Reduced brain entropy in migraine with partial restoration during attacks: A resting-state fMRI study.

Brain imaging as a prognostic biomarker in urea cycle disorders.

Artificial intelligence (AI) has the potential of reshaping GI oncology by enabling more nuanced interpretation of complex clinical, imaging, and molecular data, while supporting more timely and patient-centered decisions. This article synthesizes perspectives across the GI cancer continuum, beginning with a framework for context-aware AI that emphasizes metadata, multimodal integration, and lifecycle quality and safety as foundations for trustworthy tools that clarify, rather than conceal, uncertainty. Next, AI in endoscopy is highlighted as an example in clinic practice, focusing on computer-aided detection and diagnosis systems that not only increase adenoma detection rates but also raise questions about surveillance burden, real-world effectiveness, and the balance between skill enhancement and potential deskilling of endoscopists. Another section explores how AI can help GI oncologists design, prioritize, and implement highly innovative clinical trials-particularly multi-omic and imaging-driven approaches-while envisioning a future in which far more patients participate in trials that align with their goals and values. The final section reviews emerging AI-enabled clinical trial matching pipelines, including large language model-based retrieval and prescreening tools that operate on real-world electronic health record and protocol data, and discusses challenges related to bias, privacy, explainability, and workflow integration. Together, these contributions argue that the greatest impact of AI in GI oncology will come from deliberately aligning technical capabilities with highly relevant patient-centered clinical questions, ethical governance, and implementation strategies that expand access to trials and improve outcomes for patients with GI malignancies.

Multi-modal integration with adversarial mutual distribution matching

Fake news detection method based on multimodal fusion and generative adversarial network

Multimodal synergistic effects and theranostic integration of hafnium-based nanoradiosensitizers for enhancing precision radiotherapy.

Artificial intelligence for CO2 pipeline monitoring: Cross-domain insights from oil, gas, water, and hydrogen systems

Positron emission tomography (PET) is a widely recognized technique for diagnosing neurodegenerative diseases, offering critical functional insights. However, its high costs and radiation exposure hinder its widespread use. In contrast, magnetic resonance imaging (MRI) does not involve such limitations. While MRI also detects neurodegenerative changes, it is less sensitive for diagnosis compared to PET. To overcome such limitations, one approach is to generate synthetic PET from MRI. Recent advances in generative models have paved the way for cross-modality medical image translation; however, existing methods largely emphasize structural preservation while neglecting the critical need for pathology awareness. To address this gap, we propose PASTA, a novel image translation framework built on conditional diffusion models with enhanced pathology awareness. PASTA surpasses state-of-the-art methods by preserving both structural and pathological details through its highly interactive dual-arm architecture and multi-modal condition integration. Additionally, we introduce a novel cycle exchange consistency and volumetric generation strategy that significantly enhances PASTA's ability to produce high-quality 3D PET images. Our qualitative and quantitative results demonstrate the high quality and pathology awareness of the synthesized PET scans. For Alzheimer's diagnosis, the performance of these synthesized scans improves over MRI by 4%, almost reaching the performance of actual PET. Our code is available at https://github.com/ai-med/PASTA.

Acute Coronary Syndromes (ACS), including ST- and non-ST-segment elevation myocardial infarction (STEMI, NSTEMI), remain a leading cause of global mortality. Traditional Cardiovascular Risk Scores (CVRS) provide important insights but mainly rely on clinical data, often neglecting environmental factors (e.g.air pollution, climate) that significantly influence cardiovascular health. Integrating complex time-series environmental and clinical datasets also presents substantial challenges. We propose TabulaTime, a multimodal deep learning framework integrating clinical risk factors with environmental data to enhance ACS risk prediction. TabulaTime delivers three innovations: multimodal integration of time-series environmental and clinical data; PatchRWKV for extracting complex temporal patterns with linear computational complexity; and enhanced interpretability through attention mechanisms. TabulaTime improves prediction accuracy by 20.5% over CatBoost, with environmental data contributing a 10.1% gain. PatchRWKV outperforms state-of-the-art models (MLP-, CNN-, RNN- and Transformer-based models). Feature analysis highlights key clinical and environmental predictors. This approach advances personalised prevention and strengthens public health against cardiovascular risks.

Mapping knowledge structure and emerging trends in non-invasive brain-computer interface for stroke rehabilitation.

Multimodal neurocognitive assessment of internet gaming disorder using ERP and fNIRS during an oddball paradigm: A machine learning-based classification.

Evaluating multimodal integration of metro and feeder bus services in Bengaluru using integrated analytical and perception based methods

Foundation models (FMs), which are large-scale architectures pretrained on diverse datasets, are rapidly reshaping artificial intelligence (AI). In medical imaging, these models provide unified methods for learning rich visual and multimodal representations, facilitating accurate diagnoses, efficient clinical workflows, and equitable access to healthcare. This review surveys the FMs in medical imaging and summarizes key insights for future research. It situates medical FMs in the broader context of generalist AI, categorizing them from vision-derived adaptations to modality-specific and emerging general-purpose systems. This review systematically compares core architectures, pre-training objectives, and clinical performance, highlighting improvements in adaptability, robustness, and data efficiency. Persistent challenges, such as limited data availability, privacy regulations, interpretability issues, and real-world deployment constraints, were examined alongside practical solutions, including self-supervised learning, federated training, and parameter-efficient fine-tuning. Future directions discussed include multimodal integration, lightweight inference suitable for edge devices, and rigorous validation that adheres to regulatory standards. By consolidating the current knowledge and identifying open research questions, this review offers clear guidance for researchers and clinicians aiming to integrate FMs into routine medical practice, laying the groundwork for subsequent detailed explorations. • Foundation models improves generalization in medical imaging tasks. • Vision-, modality-, and general-purpose models are systematically compared. • FMs mitigate data scarcity via self-supervised and federated learning. • Clinical applications include diagnosis, prognosis, and workflow automation. • Future work includes multimodal fusion, deployment, and validation.

Deep learning classifies medicinal plants, driven by the need to preserve traditional knowledge and automate identification for practical uses. This review extensively summarizes 30 recent studies (2021–June 2025) on applying deep learning, primarily using image data, to classify medicinal plants. This review analyzes research distribution, dataset preparation, image preprocessing, augmentation, and deep learning architectures like convolutional neural networks, Vision Transformers, and hybrid models. Our analysis reveals a strong geographic focus, with 50% of the selected studies originating from India and Bangladesh. The focus is overwhelmingly on leaf imagery, with 29 out of the 30 studies relying on this approach. The field is also characterized by its dependence on existing data, as 56.6% of studies utilized public datasets and another 26.6% employed a hybrid of public and private data, with dataset sizes ranging from a minimum of 637 to a maximum of 13,500 images. Methodologically, the vast majority of studies rely on a transfer learning approach (36.7%), achieving robust accuracy rates between 74% and 99.9%. Furthermore, we recognize significant limitations, such as the absence of standardized and diverse datasets, insufficient inclusion of uncommon or endangered species, and inadequate representation of whole-plant imaging. The research underscores the necessity for collaborative, multidisciplinary initiatives to develop centralized, high-quality, and geographically comprehensive datasets. We delineate prospective avenues, including multimodal feature integration, the development of real-world applications, and optimization for privacy-preserving frameworks such as federated learning. This study guides academics advancing deep learning for medicinal plant classification and biodiversity conservation. Received: 13 September 2025 | Revised: 8 December 2025 | Accepted: 5 March 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Monir Hossain: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision. Fahmid Al Farid: Validation, Formal analysis, Investigation, Resources, Writing – review & editing, Visualization. Momotaz Begum: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – review & editing, Visualization. Jia Uddin: Conceptualization, Validation, Formal analysis, Investigation, Resources, Writing – review & editing, Visualization, Supervision, Project administration. Hezerul Bin Abdul Karim: Validation, Formal analysis, Investigation, Resources, Writing – review & editing, Visualization, Funding acquisition.

Adult diffuse gliomas exhibit marked heterogeneity, making comprehensive evaluation of their biological behavior difficult. This study aimed to develop and validate a multimodal machine learning framework that fuses MRI-based radiomic features, whole-slide image (WSI)-derived pathomic signatures, and clinical variables for the comprehensive assessment of adult diffuse gliomas. Radiomic features were extracted from multiparametric MRI, and pathomic features from hematoxylin-eosin (H&E) stained WSIs; both were combined with clinical variables (age, sex, tumor dimensions, anatomical location). Unimodal radiomic/pathomic models and multimodal integrated models were built using three machine learning algorithms: random forest (RF), support vector machine (SVM), and eXtreme Gradient Boosting (XGBoost). A retrospective cohort of 373 pathologically confirmed patients was used for model construction and internal validation, with an independent external cohort of 49 patients for external validation. 40 radiomic and 20 pathomic features were retained via statistical testing and RF-based feature ranking. The optimal unimodal radiomic and pathomic models achieved internal validation AUCs of 0.89 and 0.83, respectively. The multimodal model showed superior performance (internal AUC = 0.90) and stable generalizability in external validation (AUC = 0.93). The multimodal model achieved AUC 0.93 with higher balanced accuracy and sensitivity. DeLong tests showed no statistically significant difference between multimodal and radiomics (P = 0.377) or pathomics (P = 0.085) in internal test; however, the multimodal model showed clinically meaningful improvements in sensitivity and balanced accuracy. The developed and validated multimodal machine learning model integrating radiomics, pathomics, and clinical information exhibits stable and reliable performance for the grading assessment of adult diffuse gliomas.

The identification of potential associations between metabolites and diseases is crucial for understanding the onset and progression of diseases. Although numerous methods have been developed to predict metabolite-disease associations (MDAs), few methods fully mine the complementary information between different similarity relation or learn higher-order relationships between metabolites or diseases. To address these challenges, we propose a method for MDAs prediction with Progressive Orthogonal Multimodal Similarity Learning (POMSL). POMSL first constructs two hypergraphs within single-modal similarity with KNN and K-means to capture higher-order complex relationships. Then, hierarchical contrastive learning is applied to enhance the consistency of multi-similarity features by performing contrastive learning for inter-similarity and intra-similarity views. Next, a progressive orthogonal multimodal similarity integration strategy is developed to ensure the effective fusion of cross-similarity complementary information and to enhance the divergence of multi-similarity features. Finally, a bilinear decoder is used to predict MDAs. Extensive experimental results demonstrate that POMSL performs excellently in MDA prediction and, more importantly, validate the effectiveness of hierarchical contrastive learning and progressive orthogonal multimodal similarity learning. The source code and dataset are available at https://github.com/1521250466LYP/POMSL.

Spatial multi-omics techniques provide powerful tools to decipher tissue architectures in multilayer perspectives, including gene expressions, gene regulations and microenvironments. The multimodal data integration is a critical step in spatial multi-omics data processing involved in many downstream analyses, such as spatial domain clustering. Thus, it is important to learn reliable latent representation by efficiently integrating the spatial multi-omics for downstream analyses. In this work, we developed a Spatially Multimodal and Multiscale Network (SpaMM-Net) to learn the latent representations from spatial multi-omics data. Due to the intrinsic noises and the complex relationship among multiple omics, SpaMM-Net utilized spatially-guided multiscale graph attention networks to integrate the multimodal omics features at different spatial-scale levels for representation learning. We evaluated our method for downstream spatial clustering task on several spatial multi-omics datasets. The results show that SpaMM-Net achieves good performance in deciphering both detailed regions and tissue architectures. Our scale-wise weight analyses further reveal that the SpaMM-Net effectively leverages multiscale spatial information to enhance the robustness of the learned representations. SpaMM-Net is an efficient tool to capture and integrate latent representations from spatial multi-omics for spatial domain identification. This strategy can be extended to other multi-omics modalities with the rapid development of spatial multi-omics technologies.

Atypical neural synchronization in the temporal gyrus during face processing in children with autism spectrum disorder.

Neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS), Alzheimer's disease (AD), Parkinson's disease (PD) and Huntington's disease (HD) cause progressive loss of specific neuronal populations and currently lack curative therapies. Animal models and immortalized cell lines incompletely recapitulate human pathology and genetic heterogeneity, limiting drug discovery. Human induced pluripotent stem cells (iPSCs) provide a patient‑specific platform for disease modelling, drug screening and studying individual responses. Translational research (TR) uses iPSC models to identify candidate therapies that are subsequently tested in clinical trials, while reverse translational research (rTR) feeds clinical observations back to the bench by analyzing iPSCs derived from trial participants and integrating molecular data with patient phenotypes. This review summarizes recent advances in iPSC‑based TR and rTR for ALS and extends the discussion to other neurodegenerative diseases. Key clinical trials launched from iPSC screens-ropinirole, retigabine and bosutinib-are reviewed alongside emerging rTR efforts that use patient‑derived iPSCs to identify biomarkers and therapeutic mechanisms. We also survey iPSC models for AD, PD and HD, highlighting applications of three‑dimensional (3D) brain organoids and gene‑editing technologies. Finally, we discuss future directions for precision medicine, multimodal integration and technological challenges, with particular attention to how imaging biomarkers may complement iPSC-based TR/rTR frameworks in neurodegenerative diseases.

Value learning is often studied at the individual object level, yet in natural environments, values are learned within structured contexts. A dominant account-value normalization-proposes that item values are encoded relative to their local context, producing contrastive biases that suppress below-average values and amplify above-average ones. However, empirical support for value normalization has been limited to proximity-based grouping (i.e., grouping by spatial or temporal co-occurrence), leaving open how value learning operates in more structured environments. Here, we focus on structure-based grouping, in which items are related by conceptual or relational structure and test value generalization as a competing account of context-dependent value learning. Across five experiments using modified multiarmed bandit tasks, participants learned item values through trial and error, while group structure had to be inferred from experience. Contrary to the predictions of value normalization, participants systematically overestimated the value of below-average items in high-mean groups and preferred them over objectively superior items from lower mean groups. This assimilative bias was robust across groupings defined by visual features, task-defined categories, and abstract cognitive map structure and persisted even when explicit monetary reward framing was removed. Computational modeling showed that behavior was best captured by value generalization mechanisms, implemented either as feedback-driven value propagation within groups or as Bayesian integration of group-level and item-level information at decision time. Across experiments, these models consistently outperformed normalization and standard item-level learning models. Together, these findings challenge the dominance of value normalization and identify value generalization as a cognitively efficient-though imperfect-strategy for value learning in structured environments. (PsycInfo Database Record (c) 2026 APA, all rights reserved).