Multiscale Model Research Articles

Background: The cardiotoxicity of drugs, which often leads to fatal arrhythmias such as torsade de pointes (TdP), is a major cause of drug recalls. The current detection of human ether-a-go-go-related gene (hERG) potassium channels is excessively sensitive, creating an urgent need for an effective and rapid in vitro evaluation tool. Despite progress in population models of cardiomyocytes, there is a lack of direct evaluation with electrophysiological multi-scale population models. Methods and Results: We developed a multi-scale cardiac electrophysiological simulation platform to directly evaluate the cardiac toxicity of generic drugs. This platform consists of 1D cable, 2D tissue, and 3D organ scales. Initially, we stochastically generated parameters within ranges constrained by literature searches, such as ion channel conductance, and subsequently assessed pseudo-ECG morphology, myocardial cell membrane voltage, etc., to eliminate aberrant samples. We then incorporated isoprenaline (ISO) and paced at various cycle periods to evaluate the robustness of individual samples under different physiological conditions like exercise or sleep. After passing all these tests, a remaining cohort of over 400,000 individuals was identified, virtually representing the drug-free normal population. We utilized a dataset comprising 109 drugs for experimental purposes, considering the impact of these drugs on multiple ion channels. We used the ratio of individuals observed with premature ventricular contractions (PVCs) to those without PVCs as the metric to evaluate drug risk. To test the generality of this method, we also performed pipeline simulations with different cardiac models and diverse effective free therapeutic plasma concentrations (EFTPC). For example, at 1xEFTPC and 2xISO, the ToR-ORd model achieved an accuracy of 89.0%, a specificity of 93.1%, and a sensitivity of 90.2%. Further increases in EFTPC generally resulted in improved overall performance. We investigated and explained the outliers of prediction. Conclusions: Our platform presents a novel approach for detecting drug-induced cardiotoxicity. We established a robust population dataset for detecting drug toxicity via rigorous filtration. Although the cardiac model types, drug EFTPC, and specific IC50 values influence the result accuracy, our platform performance is sufficiently reliable to effectively identify high-risk drugs during early-stage development, thereby reducing costs and shortening timelines.

Achieving negative surgical margins is fundamental to curative colorectal cancer (CRC) surgery. Despite advancements in imaging, preoperative identification of margin risk remains limited. Recent developments in artificial intelligence (AI) now enable fusion of radiomics, quantitative imaging analysis and histopathology (“pathomics”) to predict microscopic tumor spread more accurately. Radiomics captures sub-visual textural and spatial features from CT and MRI, while AI-driven histopathology interprets digital slides at cellular resolution. Integrating these modalities yields a multi-scale model that reflects both macroscopic tumor architecture and microscopic invasiveness. Multicentric studies in China and the United States have demonstrated superior performance of radiopathomic models over single-modality approaches for predicting therapeutic response and margin status. As countries such as the United Kingdom and South Korea implement AI-driven precision oncology frameworks, transparent validation remains essential. By enabling more informed surgical planning and tailored resections, multimodal AI fusion could markedly enhance oncologic outcomes in CRC.

The real-time, precise monitoring of physiological signals such as intracranial pressure (ICP) and arterial blood pressure (BP) holds significant clinical importance. However, traditional methods like invasive ICP monitoring and invasive arterial blood pressure measurement present challenges including complex procedures, high infection risks, and difficulties in continuous measurement. Consequently, learning-based prediction utilizing observable signals (e.g., BP/pulse waves) has emerged as a crucial alternative approach. Existing models struggle to simultaneously capture multi-scale local features and long-range temporal dependencies, while their computational complexity remains prohibitively high for meeting real-time clinical demands. To address this, this paper proposes a physiological signal prediction method combining composite feature preprocessing with multiscale modeling. First, a seven-dimensional feature matrix is constructed based on physiological prior knowledge to enhance feature discriminative power and mitigate phase mismatch issues. Second, a network architecture CNN-LSTM-Attention (CBAnet), integrating multiscale convolutions, long short-term memory (LSTM), and attention mechanisms is designed to effectively capture both local waveform details and long-range temporal dependencies, thereby improving waveform prediction accuracy and temporal consistency. Experiments on GBIT-ABP, CHARIS, and our self-built PPG-HAF dataset show that CBAnet achieves competitive performance relative to bidirectional long short-term Memory (BiLSTM), convolutional neural network-long short-term memory network (CNN-LSTM), Transformer, and Wave-U-Net baselines across Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2). This study provides a promising, efficient approach for non-invasive, continuous physiological parameter prediction.

Tropospheric ozone (O3) is a harmful air pollutant and a short-lived greenhouse gas. To find effective O3 reduction strategies, it is essential to understand the sensitivity of O3 concentrations to its precursors, nitrogen oxides (NOx), and volatile organic compounds (VOC). This study applied the Community Multi-Scale Air Quality model (CMAQ) to assess the effects of domestic anthropogenic emissions in 2015 and 2050. The emission scenarios were based on Japan’s CO2 reduction targets, assuming an 80% decrease by 2050. Sensitivity analysis was performed by adjusting NOx and VOC emissions by ±10% and ±20%, respectively, and examining seasonal and regional variations in the O3 response. The results show that O3 levels will decrease notably in spring and summer by 2050, although concentrations will still exceed the standards in some areas. NOx reductions lead to significant O3 decreases, while VOC reductions show limited benefits, except in urban regions such as Kanto and Kansai. In winter, NOx reductions may even increase O3 levels due to weakened titration. Overall, the findings highlight the importance of prioritizing NOx control measures for effective O3 mitigation in Japan’s future energy transition.

Abstract We present a multiscale seismic attenuation tomography of a seismotectonically complex region in northern Italy hosting the well‐characterized Collalto Underground Gas Storage (UGS). Beyond its specific relevance, this site provides a natural laboratory for assessing the ability of attenuation imaging to distinguish fluid‐rich zones from highly strained, failure‐prone volumes. We integrated scattering and absorption tomography models: scattering anomalies, between the two principal thrusts, highlight localized strain near fault tips; absorption tomography images the shallow UGS and reveals a deeper fluid‐saturated volume. Seismicity concentrated around this deeper anomaly, exhibiting a pulsatory temporal pattern, suggests a fluid‐driven role in the deformation processes. These findings show that attenuation tomography, combined with multiscale and complementary geophysical models, can resolve critical subsurface features related to fluids and strain. The approach is broadly applicable to geothermal and volcanic contexts and supports seismic hazard assessment in tectonically active regions where natural and anthropogenic processes may interact.

To address the issue of insufficient resolution in remote sensing images due to limitations in sensors and transmission, this paper proposes a multi-scale feature fusion model, MSFANet, based on the Swin Transformer architecture for remote sensing image super-resolution reconstruction. The model comprises three main modules: shallow feature extraction, deep feature extraction, and high-quality image reconstruction. The deep feature extraction module innovatively introduces three core components: Feature Refinement Augmentation (FRA), Local Structure Optimization (LSO), and Residual Fusion Network (RFN), which effectively extract and adaptively aggregate multi-scale information from local to global levels. Experiments conducted on three public remote sensing datasets (RSSCN7, AID, and WHU-RS19) demonstrate that MSFANet outperforms state-of-the-art models (including HSENet and TransENet) across five evaluation metrics in ×2, ×3, and ×4 super-resolution tasks. Furthermore, MSFANet achieves superior reconstruction quality with reduced computational overhead, striking an optimal balance between efficiency and performance. This positions MSFANet as an effective solution for remote sensing image super-resolution applications.

Background: The Warburg effect, historically regarded as a hallmark of cancer metabolism, is often interpreted as a universal metabolic feature of tumor cells. However, accumulating experimental evidence challenges this paradigm, revealing a more nuanced and context-dependent metabolic landscape. Methods: In this study, we present a hybrid multiscale model of tumor metabolism that integrates cellular and environmental dynamics to explore the emergence of metabolic phenotypes under varying conditions of stress. Our model combines a reduced yet mechanistically informed description of intracellular metabolism with an agent-based framework that captures spatial and temporal heterogeneity across tumor tissue. Each cell is represented as an autonomous agent whose behavior is shaped by local concentrations of key diffusive species—oxygen, glucose, lactate, and protons—and governed by internal metabolic states, gene expression levels, and environmental feedback. Building on our previous work, we extend existing metabolic models to include the reversible transport of lactate and the regulatory role of acidity in glycolytic flux. Results: Simulations under different environmental perturbations—such as oxygen oscillations, acidic shocks, and glucose deprivation—demonstrate that the Warburg effect is neither universal nor static. Instead, metabolic phenotypes emerge dynamically from the interplay between a cell’s history and its local microenvironment, without requiring genetic alterations. Conclusions: Our findings suggest that tumor metabolic behavior is better understood as a continuum of adaptive states shaped by thermodynamic and enzymatic constraints. This systems-level perspective offers new insights into metabolic plasticity and may inform therapeutic strategies targeting the tumor microenvironment rather than intrinsic cellular properties alone.

Arteriovenous fistula (AVF) failure is a frequent clinical problem among end stage renal patients seeking durable long term dialysis access. The most common histological in vivo observation of AVF failure is endothelial injury at the juxta-anastomosis area (JAA) followed by thrombus deposition and subsequent neointimal hyperplasia (NH). While hemodynamic factors have been postulated to affect AVF remodeling and failure, the spatial correlations between changes in hemodynamics post AVF creation and in vivo physiologic observations remain poorly understood. In this work, we developed a novel computational fluid dynamics (CFD) model of an AVF using a pre-established aortocaval mouse model and integrated it with agent-based modeling for NH. The CFD simulation was performed using an animal-specific aortocaval fistula geometry derived from in vivo CTA images with prescribed boundary conditions obtained from in vivo ultrasound measurements. CFD results were validated against in vivo ultrasound velocity measurements at the level of the fistula. CFD allowed quantification of turbulence intensities throughout the fluid domain of the AVF. Turbulence was significantly elevated at the JAA and in regions of venous outflow stenosis. Turbulence intensity served as an input parameter for a simple two-rule agent-based model to test the hypothesis that non-homeostatic hemodynamic changes resulting from AVF creation drive spatial gradients in endothelial damage and proliferation of vascular smooth muscle cells (VSMC) leading to an increase in venous thickness or NH. Our findings show that increased velocity and turbulence in the JAA parallels in vivo NH formation, and that further from the JAA (both cranial and caudal) velocity and turbulence decrease incrementally. The results corroborate that perturbed hemodynamics in the JAA are potential triggers for NH and the source of thickness gradients observed in AVFs.

Handcrafted radiomics use predefined formulas to extract quantitative features from medical images, whereas deep neural networks learn de novo features through iterative training. We compared these approaches for predicting 3-month outcomes and hematoma expansion from admission non-contrast head CT in acute intracerebral hemorrhage (ICH). Training and cross-validation were performed using a multicenter trial cohort (n = 866), with external validation on a single-center dataset (n = 645). We trained multiscale U-shaped segmentation models for hematoma segmentation and extracted (i) radiomics from the segmented lesions and (ii) two latent deep feature sets—from the segmentation encoder and a generative autoencoder trained on dilated lesion patches. Features were reduced with unsupervised Non-Negative Matrix Factorization (NMF) to 128 per set and used—alone or in combination—for six machine-learning classifiers to predict 3-month clinical outcomes and (>3, >6, >9 mL) hematoma expansion thresholds. The addition of latent deep features to radiomics numerically increased model prediction performance for 3-month outcomes and hematoma expansion using Random Forest, XGBoost, Extra Trees, or Elastic Net classifiers; however, the improved accuracy only reached statistical significance in predicting >3 mL hematoma expansion. Clinically, these consistent but modest increases in prediction performance may improve risk stratification at the individual level. Nevertheless, the latent deep features show potential for extracting additional clinically relevant information from admission head CT for prognostication in hemorrhagic stroke.

Experimental and numerical investigation on low-velocity impact of plain-woven bamboo fiber reinforced epoxy resin composites based on multiscale modeling

Multiscale modeling and optimization of proton exchange membrane electrolysis cells: a review

Multiscale modeling and optimization of thermal barrier coatings for interface conditions

Carbon nanotube reinforced 3D printed PMMA filaments: Mechanical enhancement through experimental and multi-scale modeling

A multiscale finite element analysis model for functionally graded graphene reinforced composites considering graphene defects

Spatiotemporal dynamics of leachate transport in compacted clay liners: Coupled effects of dynamic slip and microbial succession mediated by multiphase interfaces.

Temporal heterogeneity in traffic crash delays: causal inference from multi-scale time factors and sample-wise structural decomposition.

Attention mechanism and multi-scale optimization-based image segmentation model in intelligent driving by transformer-DeepLabV3+

A SCA-based concurrent multiscale thermo-mechanical model for transient thermal ablative and mechanical damage properties of SiFPRCs

MSSCR: Multi-scale semantic collaborative reasoning model for explainable multimodal rumor detection

In silico multiscale computational modelling of botulinum toxin a diffusion for glabellar wrinkle treatment: Optimizing injection volumes across formulations.