Tumor metastasis is the primary cause behind the majority of cancer deaths, highlighting the urgent need for effective diagnostic and therapeutic strategies. In this paper, we rationally integrate a near-infrared (NIR) fluorescence dye IR820 and a magnetic resonance imaging (MRI) contrast agent GdL into bovine serum albumin (BSA) to construct a novel albumin-based NIR-II fluorescence/MRI bimodal imaging probe for precise diagnosis of tumor and metastatic lymph nodes (LNs). The bimodal imaging probe (BSA-IR820-GdL) could be quickly and effortlessly prepared by simply mixing the three components in accordance with a certain mix ratio. Compared to any single molecule, the bimodal imaging probe not only integrates NIR-II fluorescence and magnetic characteristics into one single nanoscale platform, but also can extend the fluorescence emission of IR820 from the NIR-I region to the NIR-II window and significantly enhance the relaxivity of GdL. Using NIR-II fluorescence and MR imaging technologies, we have successfully achieved dual-modality imaging of tumors and metastatic LNs in a lymph node metastasis mouse model. This study provides a valuable tool for preoperative tumor assessment and intraoperative surgical navigation.
Intelligent landscape recognition sits at the heart of smart coastal tourism, yet pushing vision Transformers onto resource-constrained edge hardware remains awkward-the self-attention mechanism scales quadratically with token count, and marine atmospheric haze quietly erodes recognition accuracy long before it is noticed in the laboratory. We propose a hybrid lightweight architecture that swaps standard self-attention in the early stages for depthwise separable convolution token mixers, switches to kernel-based linear attention in the deeper stages, and follows an adaptive channel reduction policy that trims feature dimensions where redundancy is empirically highest. A Fog-Aware Feature Calibration module, motivated by the Koschmieder atmospheric scattering model, is embedded between the convolutional and attention stages so that learned dehazing happens inside the network rather than as a separate preprocessing pass. Training proceeds through a three-phase pipeline that interleaves progressive structured pruning with temperature-scheduled knowledge distillation from a Swin-Base teacher; the composite objective shrinks the model to 4.8 M parameters and 0.91 GFLOPs without abrupt accuracy collapse. On a newly collected 17,565-image dataset spanning eight coastal scene categories across six shoreline regions of Zhejiang Province, our model reaches 92.6% ± 0.4% Top-1 accuracy (mean ± std over three independent runs), the best result among all baselines below 6 M parameters, including the recently released MobileViT-v2, FastViT, TinyViT, EfficientViT and RepViT. On an NVIDIA Jetson Orin Nano with TensorRT INT8 optimisation the system delivers 18.3 ms latency (54.6 FPS) within a 7.4 W average power envelope, and a 200 m visibility heavy-fog subset that mixes synthetic and real captures shows an 85.7% accuracy retention rate-6.8 percentage points above the strongest baseline. Cross-dataset zero-shot tests on Fujian, Hainan and the Places365 coastal subset, together with deployment benchmarks on Jetson Nano 4 GB and Coral Edge TPU, suggest the design transfers reasonably-though not effortlessly-beyond the original collection sites.
Food allergy, a worldwide health problem due to its rising prevalence, is an immune reaction to allergens. Among these, peanut allergy poses significant health risks and underscores the critical need for accurate detection methods. This study focuses on developing an electrochemical aptasensor based on a gold nanoparticle and Ara h 1 specific aptamer decorated screen-printed electrode for the rapid, sensitive, and portable detection of a major peanut allergen. Integrated with mobile devices, the aptasensor allows users to obtain experimental results effortlessly and conveniently. The developed electrochemical aptasensor operates by measuring changes in charge transfer on the electrode surface upon the selective binding of Ara h 1 to the immobilized aptamer. The aptasensor effectively identified Ara h 1 residues in food samples, exhibiting a linear detection range of 500-25,000 ng/mL and a detection limit of 500 ng/mL. The aptasensor's applicability was verified using food samples with measurements performed via a mobile potentiostat.
The elaborately crafted microstructure and the accommodated active species of catalyst hold great significance for improving catalytic performance. The synthesis of hierarchical multifunctional catalysts through the integration of two or more distinct active sites in porous supports (e.g., metal-organic frameworks (MOFs)) have provided promising schemes for designing sophisticated catalytic systems, especially in the tandem synthesis of complex molecules. However, the construction of such multifunctional catalytic systems is complicated. Additionally, their powder state of these catalysts inevitably exhibits poor machinability and recyclability. More importantly, the design of classical monolithic catalysts faces the challenge of constructing exquisite secondary micro-nanostructures that favor mass transfer and catalytic kinetic processes. Herein, we fabricate a composite monolithic agarose (AG) aerogel catalyst Pd@H-UiO-66-NH2/AG (H denotes the hollow structure) equipped with secondary hollow-structured amino-functionalized hafnium (Hf)-based MOF nanoreactors (Pd@H-UiO-66-NH2), which precisely encapsulates palladium nanoparticles (Pd NPs). Benefiting from the enhanced kinetic process afforded by the well-crafted secondary hollow structure and the integration of Pd NPs, MOF-positioned Lewis acid sites (Hf4+) and amino groups within the monolithic catalyst, Pd@H-UiO-66-NH2/AG exhibits high catalytic performance in the synthesis of benzylidene malononitrile (BM) from benzyl alcohol (BA) and malononitrile, achieving a 95.2 % yield. Furthermore, the composite monolithic catalyst can be effortlessly recovered for at least five times without a significant loss of activity. The monolithic catalysts fabricated by equipping multifunctional nanoreactors herein provide distinctive insights for the development of practical application-oriented monolithic catalysts for tandem synthesis.
During the first period of life, human infants rapidly and effortlessly acquire the languages they are exposed to. Although memory is central to this process, the nature of early verbal memory systems, and the factors that determine retention and forgetting, remain largely unknown. Behavioral and brain measures have demonstrated memory formation in newborns. However, word traces fade in the face of acoustic overlap, leading to interference and forgetting. Here, we investigate whether speakers' identity changes facilitate the separation into distinct acoustic episodes and the creation of non-overlapping verbal memories. Newborns (0-4 days-old) were tested in a familiarization-interference-test protocol, while neural cortical activity was recorded using functional Near-Infrared Spectroscopy (fNIRS). The results showed higher neural activation to novel words than to familiar ones during the test phase, indicating that the infants recognized the familiar words despite potentially interfering sounds. The recognition response was measured over the left inferior frontal gyrus (IFG) and superior temporal gyrus (STG) areas known to be crucial for encoding auditory information and language processing. The neural response also included the right IFG and STG, involved in interpreting vocal social cues and speaker recognition. The results indicate that speaker identity is a key feature in the formation of verbal memories from birth, facilitating separability, possibly through early source-content binding (i.e. what-who), a precursor to fully mature episodic memory. When we remember events, we often do not just recall what happened, but also where and when it was, and who was there. This is because human memory tends to merge different features of experiences into whole episodes rather than store them as separate items. Combining all these features is what makes memories stronger and easier to remember in the long term, shaping who we are and how we act. Our earliest episodic memories probably go back to the first years of life. But linguistic memory may develop earlier. Infants start learning a language before they can speak. By four months, they can respond to their own name and by six months, they can recognise common words. This suggests that they already possess the ability to store and retrieve linguistic information. But what is the nature of these earliest memories, and when are they first formed? To find out more, Visibelli et al. tested brain signatures of episodic-like, linguistic memories in infants a few hours after birth. They designed a task to examine verbal memory formation in the presence of different speakers. Newborns, while lying in their hospital cribs, were exposed to a word pronounced by one speaker and then listened to the same or a different word three minutes later, after having been exposed to interfering words pronounced by another speaker. Visibelli et al. investigated whether a change in who produces the words (the who) helps newborns create separate, retrievable memory traces for the words (the what). The researchers used a non-invasive method known as functional Near-Infrared Spectroscopy, which uses near-infrared light to monitor changes in brain oxygenation. The recordings showed that newborns’ brains responded differently to a word they had heard before than to one they had not. This difference in response indicates that the brain distinguishes the familiar word from a new one - a sign that a memory had been stored and retrieved. Crucially, these memories were maintained only when different speakers produced both the familiar and interfering words. When recognition was successful, brain regions involved in speech and voice processing became active, suggesting that infants were not only processing the words themselves, but also who was speaking. This suggests that, at birth, newborns readily link words to speaker identity, encoding both what is being said and who is speaking. The findings of Visibelli et al. open new directions for understanding memory and language development from the very first hours of life. The data suggest that speaker identity is a key feature of speech, enabling episodic-like memories of word sounds from birth and offering evolutionary advantages at the outset of human communication. They also raise the possibility that difficulties with feature binding (when the brain combines different attributes of an object or sound) may be detectable at a very early stage – before any language difficulties become apparent – thereby creating new opportunities for early identification and intervention.
Artificial intelligence (AI) systems based on Large Language Model (LLM) are becoming an increasingly important aspect of biomedical research, assisting with the tasks ranging from research design to data analysis and publication. Although AI systems increase productivity by cutting the time taken for individual tasks, they also expose their users to severe risk due to systematic distortion of outputs due to algorithmic sycophancy. The honesty of these AI systems is questionable, and can effortlessly crumble when user prompts are incorrect or when the system is under pressure. This viewpoint emphasizes the fundamental understanding of algorithmic sycophancy and the potential mechanism underlying it, which leads to systematic distortion of biological research. There is an important need to bring this issue to light in order to prevent systematic distortion of biomedical research through the cautious utilization of these LLM-based AI systems. Understanding this threat can also help to minimize the propagation of unreliable findings and literature, which pose a significant safety risk to biomedical research as a whole.
Purpose To develop and validate a deep neural network that simultaneously segments brain tumors and anatomic structures, regardless of the contrast and resolution of the input scans, and can effortlessly adapt to unseen modalities. Materials and Methods The authors included various MRI scans from patients with and without brain tumors from four different datasets. Patient data were divided into a training set and a test set. The authors' method, TumorSynth, combines a Bayesian generative model and a deep learning segmentation model. The generative model creates paired synthetic labels and images with simulated tumors and brain tissues, providing a rich dataset for training the segmentation model. The authors quantitatively compared its performance with that of other widely used methods by calculating Dice similarity coefficients (DSCs). Results A total of 1971 patients with and without tumors were included in the study (training set, n = 351 patients; test set, n = 1620 patients). The median DSCs for segmentation (authors' method vs reference standard) were 0.89 (IQR, 0.83-0.95; P < .001) for the unaffected brain volume and 0.89 (IQR, 0.84-0.94; P < .001) for the tumor region. There were no differences in parcellation performance when an MRI sequence was missing (P = .07). In cross-modality validation, the authors' method achieved DSC values of 0.88 for apparent diffusion coefficient, 0.85 for diffusion-weighted imaging, 0.80 for susceptibility-weighted imaging, and 0.79 for fractional anisotropy images. The authors observed a 4% false-positive rate when processing tumor-free MR images. Conclusion The authors developed a deep neural network for brain tumor and tissue segmentation, validated its performance across standard structural MRI sequences, and determined its generalizability to unseen data. Keywords: Segmentation, Neuro-Oncology, CNS, Deep Learning, Neurosurgery Supplemental material is available online for this article. © RSNA, 2026.
The rapid growth of born-digital PDF documents has amplified the demand for fast, precise tabular data extraction on an industrial scale. State-of-the-art deep-learning approaches have high accuracy, but at the resource expense of substantial computational complexity, data-hungry training process and black-box incomprehensibility, confining deployment in the real world. In this paper, we introduce SPARTAN (Structured Parsing and Relevant Table Analysis), an entirely open-source, heuristic-based pipeline, with high-fidelity table detection and extraction and no model training or GPU requirements. SPARTAN mixes lightweight OpenCV image-processing modules: column whitespace analysis, boundary and text-based region segmentation and line segment cell parsing, with a modular OCR layer and optional post-processing hooks for LLM-driven schema mapping. We evaluated SPARTAN on more than 20 K pages of PCN-480 (Product Change Notification and Product Discontinuance Notification), scientific papers, certificates and datasheets and reported 0.94 precision, 0.91 recall and 0.93 F1-score, with 96.7% OCR character accuracy, processing a page in 2.8 s CPU time on an average, and requiring 1.2 GB peak RAM on the most demanding PDFs. Our model outperformed Tabula, Deepdoctection, TabbyPDF and EMbTTBF in accuracy and speed. It must be noted that this comparison was conducted against standard inference configurations for the learning-based baselines; a performance evaluation against highly-optimized, edge-device deployments is a consideration for future work. Its rule transparency effortlessly copes with borderless, nested and merged-cell layouts that easily outsmart classical heuristics, without incurring the resource cost of end-to-end neural pipelines. SPARTAN's CLI-governed, swap-in-swap-out architecture encourages domain tuning, edge deployment and cloud-scalable REST service wrapping, making it a practical bridge between brittle rule systems and heavyweight AI for document-understanding pipelines. The work proves that well-crafted modernized heuristics, combined with high-quality OCR, can match or even outperform state-of-the-art deep learning while remaining within reach of small and medium enterprises, thus re-opening a critical gate to cost-efficient, explainable PDF table extraction.
Individuals with severe neurological injuries often rely on assistive technologies, but current methods have limitations in accurately decoding multi-degree-of-freedom (DoF) movements. Intracortical brain-machine interfaces (iBMIs) use neural signals to provide a more natural control method but currently struggle with higher-DoF movements: something the brain handles effortlessly. It has been theorized that the brain simplifies high-DoF movement through muscle synergies, which link multiple muscles to function as a single unit. These synergies have been studied using dimensionality reduction techniques like principal component analysis (PCA), non-negative matrix factorization (NMF), and demixed PCA (dPCA) and successfully used to reduce noise and improve offline decoder stability in non-invasive applications. However, their effectiveness in improving decoding and generalizability for implanted recordings across varied tasks is unclear. Here, we evaluated whether brain and muscle synergies can enhance iBMI performance in non-human primates performing a two-DoF finger task. Specifically, we tested if PCA, dPCA, and NMF could compress and denoise brain and muscle data and improve decoder generalization across tasks. Our results showed that while all methods effectively compressed data with minimal loss in decoding accuracy, none improved performance through denoising in our datasets. Additionally, none of the methods enhanced generalization across tasks. These findings suggest that while dimensionality reduction can aid data compression, extracting synergies alone did not provide an advantageous or cleaner control space for linear decoding in our study. Further research with larger sample sizes and more channels in muscle recordings is required to determine whether synergies can be leveraged as an optimal control framework or if alternative approaches are required to enhance decoder robustness in iBMI applications.
Highly stretchable triboelectric nanogenerators (TENGs) are indispensable for conformal energy harvesting and self-powered sensing. The hydrogel-based TENGs have demonstrated encouraging performance in the fabrication of flexible and transparent devices. Here, we introduce a transparent and stretchable conductive organohydrogel which was synthesized in a water/glycerol co-solvent system via cross-linking of poly(vinyl alcohol) (PVA) and borax. The electrical conductivity of the PVA/borax organohydrogel can be tuned over a broad range simply by adjusting the borax concentration. The prepared organohydrogel can be utilized as a resistance sensor to monitor human motions. A single-electrode TENG was developed by employing the PVA/borax organohydrogel as an electrode and silicone rubber as a triboelectric layer. An optimally formulated organohydrogel-based TENG (OH-TENG) delivers a peak-to-peak voltage of approximately 500 V, a short-circuit current of 3.0 µA, and a transferred charge of 145 nC under a 3.0 Hz mechanical excitation. Demonstrations show that the device rapidly charges an electrolytic capacitor, effortlessly illuminates a string of green LEDs, and powers portable electronics. When interfaced with Darlington transistors and relay modules, the OH-TENG can reliably switch external circuits on and off. It has also been integrated with a Bluetooth oscilloscope module, enabling real-time monitoring of human movements. These results highlight its potential applications in human-machine interfaces and safety systems. This study elucidates how organohydrogel properties govern the performance of OH-TENGs and provides a general blueprint for designing next-generation, highly stretchable TENGs.
Humans effortlessly interpret images by parsing them into part-whole hierarchies. Yet, deep learning models, despite excelling at capturing multi-level features, often fail to explicitly encode these part-whole hierarchies-an essential aspect of medical imaging, which boasts anatomical hierarchies in nature. To address this limitation, we introduce Adam-v2, a self-supervised learning framework that explicitly learns to encode inherent part-whole hierarchies within medical images through three key branches: (1) "localizability", which acquires discriminative representations to distinguish different anatomical structures; (2) "composability", which learns each anatomical structure in a parts-to-whole manner; and (3) "decomposability", which comprehends each anatomical structure in a whole-to-parts manner. Our extensive experiments showcase Adam-v2's advanced capability in anatomy understanding, unveiled through its embeddings (Eve-v2). Particularly, Eve-v2 demonstrates a zero-shot understanding of anatomy as revealed through its learned properties: ➀ preserving localizability of anatomical structures and ➁ encoding part-whole relations of anatomical structures as well as its emergent properties: ➂ understanding anatomical layouts via interpolation and extrapolation, ➃ associating each image pixel with (dense) semantic embeddings, ➄ recognizing anatomical symmetries, ➅ generating embeddings consistent across scales, and ➆ matching anatomical structures across images of the same patient with different diseases, images of different patients, and augmented views of the same image. Adam-v2 also offers robust and generalizable representations and stands out in ➇ few-shot learning, ➈ full-transfer learning, and ➉ novelty and anomaly detection. These capabilities and performance directly stem from our crafted anatomy learning strategy, which explicitly constructs hierarchies for distinct anatomical structures from unlabeled medical images. Project page: GitHub.com/JLiangLab/Eden.
Omniphobic NP-GLIDE (no-problem-to-glide) coatings allow virtually any everyday liquid─water, oils, sauces, biological fluids─to roll off effortlessly, making them powerful candidates for self-cleaning and protective surface technologies. Yet, as with all coatings, surface wear and microdamage remain unavoidable during use, creating a pressing need for materials that can reliably repair themselves. To address this challenge, we introduce a new class of inorganic-organic self-healing NP-GLIDE coatings based on polyhedral oligomeric silsesquioxane (POSS) cages bearing thiol (SH) and tert-butylamino (tA) groups, with only ∼1% of the thiol groups grafted with 2.0 kDa antismudge poly(dimethylsiloxane) chains. Cross-linking these multifunctional POSS precursors with either a flexible hexamethylene diisocyanate trimer (HDIT) or a rigid isophorone diisocyanate (IPDI) generates networks containing dynamic thiourethane (from SH) and hindered urea (from tA) linkages. Systematic comparisons reveal that IPDI-cross-linked coatings achieve markedly higher nanoindentation hardness and more robust antismudge performance than their HDIT analogues. Furthermore, tuning the tA/SH ratio provides a molecular handle to balance mechanical strength and autonomous repair: higher tA content accelerates both surface and bulk healing at the expense of hardness. Strikingly, the IPDI-cross-linked materials combine hardness values between those of poly(ethylene terephthalate) and polystyrene with rapid self-healing, establishing a versatile design platform for durable, high-performance NP-GLIDE coatings.
Noise ranks as the world's second-largest environmental risk factor according to the WHO. The simultaneous achievement of ultrabroadband and perfect or near-perfect noise absorption is a quite significant yet long-standing challenge. Here, we propose a "gradient pore circulation (GPC)" strategy for building a hierarchical ordered architecture of green aerogels by using highly active microfibers precisely dissociated from a wood S2 sublayer as basic units. The aerogels comprise anisotropic, parallelly layered microchannels enriched with multilevel pores within each layer, alongside abundant spring-shaped strips bridging these adjacent layers. Under the "GPC" strategy, the soundwaves effortlessly enter the parallelly layered microchannels possessing moderate flow resistance, while the synergy of long microchannels, multilevel pores, and abundant interlayer strips creates plentiful closed loops, fostering a repetitive cyclic reflection-friction-dissipation of soundwaves. Under these synergies, the aerogels achieve near-perfect acoustic absorption properties, with a sound-absorption coefficient (SAC) of 0.95 to 1 across an ultrabroad frequency range of 520 to 6300 Hz and a superb noise-reduction coefficient of 0.82, the highest recorded to date. More significantly, the aerogels retain excellent sound absorption (SAC > 0.85) even under extreme temperatures (-196 to +80 °C), and high humidities (up to 98%) and salt spray environments with mild modifications. Moreover, the aerogels are biodegradable, superelastic, and have strong compression fatigue resistance over 1000 cycles, manifesting great potential as sustainable sound absorption materials for diverse applications.
Conversations can feel effortlessly engaging or, conversely, difficult and unrewarding. Multiple factors contribute to the experienced quality and outcomes of a conversation, among them how interlocutors align with each other. The present study investigated speech-to-speech, brain-to-speech, and brain-to-brain coordination as markers of interpersonal alignment, examining their relationship with jointly perceived interaction quality and mutual affinity between conversational partners. Pairs of previously unacquainted participants (dyads) engaged in multiple short, free-form conversations on topics of varying interest while their vocal and neural activity were simultaneously recorded in a dual-EEG ("hyperscanning") setup. We analyzed interlocutors' prosodic adaptation, neural speech tracking, and neural coordination during each conversation. At the speech-to-speech level, our findings reveal that partners with more positive mutual impressions became more similar in their volume and voice quality over the course of the experiment session, reflecting greater prosodic convergence. At the brain-to-speech level, we found no reliable effect of interaction quality on neural tracking of unfolding speech within any individual region, although topographical differences suggested relative modulation across scalp sites. Finally, at the brain-to-brain level, our findings show that higher perceived interaction quality enhanced inter-brain relationships across frequency bands (alpha and theta) and temporal dependencies (concurrent/near-instantaneous and recurrent/listener-lagging), with the strongest effects observed for concurrent alpha-band coupling. These findings suggest that distinct coordination processes are involved in how interlocutors experience an interaction and how they establish relational affinity, casting new light into the mechanisms that make a conversation worthwhile.
Blockchain-based health information exchange (HIE) has received increased attention in health care research and practice over the last years. It enables the sharing of patient information across healthcare organizations, provides higher levels of data confidentiality and security, and reduces time and costs in collaborative medical decision-making. To make informed decisions on the implementation of blockchain-based HIE in practice and to fully understand the implications of its use for patient care, it is important to gain insights into patient perceptions of and interactions with blockchain-based HIE. This study aimed to assess patient perceptions of a blockchain-based HIE mobile app to inform its iterative development. We used a mixed methods user-centered design in 3 phases to iteratively assess patient perceptions of blockchain-based HIE: (1) structured questionnaires collecting patient requirements for blockchain-based HIE, (2) semistructured interviews evaluating mobile app mock-ups, and (3) a survey with blockchain-based HIE scenarios related to their patient care using the technology acceptance model, System Usability Scale, and open feedback. Both the semistructured interviews and the survey were conducted in a clinical setting with patients with cancer undergoing treatment at a major university hospital in Germany. As an exemplary case, we deem patients with cancer as well-positioned to evaluate a blockchain-based HIE mobile app since their treatment requires extensive coordination and data sharing across health care providers. Our findings support that patients have a high intention to use the blockchain-based functions that enable them to define, track, and revoke access to their health data per health care facility and service provider. Patients rated the 4 key functionalities (connection with providers, document sharing, a health diary, and a health care service provider search) as both useful and easy to use. The overall System Usability Scale of the blockchain-based HIE mobile app improved over the 3 phases up to 77.34, showing a good overall usability. The open feedback showed that patients' perceived usefulness of a blockchain-based HIE mobile app is especially influenced by 3 factors: the acceleration of the process of data sharing, patient-centered access control, and alignment with the respective health care settings. Moreover, patients' perceived ease of use of a blockchain-based HIE mobile app is impacted by 3 additional factors: the intuitiveness of the interaction, an aesthetic and functional design, and individual differences such as age or literacy with document management systems. The evaluation demonstrates that patients are inclined to use blockchain-based HIE to manage their health data, as it empowers them to control which health care providers or individuals can access their information. To foster the use of a blockchain-based HIE mobile app, the app should allow patients to effortlessly establish connections with health care providers, offer an overview of all patient data, and enable patients to share medical documents individually via the app.
The "BlockCare" addresses the growing challenge of managing and accessing personal health records scattered across various healthcare providers. Patients often struggle with fragmented and inaccessible health information, leading to delays and potential errors in their care.1,2 We have expanded the system's technical architecture to ensure interoperability and compliance with global healthcare regulations. BlockCare offers a secure, centralized platform where users can store, manage, and easily access their complete health data, including medical history, lab results, and treatment records.3,4 Detailed security mechanisms such as cryptographic hashing, multi-signature authentication, and a decentralized access control model have been incorporated.5 Utilizing advanced encryption and robust authentication methods, the app ensures the highest level of data privacy and security.6,7 It integrates seamlessly with different healthcare systems, providing real-time updates and allowing users to share their information effortlessly with healthcare professionals.8 The intuitive interface simplifies the process of retrieving and managing health records, empowering patients to make informed decisions about their care and improving overall care coordination.9,10 The manuscript now discusses compliance with Health Insurance Portability and Accountability and General Data Protection Regulation, ensuring legal and ethical handling of sensitive health data.11 Countries like Greece, where comprehensive regulatory frameworks for electronic health record (her) adoption are still emerging, could greatly benefit from decentralized health record solutions like BlockCare.
This paper presents a novel approach for generating high-quality, cross-category 3D models from free-hand sketches with limited training data. We propose the first semi-supervised learning method to our knowledge for sketch-to-3D model conversion. Innovatively, we design a coarse-to-fine pipeline to perform the semi-supervised learning in the coarse stage and train a diffusion-based refiner to get a high-resolution 3D model. We designed a sketch-augmentation method for semi-supervised learning and integrated priors such as CLIP loss, shape prototypes, and adversarial loss to help generate high-quality results even with abstract and imprecise sketches. We also introduce an innovative procedural 3D generation method based on CAD code, which helps pre-train part of the network before fine-tuning with limited real data. Our approach, coupled with a specifically designed curriculum learning, allows us to generate high-quality 3D models across multiple categories with as few as 300 sketch-3D model pairs, marking a significant advancement over previous single-category approaches. In addition, we introduce the KO2D dataset, the largest collection of hand-drawn sketch-3D pairs to support further research in this area. As sketches are a far more intuitive and detailed way for users to express their unique ideas, we believe that this paper can move us closer to democratizing 3D content creation, enabling anyone to transform their ideas into 3D models effortlessly.
Humans effortlessly relate what they see to what they know, drawing on existing knowledge of the perceptual, conceptual, and contextual attributes of objects while searching for and recognizing objects. Although prior studies have investigated the temporal dynamics of perceptual and conceptual object properties in the neural signal, it remains unclear whether and when contextual associations are uniquely represented. In this study, we used representational similarity analysis on electroencephalography (EEG) data to explore how the brain processes the perceptual, conceptual, and contextual dimensions of object knowledge over time. Using human similarity judgments of 190 naturalistic object concepts presented as either images or words, we constructed separate behavioral models of the perceptual, conceptual, and contextual properties of objects. We correlated these models with neural patterns from two EEG datasets, one publicly available and one newly collected, both recorded while participants passively viewed the same object stimuli. Across both datasets, we found that perceptual features dominated the early EEG response to object images, and conceptual features emerged later. Contextual associations were also reflected in neural patterns, but their explanatory power largely overlapped with that of conceptual models, suggesting limited unique representation of the contextual attributes of objects under passive viewing conditions. These results highlight the integration of perceptual and conceptual information by the brain when processing visual objects. By combining high temporal resolution EEG with behaviorally derived models, this study advances our understanding of how distinct dimensions of object knowledge are encoded in the human brain.
People effortlessly recognize objects of various materials and predict their behavior from visual information. However, research on causal perception has largely underexplored dynamic interactions involving nonrigid objects, often treating spatiotemporal causality as independent of object properties. To address this gap, we introduce a novel causal perception phenomenon in which speed profiles alone evoke the perception of elastic, nonrigid motion in interactions between simple geometric figures, ultimately shaping causal perception. Using an ambiguous-motion stimulus involving a separation event, we show that a line segment elongating with a disc at its end before separating can be interpreted in two distinct ways-as a rigid stick pushing a disc or as a disc pulling an elastic band. Across four experiments, deceleration before separation followed by rapid postseparation motion consistently biased perception toward the elastic band interpretation, demonstrating the critical role of kinematic regularities in shaping both causal and material perception. Four follow-up experiments using a pause-detection task revealed sensitivity to motion dynamics inconsistent with a stretched elastic band, even when causal and material perception was entirely task-irrelevant, further indicating the perceptual nature of this phenomenon. These findings illustrate how subtle kinematic patterns can simultaneously reverse perceived force dynamics and causal roles, accompanied by corresponding shifts in material perception, contributing to a unified framework for material and causal perception. Ultimately, this work provides new insights into how the visual system uses kinematic information to assign causal agents and patients in dynamic interactions. (PsycInfo Database Record (c) 2026 APA, all rights reserved).