient of the MLP. We empirically find that this ratio decreases during training, leading to sparse activations. 2) We also propose the notion of derivative sparsity, which reduces to activation sparsity under ReLU, but further enables pruning in the backward propagation and is more stable than activation sparsity. With the theoretical findings, we can further encourage activation sparsity by decreasing the numerator and increasing the denominator of the ratio: 1) To improve (lower) the flatness, we add different bias vectors to input tokens of MLP blocks to strengthen stochastic gradient noises that drive the model to a flat area. 2) We restrict the lower bound of affine parameters in LayerNorm to increase the input norm of MLPs. 3) To increase the activation sparsity, we propose an activation function JSReLU to encourage the search of parameters with sparse derivatives and sparse activations. These plug-and-play modifications can effectively reduce the ratio and produce sparser activations. Experiments on ImageNet-1K and C4 demonstrate relative improvements of at least 36% on inference sparsity and at least 50% on training sparsity over vanilla Transformers, indicating further potential cost reduction in both inference and training.

Understanding sickle cell dynamics requires accurate identification of morphological transitions under diverse biophysical conditions, particularly in densely packed and overlapping cell populations. In microfluidic sickling assays, simple dilution to reduce overlap is often undesirable because it reduces statistical power per experiment, and does not eliminate aggregation-driven clustering under hypoxic conditions. Moreover, longitudinal and cyclic deoxygenation-reoxygenation studies require tracking large cell populations within a single field of view, as all cells in the sample may undergo cumulative history-dependent changes. These experimental constraints necessitate robust quantification directly in dense suspensions. Here, we present an automated deep learning framework that integrates AI-assisted annotation, segmentation, classification, and instance counting to quantify red blood cell (RBC) populations across varying density regimes in time-lapse microscopy data. Experimental images were annotated using the Roboflow platform to generate labeled dataset for training an nnU-Net segmentation model. The trained network enables prediction of the temporal evolution of the sickle cell fraction, while a watershed algorithm separates overlapping cells to enhance quantification accuracy. Despite requiring only a limited amount of labeled data for training, the framework achieves high segmentation performance, effectively addressing challenges associated with scarce manual annotations and cell overlap. By quantitatively tracking dynamic changes in RBC morphology, this approach can more than double the experimental throughput via densely packed cell suspensions, capture drug-dependent sickling behavior, and reveal distinct mechanobiological signatures of cellular morphological evolution. Overall, this AI-driven framework establishes a scalable and reproducible computational platform for investigating cellular biomechanics and assessing therapeutic efficacy in microphysiological systems.

Abstract Reconstructive microsurgery is central to contemporary reconstructive surgery, enabling complex restoration of form and function across multiple anatomical regions. As clinical demands increase and operative exposure declines, the need for effective, structured microsurgical training has become more pressing. Despite advances in simulation, assessment, and curriculum design, substantial variability persists in how microsurgical training is delivered, validated, and sustained worldwide. This narrative review synthesizes the current global landscape of microsurgical training, examining key challenges and evidence-based educational strategies. Core components of effective training programs are identified, including technical skills acquisition, objective assessment, theoretical knowledge, research engagement, structured training pathways, and mentorship. A high-volume, longitudinal training model at Chang Gung Memorial Hospital is presented as an illustrative example. Future directions in microsurgical education are explored, with particular emphasis on competency-based curricula and entrustable professional activities. Key challenges include global variability and inequity in training access, limited longitudinal validation of training outcomes, reduced operative exposure, overemphasis on isolated technical skills, and shortages in mentorship capacity. Effective microsurgical training requires integration of technical, cognitive, non-technical, and professional competencies within longitudinal, context-rich educational frameworks. Emerging approaches—including distributed simulation, advanced virtual reality technologies, structured supervision, and international training networks—offer potential solutions to current limitations. Microsurgical education must evolve beyond episodic technical training toward integrated, competency-based frameworks that support progressive autonomy, patient safety, and independent practice. Although no single model is universally replicable, transferable principles from established programs can inform curriculum design across diverse settings. Continued investment in trainer development, outcome-driven educational research, and international collaboration will be essential to ensure the sustainability, equity, and effectiveness of future microsurgical training.

The rapid proliferation of sophisticated cyber threats has exposed critical limitations in conventional security architectures that rely on isolated, reactive tools. This paper presents ZeroGuardian-XDR, an intelligent and lightweight Extended Detection and Response (XDR) framework engineered to deliver real-time network threat detection, automated vulnerability assessment, and proactive incident alerting through a unified platform. The proposed system employs a trained autoencoder neural network for behavioral anomaly detection, enabling the identification of zero-day and previously unknown threats without reliance on static signature databases. ZeroGuardian-XDR integrates nine live global threat intelligence feeds including AlienVault OTX, Abuse.ch, Feodo Tracker, URLhaus, Blocklist.de, ThreatFox, NVD CVEs, MITRE ATT&CK, and EmergingThreats, collectively maintaining over 22,000 dynamic threat indicators automatically refreshed every six hours. The system maps all detections to the MITRE ATT&CK framework with 87% technique coverage across 8 tactical phases and 691 monitored techniques. A professional SOC-style web dashboard, multi-channel alert delivery via Telegram and email, automated PDF report generation, and an Nmap-powered CVE vulnerability scanner complete the integrated architecture. Experimental evaluation using five simulated zero-day attack scenarios demonstrated 100% detection accuracy with minimal false positive rates. The framework is deployed on Ubuntu Server 24.04 and made publicly available through open-source distribution with Windows and Linux installer packages. ZeroGuardian-XDR represents a scalable, cost-effective, and academically reproducible cybersecurity solution for modern network protection

DLR-YOLO: Dynamic low-rank training for a lightweight power tower object detection network in multi-scenario remote sensing images

Sub-visible particle analysis using flow imaging microscopy combined with deep learning has proven effective in identifying particle types, enabling the distinction of harmless components such as silicone oil from protein particles. However, the scarcity of available data and severe imbalance between particle types within datasets remain substantial hurdles when applying multi-class classifiers to such problems, often forcing researchers to rely on less effective methods. The aforementioned issue is particularly challenging for particle types that appear unintentionally and in lower numbers, such as silicone oil and air bubbles, as opposed to protein particles, where obtaining large numbers of images through controlled settings is comparatively straightforward. In this work, we develop a state-of-the-art diffusion model to address data imbalance by generating high-fidelity images that can augment training datasets, enabling the effective training of multi-class deep neural networks. We validate this approach by demonstrating that the generated samples closely resemble real particle images in terms of visual quality and structure. To assess the effectiveness of using diffusion-generated images in training datasets, we conduct large-scale experiments on a validation dataset comprising 500,000 protein particle images and demonstrate that this approach improves classification performance with no observable downside. Finally, to promote open research and reproducibility, we publicly release both our diffusion models and the trained multi-class deep neural network classifiers, along with a straightforward interface for easy integration into future studies, at https://github.com/utkuozbulak/svp-generative-ai.