Oxidative-stress-related diseases generate elevated levels of hydrogen peroxide (H2O2) and other redox-active species in exhaled breath condensate (EBC), offering a promising avenue for non-invasive disease assessment. Despite its diagnostic potential, clinical translation of EBC redox species detection has been constrained by bulky analyzers, labor-intensive operation, and limited sensitivity. By harnessing the redox potential, we present a Redox-mediated Optochemical Plasmonic Sensing (ROPS) system, an ultrasensitive platform that enables real- time, easy-to-use, point-of-care quantitative analysis of H2O2-equivalent reducing capacity in EBC. By coupling H2O2-triggered gold nanoparticle growth on a phase-modulated plasmonic sensor, the ROPS system based on the phase-modulated interferometry achieved a detection limit of 1.19 pM and a wide dynamic range across 4 orders of magnitude. Moreover, the ROPS system demonstrated excellent reusability, maintaining stable H2O2 sensing performance over 40 reuse cycles. To enable point-of-care applications, we further developed a handheld high-efficiency EBC sampler, which harvested exhaled condensate at 2.49 μL/s and enabled an EBC- ROPS sampling-to-biosensing workflow down to 15 min. Clinical evaluation using EBC samples from lung and esophageal cancer patients and matched controls confirmed reliable quantification of trace H2O2-equivalent Au (III)-reducing capacity in complex EBC matrices, highlighting the potential of the integrated ROPS platform as a practical tool for point-of-care and non-invasive oxidative stress-related disease assessment.

Metabolic fingerprinting offers a rapid, minimally invasive route to in vitro diagnostics because small-molecule metabolites respond quickly to disease and therapy. Laser desorption/ionization mass spectrometry (LDI-MS) is attractive for this purpose due to simple workflows and high throughput, yet conventional organic matrices generate strong low-mass background and variable crystallization that can mask metabolites and undermine reproducibility. Semiconductor nanomaterials provide an alternative substrate class with low background and engineerable electronic/thermal properties. Recent progress shows that the LDI performance can be actively tuned by materials design, rather than trial- and-error matrix selection. By adjusting band structures, introducing defects, and building heterojunction/ metal-semiconductor interfaces, semiconductor substrates can couple laser energy into photothermal heating and interfacial charge transfer, improving desorption/ionization efficiency while maintaining clean spectra. These advances have enabled direct metabolic fingerprinting of complex biofluids with tiny sample volumes, short acquisition times, and growing compatibility with machine learning models for diseases including cancers, neuropsychiatric disorders, and metabolic/systemic conditions.

Dexterous manipulation is a fundamental yet unresolved topic in robotics, particularly in the realm of micro-manipulation of flexible objects. The complex interplay among external forces, intrinsic material properties, and environmental factors that govern shape deformation presents significant challenges to learning and control. This paper introduces a mechano-flow co-manipulation scheme for flexible micro-objects. It achieves precise handling of floating micro-objects by dynamically adjusting the flow field at the air-liquid interface, enabling simultaneous control of both manipulation and focal planes. A real-time flow prediction framework is proposed based on a reversible neural network, termed FlowNet, achieving bidirectional prediction of input flow and resulting spatio-temporal flow distribution with high accuracy. Through experimental validation, including linear-drive characterization, open-loop trajectory following, and manipulation of different flexible objects, we demonstrate the system’s versatility, precision, and stability. The proposed system offers a promising approach for micro-manipulation through mechano-flow interaction for flexible micro-objects.

Mechanical stress is a fundamental aspect of soft tissues that influences cellular behavior and tissue integrity, yet noninvasive imaging of stress in vivo remains a major challenge. Conventional magnetic resonance elastography (MRE) measures tissue stiffness but not stress. Employing a physics-informed model, we analyze reverberant shear wave fields recorded by MRE to capture the 3D mapping of tissue stress. By decomposing reverberant waves into anisotropic traveling wave components, we establish a direct relationship between wave speed, polarization, and local stress, without requiring wave direction knowledge. This method was validated with numerical simulations and phantom experiments. Applied to patients with meningiomas and pituitary adenomas, it generated high-resolution stress maps consistent with anatomy and physiology. Cortical stress measurements enabled intracranial pressure estimation, uncovering trends related to patient age and pathology. Our results demonstrate that noninvasive stress imaging provides a novel quantitative biomarker for mechanobiological research and suggests new possibilities for clinical applications.

Surgical instrument recognition and tissue motion analysis are two critical tasks for instrument-tissue characterization (ITC) in robot-assisted minimally invasive surgery, with broad applications in surgical scene understanding and 3D reconstruction. While precise instrument segmentation and detailed motion map estimation are the foundation of these tasks, existing methods commonly address these two tasks independently, and little attention has been paid to the fusion of these two tasks for exploring the intrinsic relationships between spatial segmentation and temporal motion. To address this, we propose ITCNet, a multi-task self supervised instrument segmentation and motion estimation network for instrument-tissue characterization, which simultaneously predicts motion maps and instrument masks from consecutive images. The proposed ITCNet is built upon a shared feature encoder for extracting unified representations and two task-specific decoders for each individual task. To dive into the inherent relationships between spatial segmentation and temporal motion, several fusion constraints are designed to facilitate the multi-task training process, including a motion-based geometric segmentation consistency and a segmentation-based edge-aware motion consistency. Meanwhile, the proposed multi-task fusion scheme enables the adaptation across different surgical scenes with few-shot and zero-shot learning strategies. Extensive experiments have been conducted on three public datasets for comparison with existing methods. Quantitative and qualitative results demonstrate that our proposed method can accurately segment surgical instruments and predict tissue motions across various surgical scenarios. Furthermore, application results on our established surgical robotic system emphasize its potential for real-world surgical robotics.

Multimodal large language models (MLLMs)have achieved remarkable success in general per-ception, yet complex multi-step visual reasoningremains a persistent challenge. Although recentagentic approaches incorporate tool use, they of-ten neglect critical execution feedback. Conse-quently, they suffer from the imagination-action-observer (IAO) bias, a misalignment betweenprior imagination and observer feedback that un-dermines reasoning stability and optimality. Tobridge this gap, we introduce V-ABS, an action-observer driven beam search framework that en-ables deliberate reasoning through thinker-actor-observer iterations. We also propose an entropy-based adaptive weighting algorithm to mitigatethe IAO bias by dynamically balancing the con-fidence scores between the policy priors and theobservational feedback. Moreover, we constructa large-scale supervised fine-tuning (SFT) datasetcomprising over 80k samples to guide the modelto assign higher prior confidence to correct ac-tion paths. Extensive experiments across eightdiverse benchmarks show that V-ABS achievesstate-of-the-art performance, delivering an aver-age improvement of 19.7% on the Qwen3-VL-8Bbaseline and consistent gains across both open-source and proprietary models. Code is availableat https://github.com/pami-zwning/V-ABS.

Implicit neural representation has demonstrated promising results in 3D reconstruction in various scenes. However, existing approaches either struggle to model fast-moving objects or are incapable of handling large-scale camera ego-motion in urban environments. This leads to low-quality synthesized views of the large-scale urban scenes. In this paper, we aim to jointly solve the problems caused by large-scale scenes and fast-moving vehicles, which are more practical and challenging. To this end, we propose a progressive scene graph network architecture to learn the local scene representations of dynamic objects and global urban scenes. The progressive learning architecture dynamically allocates a new local scene graph trained on frames within a temporal window, with the window size automatically determined, allowing us to scale up the representation to large-scale scenes. Besides, according to our observations, fast-moving objects are observed only in a few frames, which leads to a significant decline in reconstruction accuracy for dynamic objects. Therefore, We introduce a foundation-guided object representation that extracts object-centric visual priors and conditions the density and color decoders in normalized object coordinates. We further propose a frequency-progressive regularization strategy that gradually exposes high-frequency positional, directional, and pose encodings during training, reducing overfitting to sparse object observations. Experimental results demonstrate that our method achieves state-of-the-art view synthesis accuracy, object manipulation, and scene roaming ability in various scenes.

Restoring high-quality images from blurred videos is a highly challenging task, especially in severely blurred scenes. In recent years, event-based methods have achieved significant progress in video deblurring. However, the modal differences between the event and image increase the difficulty of feature fusion. Additionally, the sparsity of event makes it difficult to restore some local details. To address these issues, we propose a new video deblurring method. Firstly, we design a cross modal collaborative attention mechanism to effectively fuse features from blurred frames and event frames, thereby deeply extracting motion information from event frames. Secondly, we utilize a diffusion model to generate spatial guiding prior feature, enhancing local details and textures. Furthermore, we propose an event-guided dynamic feature fusion module that adaptively integrates spatio-temporal information from neigh boring frames. Experimental results on both synthetic and real datasets demonstrate that our method outperforms the current state-of-the-art approaches. The code is available at: https://github.com/Frank-Zhou-01/EDVD-main

Multimodal Emotion Recognition in Conversations (MERC) aims to identify the emotional states expressed in utterances by modeling the complex interactions of textual, acoustic, and visual modalities within dynamic dialogue contexts. While recent advances leverage graph neural networks and contrastive learning to improve semantic alignment and representation robustness, they often suffer from rigid cross-modal alignment, which suppresses modality-specific nuances and degrades performance in dynamic or semantically ambiguous settings with noise. To address these limitations, we propose TriEMO, a novel Triple Semantic Alignment framework based on Modality Pair Contrastive Learning Graph Network for MERC. TriEMO introduces three key modules to enhance cross-modal representation and alignment: Modality Pair Data Augmentation (MPDA), which constructs contrastive learning units based on modality pairs to enrich pairwise semantic interactions; Rigid Modality Pair-based Contrastive Learning (RMPCL) combined with Modality Pair Hypergraph Learning (MPHL) and a Virtual Emotional Modality (VEM), which serves as a semantic anchor to coordinate information transmission both within temporal segments and across modality pairs, ensuring robust semantic alignment; and Triple Adversarial Contrastive Learning (TACL), which mitigates over-alignment in deep graph propagation by pulling modality pairs closer while pushing single-instance modality pairs away, thereby preserving discriminative features. Finally, a lightweight linear classifier predicts the emotion for each utterance. Extensive experiments on the IEMOCAP and MELD benchmarks demonstrate the effectiveness of TriEMO. On IEMOCAP, TriEMO outperforms state-of-the-art graph learning-based methods by 2.21% weighted F1 and contrastive learning-based methods by 0.69%, with a notable improvement of 2.13% on the Neutral category. On MELD, it achieves 3.18% weighted F1 improvement over the strongest graph learning baselines and 0.97% over contrastive learning approaches, with a 2.45% gain on the challenging Disgust emotion. These consistent improvements highlight TriEMO’s superior robustness and generalization for multimodal emotion recognition in conversational settings.