Hong Kong University of Science and Technology
Hong Kong University of Science and Technology Institutional RepositoryNot a member yet
162821 research outputs found
Sort by
Tumor microenvironment-responsive NanoShield for precision photodynamic and photothermal synergistic cancer therapy with mitigated skin phototoxicity
No full textPhotodynamic therapy (PDT) is a highly promising non-invasive cancer treatment but is often hampered by severe skin phototoxicity arising from the nonspecific distribution of photosensitizers (PSs). Stimuli-responsive aggregation-induced emission (AIE) PSs show great promise for enhancing selectivity and efficacy. However, challenges such as complex synthesis and inadequate in vivo phototoxicity evaluations still need to be addressed. Furthermore, single-modality PDT frequently yields suboptimal therapeutic outcomes. To address these limitations, we present a tumor microenvironment (TME)-responsive AIE NanoShield (P-AD@PD), a precision theranostic nanoplatform engineered for synergistic phototherapy with mitigated skin phototoxicity. The NanoShield comprises a poly(N-isopropylacrylamide-co-acrylic acid) (PNA) nanogel system encapsulating an AIE PS (AD), shielded by a polydopamine (PDA) coating to enhance photoprotection. During systemic circulation, the PDA layer quenches AD photoactivity, minimizing skin phototoxicity. Upon tumor accumulation, it triggers hyperthermia for photothermal therapy (PTT) and gradually degrades within the TME to reactivate AD for precise PDT. This sequential PTT-PDT regimen amplifies therapeutic efficacy through dual-mode imaging while mitigating off-target toxicity. The PDA-based shielding strategy offers broad applicability across PSs, providing a universal approach to enhance PDT efficacy and safety.</p
Toward Medical Deepfake Detection: A Comprehensive Dataset and Novel Method
No full textThe rapid advancement of generative AI in medical imaging has introduced both significant opportunities and serious challenges, especially the risk that fake medical images could undermine healthcare systems. These synthetic images pose serious risks, such as diagnostic deception, financial fraud, and misinformation. However, research on medical forensics to counter these threats remains limited, and there is a critical lack of comprehensive datasets specifically tailored for this field. Additionally, existing media forensic methods, which are primarily designed for natural or facial images, are inadequate for capturing the distinct characteristics and subtle artifacts of AI-generated medical images. To tackle these challenges, we introduce MedForensics, a large-scale medical forensics dataset encompassing six medical modalities and twelve state-of-the-art medical generative models. We also propose DSKI, a novel Dual-Stage Knowledge Infusing detector that constructs a vision-language feature space tailored for the detection of AI-generated medical images. DSKI comprises two core components: 1) a cross-domain fine-trace adapter (CDFA) for extracting subtle forgery clues from both spatial and noise domains during training, and 2) a medical forensic retrieval module (MFRM) that boosts detection accuracy through few-shot retrieval during testing. Experimental results demonstrate that DSKI significantly outperforms both existing methods and human experts, achieving superior accuracy across multiple medical modalities.</p
Surgical-MambaLLM: Mamba2-Enhanced Multimodal Large Language Model for VQLA in Robotic Surgery
No full textIn recent years, Visual Question Localized-Answering in robotic surgery (Surgical-VQLA) has gained significant attention for its potential to assist medical students and junior doctors in understanding surgical scenes. Recently, the rapid development of Large Language Models (LLMs) has provided more promising solutions for this task. However, current methods struggle to establish complex dependencies between text and visual details, and have difficulty perceiving the spatial information of surgical scenes. To address these challenges, we propose a novel method, Surgical-MambaLLM, which is the first to combine Mamba2 with LLM in the surgical domain, that leverages Mamba2’s ability to effectively capture cross-modal dependencies and perceive spatial information in surgical scenes, thereby enhancing the LLMs’ understanding of surgical images. Specifically, we propose the Cross-modal Bidirectional Mamba2 Integration (CBMI) module to leverage Mamba2 for effective multimodal fusion, with its cross-modal integration capabilities. Additionally, tailored to the geometric characteristics of surgical scenes, we design the Surgical Instrument Perception (SIP) scanning mode for Mamba2 to scan the surgical images, enhancing the model’s spatial understanding of the surgical scene. Extensive experiments demonstrate that our Surgical-MambaLLM model outperforms the state-of-the-art methods on the EndoVis17-VQLA and EndoVis18-VQLA datasets, significantly improving the performance of the Surgical-VQLA task.</p
Optimal battery sizing using stochastic programming to consider building load variation and peak demand charge
No full textDemand charges and Time-of-use pricing are fundamental elements of contemporary electricity markets, introducing complexities in the operation of microgrids. Time-of-use pricing incentivizes energy consumption during off-peak hours, while demand charges impose fees based on peak power usage, significantly impacting electricity costs for both residential and commercial users. This research investigates the potential of battery energy storage systems to mitigate these costs by reducing demand charges and facilitating energy arbitrage. A significant challenge in determining optimal battery size lies in the uncertainties associated with building load predictions. Therefore, the study addresses critical uncertainties in load forecasts driven by climate change and occupant behavior. A novel stochastic framework is proposed that integrates these uncertainties into building load forecasts and considers demand charges in the optimization process. By employing the K-medoids clustering method in conjunction with the Bayesian information criterion, the framework achieves a remarkable reduction in computation time of 75.4% to 87.4%, while preserving essential load variability. The stochastic framework results in an overall cost reduction of 5.7%, alongside a 13.3% increase in the optimal battery size. Furthermore, implementing the proposed framework leads to a peak demand reduction of up to 25.8%.</p
GeoDualSPHysics: a high-performance SPH solver for large deformation modelling of geomaterials with two-way coupling to multi-body systems
No full textThis paper presents GeoDualSPHysics, an open-source, graphics processing unit (GPU)-accelerated smoothed particle hydrodynamics (SPH) solver designed for simulating large-deformation geomaterial and their interactions with multi-body systems. Built upon the popular open-source SPH solver DualSPHysics, the solver leverages its highly parallelised SPH scheme empowered by the CUDA parallelisation while extending its capabilities to large-deformation geomechanics problems with particles up to the order of 10⁸ on a single GPU. The SPH geomechanics model is enhanced by a noise-free stress treatment technique that stabilizes and accurately resolves stress fields, as well as an extended modified Dynamic Boundary Condition (mDBC) ensuring first-order consistency in solid boundary modelling. Additionally, the coupling interface between DualSPHysics and the multi-body dynamics solver Project Chrono is adapted for simulating interactions between geomaterials and multiple interacting rigid bodies. Benchmark validations confirm the solver's accuracy in resolving geotechnical failures, impact forces on solid boundaries, and geomaterial-multibody system interactions. GPU profiling of the newly implemented CUDA kernels demonstrates their performance metrics are similar to those of the original DualSPHysics solver. Performance evaluations demonstrate its saving in memory usage of 30-50% and improvements in computational efficiency over existing SPH geomechanics solvers, achieving practical simulation speeds for systems with tens of millions of particles and showing a speedup of up to 180x compared to the optimised multi-core CPU implementation. These advances position GeoDualSPHysics as a versatile, efficient tool for high-fidelity simulations of complex geotechnical systems. Program summary: Program title: GeoDualSPHysics CPC Library link to program files: https://doi.org/10.17632/z4sh62y97g.1 Licensing provisions: GNU Lesser General Public License Programming language: C++ and CUDA Nature of problem: Simulating large deformations in geomaterials and their interactions with movable or fixed solid bodies is critical for addressing engineering challenges such as landslides, soil-machine interactions, and off-road vehicle mobility. While the Smoothed Particle Hydrodynamics (SPH) method is well-suited for modelling continuum-based geomaterial behaviour in these scenarios, critical computational barriers persist, including: (1) numerical instabilities and unphysical noise in large-deformation regimes, (2) inefficiency in scaling simulations to millions of particles for real-world systems, and (3) inadequate frameworks for robust, two-way coupling between deformable geomaterials and multi-body systems. Overcoming these limitations demands stabilized SPH formulations, high-performance computing architectures, and two-way coupling with multibody dynamics solvers. Solution method: The GeoDualSPHysics solver addresses the above challenges by combining (1) a stabilised SPH formulation for geomaterials, featuring a noise-free stress treatment to eliminate spurious oscillations in large deformations and an extended modified Dynamic Boundary Condition (mDBC) for first-order consistent solid boundary modelling; (2) high-performance CUDA-based GPU parallelization inherited from DualSPHysics, enabling efficient simulations of tens of millions of particles; and (3) two-way coupling with Project Chrono via the DSPHChronoLib library, which integrates collision detection, frictional contact models, and joint constraints to resolve interactions between deformable geomaterials and multi-body systems.</p
Energy recovery from corn straw-based biochar@MIL-88A(Fe)-mediated anaerobic digestion of waste activated sludge under norfloxacin: Metabolism and antibiotic resistance gene fates
No full textNorfloxacin (NOR), a commonly detected antibiotic in waste activated sludge (WAS), remains understudied in anaerobic digestion (AD). This study investigated the effect of NOR on WAS AD, with corn straw-based biochar modified with MIL-88A(Fe) (BM) added to enhance energy recovery during digestion. Accumulated methane production was inhibited by 41.86 % in the BM-mediated digestion system under 1 mg/L NOR. Moreover, NOR induced the build-up of volatile fatty acids (VFAs), hindering methanogenic pathways subsequently. Microbial community structure was altered, with an enrichment of bacteria responsible for NOR degradation and a 13.20 % reduction in the abundance of hydrogenotrophic methanogens under antibiotic stress. Methanogenesis was inhibited with the expression of related genes and enzymes suppressed. The high enzymatic activities of cytochrome P-450 (CYP450) and acetate kinase contributed to the high NOR biodegradation efficiency (88.79 %). Twelve typical antibiotic resistant genes (ARGs) types, including multidrug, aminoglycoside, macrolides (MLs), etc., were examined in the AD system. The total abundance of ARGs type and subtype increased under NOR addition, implying ARGs removal was inhibited by NOR stress. Resistance to NOR exposure was primarily associated with antibiotic efflux and alterations in antibiotic target. Horizontal gene transfer (HGT) and vertical gene transfer (VGT) were the mechanistic routes for ARG evolution, with HGT inhibited and VGT promoted following NOR addition. The dominant genus Acinetobacter was the potential host for nearly all ARGs. This study advanced understanding of the impact of NOR on WAS digestion with BM mediation, providing new insights for optimizing WAS digestion.</p
Splurging with Alexa: How voicebots increase product upgrades
No full textDo AI-enabled voicebots, such as Amazon Alexa and Google Assistant, influence consumer choice, and if so, how and why? We demonstrate that consumers are more likely to choose expensive upgrades over more basic options when shopping using voicebots, compared to screen-based or text-based online shopping interfaces. Eleven studies using real voicebots, including secondary data from an online vendor, suggest that this tendency to upgrade arises from the cognitive demands of interacting with voicebots, which compromise processing of cost information while not influencing processing of benefit information. This effect attenuates for voice interactions with humans, suggesting a potential boundary condition. The findings make important contributions in the emerging field of consumer-AI interaction.</p
A well-balanced gas-kinetic scheme with adaptive mesh refinement for shallow water equations
No full textThis paper presents the development of a well-balanced gas-kinetic scheme (GKS) with space-time adaptive mesh refinement (STAMR) for the shallow water equations (SWE). While well-balanced GKS have been established on Cartesian and triangular meshes, the proposed STAMR framework utilizes arbitrary quadrilateral meshes with hanging nodes, introducing additional challenges for maintaining well-balanced properties. In addition to spatial adaptivity, temporal adaptivity is incorporated by assigning adaptive time steps to cells at different refinement levels, further enhancing computational efficiency. Furthermore, the numerical flux in the GKS adaptively transitions between equilibrium fluxes for smooth flows and non-equilibrium fluxes for discontinuities, providing the proposed GKS-based STAMR method with strong robustness, high accuracy, and high resolution. Standard benchmark tests and real-world case studies validate the effectiveness of the GKS-based STAMR and demonstrate its potential for interface capturing and the simulation of complex flows.</p
Cooling photovoltaic surfaces with vertical or rooftop greenery: a review of mechanisms, key factors, methods and future research trends
No full textBuilding-integrated photovoltaic (PV) systems and greenery are effective strategies for enhancing energy efficiency and ecological value in building façades and rooftops. Integrating PV with greenery not only provides environmental and urban-scale benefits but also helps reduce PV surface temperatures, mitigating efficiency losses caused by overheating. This review summarizes the cooling effects and efficiency improvements of two common systems: PV-green roofs (PV-GR) and façade-integrated PV-vertical greenery (FIPV-VG). Key influencing factors such as distance between PV and greenery, plant species, and climate zones are discussed. Experimental and simulation-based research methods are also outlined. Plant species used in previous literature and potential plant selection with high evapotranspiration rate are summarized and recommend, respectively. This article is a comprehensive critical literature review that synthesizes existing studies and does not collect or analyze primary data. FIPV-VG systems show surface temperature reductions of 1.65 °C to 4 °C and relative efficiency increases of 0.4 % to 3 %. PV-GR systems exhibit broader ranges, with cooling from 1 °C to 11 °C and relative efficiency gains between 0.08 % and 18 %. The temperate oceanic (Cfb) and tropical rainforest (Af) zones demonstrate the highest PV yield enhancements, ranging from 0.4 % to 17.1 % and 0.1 % to 8.6 %, respectively. Sedum is the most common used species in PV-GR while Thunbergia grandiflora is recommended for FIPV-VG due to its strong evapotranspiration capacity (6 L/day/m2) and vigorous growth. The optimal distance between PV panels and greenery requires further in-depth investigation, as it significantly influences both convective and latent heat transfer.</p
NIR-driven electron transfer for in situ gelation and enhanced hydrogen therapy
No full textHydrogen therapy has shown significant promise in improving wound healing by mitigating oxidative stress and inflammation. However, its therapeutic efficacy is constrained by limited delivery methods and insufficient bioavailability of hydrogen at wound sites. Herein, we design NIR-light triggered in situ gelation platform comprising ternary polymer dots as the photocatalyst, ascorbic acid as an electron mediator, and poly(ethylene glycol) diacrylate (PEGDA) as the polymeric matrix. The blended components of ternary Pdots enable the extended light absorption and cascading energy level alignment, leading to a marked increase in hydrogen generation compared to binary Pdots. Following local injection of the mixed precursor solution at the wound site and subsequent 700 nm light exposure, the in-situ gelation of PEGDA is initiated by ascorbate free radicals, obviating the need for commercial photoinitiators. The resulting hybrid hydrogel retains water content and photocatalysts, enabling prolonged hydrogen evolution under NIR light. The hydrogen produced by the catalytic action of the ternary Pdots effectively scavenges reactive oxygen species at the wound site and promotes macrophage M1-to-M2 phenotype transition. The immunomodulatory effects of this light-triggered platform demonstrate significant therapeutic potential, accelerating wound repair through enhanced hydrogen delivery strategy.</p