Abstract: Hypoxia is a prevalent characteristic of neurological diseases, including ischemic injury, neurodegeneration and infectious disease complications. Concurrently, hypoxia shapes both protective and pathological responses within the central nervous system (CNS). Central to this process is hypoxia-inducible factor 1α (HIF1α), a transcription factor that regu-lates cellular adaptation to reduced oxygen availability through coordinated glycolytic, inflammatory and cell survival pathways. Under hypoxic conditions, HIF1α transcriptional activity influences microglial activation, mitochondrial quality control, and cytokine production, thereby modulating neuroinflammation and neuroprotection. Preclinical evidence points toward hypoxia preconditioning being neuroprotective through HIF1α-dependent mechanisms in a context-dependent matter. This review synthesizes the current understanding of the role of HIF1α across neurological disease contexts, highlighting the intersection of hypoxia, neuroinflammation and neuronal survival. Ultimately, defining the cell-specific and context-dependent involvement of HIF1α will be critical for targeted therapeutic approaches to alleviate neuronal death and slow disease progression.

Abstract: Let R be a commutative ring with a nonzero identity and M a nonzero unital R-module. We introduce the concepts of weakly n-submodules and weakly (1, n)-submodules as module-theoretic generalisations of the weakly n-ideal and weakly (1, n)-ideal. A proper submodule N of M is called a weakly n-submoduleif whenever 0 ̸= am ∈N for some a ∈R and m ∈M , then a ∈(N :R M )or m ∈Nil(M )·M , where Nil(M ) = annR(M ). Similarly, N is called a weakly (1, n)-submodule if whenever 0 ̸= abm ∈N for some nonunit elementsa, b ∈R and m ∈M , then ab ∈(N :R M ) or m ∈Nil(M )·M . Every weaklyn-submodule is a weakly (1, n)-submodule, and every weakly (1, n)-submoduleis weakly 1-absorbing primary. We provide a six-fold characterisation, provestructure theorems classifying the rings and modules over which every propersubmodule belongs to these classes in particular, we show that for faithfulnitely generated multiplication modules, every proper submodule is weakly (1, n) if and only if R is a UN-ring or a product of two elds and investigatebehaviour under homomorphisms, localisations, and quotient modules.

Abstract: Zinc deficiency is increasingly recognized as a risk factor for neurodegenerative diseases, yet the underlying molecular mechanisms remain incompletely understood. In this study, we investigated the impact of intracellular zinc depletion on oxidative stress and in-flammasome activation in microglial (SIM-A9) and neuronal (SH-SY5Y) cell models, and evaluated the protective effects of polyphenolic compounds. Intracellular zinc chelation with the membrane-permeable chelator TPEN markedly increased reactive oxygen species (ROS) production, reduced cell viability, and upregulated the mRNA expression of NLRP3 inflammasome–related genes and pro-inflammatory cytokines. In contrast, extracellular zinc chelation had no effect, highlighting the critical role of intracellular zinc homeostasis in maintaining redox balance. Zinc supplementation significantly attenuated these responses. Among 32 polyphenols screened by DPPH radical scavenging assay, caffeic acid derivatives—chicoric acid (ChA), rosmarinic acid (RA), and caffeic acid phenethyl ester (CAPE)—exhibited the most potent antioxidant activity, surpassing that of edaravone. These compounds suppressed ROS production and differentially protected against zinc deficiency–induced cellular damage. ChA showed the strongest ROS in-hibitory activity (IC50: 1.9 µM in SIM-A9), RA provided robust cytoprotection even at low concentrations, and CAPE most effectively suppressed inflammasome-related gene ex-pression and inhibited aggregation of both Aβ1–42 and the highly neurotoxic py-roglutamate-modified variant pEAβ3–42. These findings demonstrate that intracellular zinc deficiency drives ROS-dependent NLRP3 inflammasome activation, and suggest that caffeic acid derivative polyphenols may serve as complementary agents for mitigating neuroinflammatory and amyloidogenic processes relevant to Alzheimer's disease.

Abstract: Robots deployed in disaster environments—such as collapsed buildings, flooded tunnels, and conflict-damaged urban areas—must navigate without GPS, operate under degraded sensing conditions including dust, smoke, and darkness, and adapt rapidly to changing mission conditions. Most existing learning-based navigation approaches rely on a single policy, which often fails when the environment shifts in unexpected ways. This paper presents UC-MESL (Uncertainty-Conditioned MAP-Elites Skill Library), a framework that learns a diverse library of specialized navigation behaviors and dynamically switches between them in real time based on environmental uncertainty. Each skill is optimized for a specific operating condition and characterized by three interpretable traits: risk tolerance, exploration preference, and movement style. A lightweight selector uses live uncertainty estimates from the robot’s onboard map to choose the most appropriate skill during deployment. We evaluate UC-MESL across three simulated rescue scenarios—collapsed rubble, flooded tunnels, and war-damaged urban blocks—under four levels of sensor degradation and realistic communication outages. Compared with the strongest single-policy baseline, UC-MESL finds 18.4% more victims within the mission time budget, reaches the first victim 31.2% faster, reduces hazard exposure by 24.6%, and loses only 8.3% of performance under severe sensor noise, compared with 29.7% for a single-policy NEAT baseline. These results demonstrate that maintaining a diverse repertoire of specialized navigation skills, combined with uncertainty-aware skill selection, provides more robust and reliable autonomy for disaster-response robotics than optimizing a single general-purpose policy.

Abstract: The YOLO (You Only Look Once) object detection models have undergone rapid evolution, with each version introducing architectural enhancements aiming to improve speed, accuracy, and deployment. Simultaneously, Single-Board Computers (SBCs) have advanced to support increasingly complex AI models in edge environments. This study presents a comprehensive benchmarking of YOLO versions 8 through 12 across a range of SBCs, including Raspberry Pi4/5, NVIDIA Jetson Nano, Jetson Orin, and LattePanda, under different power modes. Key performance metrics, including inference speed (FPS), detection accuracy (mAP), RAM usage, and computational complexity (FLOPs), are evaluated. These findings offer practical insights for developers and researchers to select optimal YOLO variants and SBC configurations for real-time edge deployment.

Abstract: Background: Hantavirus pulmonary syndrome (HPS), also designated hantavirus cardiopulmonary syndrome, is caused by New World hantaviruses, principally Sin Nombre virus in North America and Andes virus in South America. The syndrome is characterized by rapidly progressive noncardiogenic pulmonary edema and myocardial depression, with case fatality rates of 30% to 50%. Methods: This review synthesizes peer-reviewed literature on the virological, pathophysiological, clinical, and therapeutic aspects of HPS, with emphasis on cardiopulmonary mechanisms. Sources were identified through PubMed, prioritizing original research, clinical series, and controlled trials published through 2025. Results: Pathogenic hantaviruses enter endothelial cells and platelets via αvβ3 integrins, disrupting the VEGF-VEGFR2 signaling axis and rendering endothelial cells hypersensitive to physiological VEGF concentrations. Expansion of CD8+ T cells and activated macrophages releases TNF-alpha, IFN-gamma, and nitric oxide, amplifying microvascular permeability and contributing to myocardial depression. Autopsy studies demonstrate direct hantaviral myocarditis with viral antigen in cardiac endothelium and interstitial macrophages. Transpulmonary thermodilution confirms simultaneous hypovolemia, reduced global ejection fraction, and elevated extravascular lung water. VA-ECMO initiated at the first signs of cardiopulmonary decompensation has reported survival rates approaching 80% in selected experienced centers. No antiviral has demonstrated efficacy in controlled trials during the cardiopulmonary phase. Conclusions: HPS produces a mixed shock state through increased microvascular permeability, T cell-mediated immunopathology, and direct myocarditis. Management follows a stepwise algorithm: suspected HPS triggers immediate complete blood count with peripheral blood smear and hantavirus IgM serology or RT-PCR, followed by ICU admission, conservative fluid resuscitation guided by transpulmonary thermodilution, and early contact with an ECMO-capable center at the first sign of rising lactate, falling cardiac index, refractory shock, arrhythmia, or rapid oxygenation failure.

Abstract: AGI alignment is often evaluated at a snapshot: a system is judged by its current outputs, policy profile, benchmark behavior, or apparent corrigibility. Snapshot evaluation misses a central risk of advanced deployment: a good endpoint can still be reached by a bad journey. Two trajectories may arrive in similar behavioral regions while differing in reversibility, opacity, intervention cost, memory entanglement, institutional dependency, and the quality of human judgment left available for oversight. This paper develops a path-sensitive alternative. It represents AGI development as motion through an augmented state space Z containing model and environment state, world-model structure, policy state, memory and provenance traces, governance affordances, institutional embedding, and human evaluative capacity. Cognitive integrity — the capacity of individuals, teams, or institutions to sustain calibrated attention, trust, contestability, and decision under pressure [1] — is introduced here as an alignment-relevant state variable rather than assumed as a familiar metric. The formal contribution is a scaffold of definitions: controlled transition laws over augmented state, escape cost, path-level alignment functionals, viability floors, forbidden regions, and trajectory classes distinguished by lock-in, basin structure, retargetability, and integrity preservation. The result does not supply a calibrated empirical model of deployed AGI systems. It specifies what such a model must track if alignment evidence is to cover both present behavior and the remaining possibility of legible, reversible, and cognitively intact correction.

Abstract: Introduction: Precision oncology has revolutionized cancer care in high-income countries, but its implementation in Latin American low-resource settings faces profound bioethical dilemmas. This study analyzes these challenges through the lens of social justice and equity. Methods: An integrative review was conducted following the Whittemore and Knafl framework. A systematic search was performed across PubMed, Scopus, SciELO, and LILACS (2015–2025). Thematic synthesis was applied to integrate empirical data with normative bioethical theories. Results: Four major analytical themes were identified: 1) The Innovation Paradox and Financial Toxicity, where prohibitive pricing (exceeding $100,000 USD/year) violates distributive justice and leads to a biological penalty in survival; 2) Infrastructure Deficits and Epistemic Injustice, highlighted by a 9.4% access rate to Next-Generation Sequencing (NGS) and the risks of applying Eurocentric genomic data to admixed LA populations; 3) Research Vulnerability, where clinical trials serve as survival strategies, compromising autonomy and informed consent; and 4) The Judicialization Dilemma, where individual court orders for high-cost drugs threaten systemic sustainability and equity. Conclusions: To prevent a genomic apartheid, Latin America must transition toward genomic sovereignty and frugal precision oncology. Bioethical frameworks in the region must prioritize protection ethics and social justice to ensure that scientific innovation does not exacerbate existing health inequities.

Abstract: Recent studies have investigated whether the rate of global warming has changed since the 1970s, with particular attention to the role of natural variability and its removal from temperature time series. In particular, Foster and Rahmstorf (2026) analyzed global mean surface temperature series, adjusted for natural variability. However, their procedure might produce spurious changepoints, since it does not appropriately handle the autocorrelation present in the residuals of the models considered. In this study, we revisit the same adjusted temperature series using a different methodology (the Quandt likelihood ratio test) while properly accounting for the presence of autocorrelation. We find evidence that global temperature has departed from its previous path since around 2013-2014. Our results provide a robust proof of a clear recent increase in the temperature trend, at a rate of warming that has doubled since that date.

Abstract: Chagas disease, caused by Trypanosoma cruzi, remains a major public health problem in Latin America and an emerging global health concern due to population mobility. Alt-hough benznidazole and nifurtimox remain the only approved antiparasitic drugs, their limited efficacy in chronic infection, prolonged treatment regimens, frequent adverse ef-fects, and variable activity across parasite strains highlight the need for new therapeutic strategies. In addition, the pathogenesis of chronic Chagas disease is driven not only by parasite persistence but also by immune-mediated tissue damage, particularly in chronic Chagas cardiomyopathy. In this review, we examine emerging therapeutic approaches that extend beyond conventional trypanocidal chemotherapy, with emphasis on plant-derived extracts, essential oils, antimicrobial peptides, and cell-based immuno-modulatory strategies. Plant compounds and essential oils have shown antiparasitic ac-tivity through mechanisms including oxidative stress induction, membrane disruption, interference with sterol biosynthesis, and mitochondrial dysfunction, while some extracts also modulate host immune responses. Antimicrobial peptides display dual potential by directly damaging parasite membranes and organelles or by reshaping infec-tion-associated inflammatory responses. In parallel, cell-based therapies such as mesen-chymal stromal cells, tolerogenic dendritic cells, and bone marrow-derived cells have demonstrated promising cardioprotective and immunoregulatory effects in experimental chronic Chagas disease. Collectively, these approaches support a multitarget therapeutic framework in which parasite-directed and host-directed interventions may complement each other. Further mechanistic studies, standardization, and translational validation will be essential to advance these candidates toward clinically useful therapies for Chagas disease.

Abstract: Generative artificial intelligence (GAI) is increasingly embedded in personal financial, yet little is known about how models make recommendations using financial information and demographic cues. This study audits three frontier GAI models, GPT 5.5, Gemini 3.1 Pro, and Claude Opus 4.7, using a full-profile conjoint experiment in which each model evaluated the same 1,000 hypothetical investor profiles and selected among standardized conservative, balanced, and aggressive portfolios. Investor profiles systematically varied attributes, including risk tolerance, time horizon, goal type, income, and age, gender, ethnicity, marital status, and employment type. Ordered logistic regressions and matched-profile comparisons show that all three models base recommendations primarily on legitimate financial inputs, especially risk tolerance and time horizon. Gender and ethnicity do not significantly influence recommendations, although age affects all models and marital status affects ChatGPT. However, the models are not interchangeable: they differ significantly in overall risk appetite and in how they translate risk tolerance, time horizon, goal type, and age into portfolio choices, with economically meaningful differences in predicted recommendations for identical clients. These findings suggest that contemporary GAI investment advice exhibits limited evidence of conventional demographic bias but introduces a distinct form of platform risk arising from model-specific advisory logic.

Abstract: Contrastive learning–based models such as DrugCLIP have recently emerged as scalable tools for structure-based virtual screening by embedding protein structures and small molecules into a shared representation space. While these approaches demonstrate high throughput and competitive screening performance in ligand retrieval tasks, their ability to correctly identify biologically relevant ligand-binding pockets has not been systematically evaluated. Here, we construct a benchmarking dataset comprising 42 pharmacologically diverse human protein targets with experimentally validated drug-bound structures spanning multiple target families. Using this dataset, we evaluate the pocket recognition capability of DrugCLIP and compare its performance with a traditional structure-based workflow (Fpocket combined with ESSA) and a machine learning-based method (P2Rank). DrugCLIP shows robust performance for well-characterized target classes, including kinases (10/10) and nuclear receptors (5/5), but exhibits markedly reduced accuracy for ion channels (1/4), GPCRs (3/5) and transporters (3/5). Notably, pocket prediction accuracy does not strongly correlate with structural data availability, suggesting that intrinsic pocket characteristics rather than training data abundance primarily affect model performance. Across the benchmark, DrugCLIP achieves an overall success rate of 71% (95% CI: 56-83%), compared with 79% (95% CI: 64-88%) for Fpocket+ESSA and 93% (95% CI: 81-98%) for P2Rank. McNemar’s test showed no significant difference between DrugCLIP and Fpocket+ESSA (p=0.508), whereas P2Rank significantly outperformed DrugCLIP (p=0.012). Together, these results provide a quantitative evaluation of pocket recognition by contrastive learning–based models and highlight key limitations of embedding-based approaches for pocket localization.

Abstract: Background: Marfan syndrome (MFS) is a connective tissue disorder caused by FBN1 mutations, leading to elastic fiber disarray and early thoracic aortic aneurysm (TAA) formation. Currently, pharmacological treatments lack specificity and only delay progression. We previously reported a specific TGFβ-driven miR-632 up-regulation in MFS TAA tissues and blood causing smooth muscle cell dedifferentiation and aortic wall degeneration. This study evaluated the effects of three conventional antihypertensive drugs (β-blocker, ACE inhibitor and sartan) on parietal remodeling comparing them with a miR-632 inhibitor in an ex vivo TGFβ –induced model of MFS TAA. Methods and Results: Gene expression and western blot analyses demonstrated that only losartan significantly reduced miR-632 and vascular degeneration markers. Notably, combined treatment with ramipril and carvediol compromised losartan’s efficacy, highlighting the need for careful therapeutic selection. miR-632 inhibitor was the most effective strategy in this ex vivo setting, although further preclinical validation is needed to confirm its therapeutic potential in vivo. Conclusions: Our data emphasize that choosing the right treatment in MFS aortopathy requires understanding its specific impact on cellular pathways. Our findings identify losartan as the most effective standard drug while suggesting miR-632 as a promising future target to stabilize the aortic wall and delay surgery.

Abstract: Alzheimer's disease (AD), a leading cause of dementia worldwide, is a neurological disorder characterized by progressive cognitive decline. AD is also considered a significant socioeconomic burden. While definitive diagnostic tools such as positron emission tomography (PET) imaging and cerebrospinal fluid (CSF) biomarker analysis offer high sensitivity and specificity, they are limited by high cost, invasiveness, and limited accessibility. Consequently, these gold standard approaches hinder their applicability for large-scale screening and longitudinal follow-up. Recent advances in blood-based biomarkers hold promise in capturing systemic molecular changes associated with AD. In particular, transcriptomic signatures derived from RNA sequencing (RNA-seq) are promising in capturing systemic molecular changes associated with AD. Gene expression profiles in peripheral blood reveal underlying pathological processes. These pathological processes can be listed as synaptic dysfunction, neuroinflammation, and metabolic dysregulation. Together with the high-dimensional datasets and AI approaches enable the identification of robust predictive models which has the assistance of estimating AD-related biomarker status. We further discussed the integration of multiple omics data, including genomics, proteomics, and metabolomics to improve biomarker robustness. We also addressed key challenges related to reproducibility, repeatibility, cohort heterogeneity, and clinical application. And we outline future directions of standardized, scalable, and clinically applicable diagnostic machineries.

Abstract: Background: Psychosocial assessment is a critical component of transplant candidate evaluation, yet its clinical utility is often limited by the descriptive nature of existing tools such as the Stanford Integrated Psychosocial Assessment for Transplantation (SIPAT). Translating multidimensional assessment data into actionable clinical insights remains a challenge in routine practice. Methods: We developed a clinical decision support application that integrates SIPAT item-level data with probabilistic risk estimation, visualization, and cohort-referenced interpretation. The application was based on a retrospective dataset of 496 lung transplant candidates evaluated at a single tertiary transplant center. Random forest–based models were used to transform SIPAT item-level data into probabilistic risk representations to estimate domain-specific risks, including depression, anxiety, nicotine-related risk, alcohol-related risk, illicit drug use, social support deficits, and non-adherence. Risk estimates were expressed as calibrated probabilities and categorized into clinically interpretable levels. Additional components included domain-level burden scoring and unsupervised clustering of multidomain risk profiles. Results: Estimated risks were predominantly low across the cohort, with high-risk subgroups identified for depression (6.5%), anxiety (2.2%), nicotine-related risk (11.3%), alcohol-related risk (4.4%), illicit drug use (2.2%), social support deficits (8.1%), and non-adherence (1.4%). Clustering analysis revealed three distinct profiles: a low-risk majority group, a subgroup characterized by elevated nicotine-related risk, and a small high-burden group with substantially elevated psychological distress, reduced social support, and increased non-adherence risk. Risk estimates showed strong and domain-consistent correlations with SIPAT scores (Spearman rho up to 0.80, p < 0.001). Feature importance analyses confirmed that risk estimation was primarily driven by clinically relevant SIPAT items. The application generated structured outputs integrating risk estimates, visualization, and intervention prioritization. Conclusions: The proposed application translates SIPAT-based psychosocial assessment into structured, multidomain risk profiles that enhance clinical interpretability and support targeted psychosocial prehabilitation. This approach provides a practical framework for translating psychosocial assessment into individualized intervention planning in lung transplant settings.

Abstract: Aiming at the core limitations of single large language models in complex task solving, including coarse task decomposition, cumulative long-chain reasoning errors, and the lack of explicit cross-agent collaboration, this paper proposes a large language model-driven multi-agent collaborative method. A hierarchical and role-based agent architecture is designed to separate task decomposition, specialized reasoning, result verification, and decision fusion, thereby enabling modular task solving and closed-loop orchestration over the full execution process. In addition, an efficient semantic communication mechanism is introduced to transmit compressed reasoning states across agents without breaking intermediate logical dependencies. A dynamic feedback iteration module is further employed to adjust routing strategy, collaboration intensity, and reasoning paths in real time according to subtask progress and verification outcomes. Comparative experiments on mathematical reasoning, multi-step planning, and complex information integration show that, relative to a single large language model, the proposed method improves the average completion rate by 21.3%, reduces the long-chain reasoning error rate by 18.7%, and reaches 92.6% cross-agent decision consistency. These results demonstrate that structured collaboration substantially improves robustness and accuracy for complex task solving and provides a practical technical path for diverse intelligent systems.

Abstract: Drug-related criminal activities on social media increasingly employ dynamic coded language—such as fruit substitutions, numeric homophones, and dialectal metaphors—to evade detection. This linguistic obfuscation poses significant challenges to conventional keyword-based monitoring systems. Furthermore, the scarcity of open-source datasets capturing these specific evasive expressions severely impedes automated detection research. To address these limitations, we construct a dedicated dataset of 10000 samples of drug-related coded texts sourced from mainstream Chinese social media platforms. Concurrently, we propose an optimized, TextCNN-based deep learning framework tailored for the automated identification of such illicit content. By leveraging multi-scale convolutional feature extraction, our model effectively captures intricate local semantic patterns and morphological variations inherent in short, highly noisy social media texts. Experimental results demonstrate that the proposed method achieves an F1-score of 99.3%, significantly outperforming established baseline approaches in the semantic representation of coded language. These findings indicate that our framework provides an efficient, robust, and scalable computational solution for intelligent drug-related content monitoring in complex online environments.

Abstract: Autonomous driving systems rely on a sophisticated pipeline of artificial intelligence models to perceive, predict, and plan in dynamic environments. This review presents a systematic analysis of the machine learning and deep learning models underpinning vehicle autonomy, spanning classical convolutional neural networks (CNNs) for object detection and semantic segmentation, to recurrent and Transformer-based architectures for trajectory prediction and motion planning. In this review, a critical examination of the autonomous vehicle sensor stack—including cameras, LiDAR, radar, ultrasonics, and GNSS/IMU as data acquisition systems, highlighting modality-specific AI challenges such as monocular depth estimation, 3D point cloud processing, and radar Doppler interpretation. The evolution of perception and decision-making pipelines is reviewed, contrasting modular architectures with end-to-end learning paradigms that directly map raw sensor data to control commands, and discussing their trade-offs in interpretability, safety assurance, and robustness to rare edge cases. We further survey specialized hardware accelerators and heterogeneous automotive SoCs designed to meet stringent real-time and power constraints. Industrial strategies are compared, including multi-modal sensor fusion and vision-centric approaches based on large-scale imitation learning. Finally, we identify open challenges related to robustness under adverse conditions, domain shift, causal ambiguity, and the need for interpretable and certifiable AI in safety-critical autonomous driving systems.

Abstract: Ultraweak photon emission (UPE) refers to spontaneous, low-intensity photon release from biological systems, generated largely through oxidative metabolic reactions involving reactive oxygen species, lipid peroxidation, mitochondrial activity, and electronically excited molecular intermediates. Because the nervous system is highly metabolically active and vulnerable to oxidative stress, hypoxia, excitotoxicity, inflammation, and mitochondrial dysfunction, UPE may offer a noninvasive optical window into neural physiology and disease. In this narrative review, we examine experimental and translational evidence linking UPE to nervous system function, with emphasis on neuronal excitation, glutamate-mediated activity, ischemia-reperfusion injury, stroke, neurodegeneration, mental-state and anesthesia paradigms, photobiomodulation, demyelinating disease, Parkinson disease, amyotrophic lateral sclerosis, and neuro-oncology. Across these domains, UPE appears most consistently associated with redox metabolism, mitochondrial function, oxidative stress, and excitation–metabolism coupling, whereas evidence that endogenous photons mediate functional neural signaling remains preliminary. Current data suggest that UPE may be most promising as a preclinical biomarker of tissue metabolic state, delayed post-ischemic dysfunction, and early neurodegenerative change, particularly when integrated with electrophysiology, perfusion imaging, molecular assays, and other physiologic measures. However, clinical translation is limited by low photon flux, limited temporal and spectral resolution, difficulty localizing signals from deep tissue, heterogeneous experimental protocols, and incomplete source attribution. Overall, UPE represents a promising but still early-stage framework for studying nervous system metabolism and disease, with future progress dependent on standardized methods, multimodal validation, and disease-specific investigation.

Abstract: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder for which early and accurate diagnosis remains a critical challenge. In this work, we propose a Multi-ROI Multimodal 3D Vision Transformer for AD classification that integrates structural MRI data with clinical and volumetric biomarkers within a unified attention-based framework. The proposed approach leverages anatomically guided multi-region-of-interest (ROI) decomposition to focus on disease-relevant brain structures, including the hippocampus, entorhinal cortex, fornix, and major cortical lobes. Each ROI is encoded using 3D tubelet embeddings, while clinical and volumetric features are transformed into feature-wise tokens, enabling seamless multimodal fusion through self-attention mechanisms. A hemisphere-aware selection strategy is introduced to identify the most discriminative ROI representations, enhancing both performance and interpretability. The model is evaluated on a merged multi-cohort dataset combining ADNI, AIBL, and OASIS, using a 7-fold cross-validation protocol. Experimental results demonstrate that the proposed method achieves high classification performance, reaching an accuracy of 97.62% and an AUC of 0.9940, outperforming single-modality and whole-brain baselines. Furthermore, attention-based analysis provides interpretable insights into the relative importance of clinical and neuroanatomical features, revealing consistency with established AD biomarkers. These findings highlight the effectiveness of multimodal integration and ROI-based representation for robust and explainable AD classification.