Brain Informatics最新文献

The integration of multimodal data has emerged as a powerful strategy for enhancing the accuracy and interpretability of artificial intelligence (AI) models in the diagnosis and prognosis of Alzheimer's Disease (AD). This systematic review presents a comprehensive synthesis of recent advances in AI-driven multimodal fusion approaches for AD prediction. A detailed examination of widely used datasets-including their modalities, preprocessing pipelines, and accessibility-is provided to aid reproducibility and methodological transparency. We analyze and categorize the various data harmonization and preprocessing techniques employed across neuroimaging (e.g., fMRI, sMRI, PET), electrophysiological (EEG), and genomic modalities, highlighting domain-specific practices and challenges. Furthermore, fusion strategies are classified into data-level, feature-level, decision-level, and temporal (early, intermediate, and late) paradigms, offering insights into their implementation and diagnostic impact. The review also investigates the adoption of explainable AI (XAI) techniques across studies and identifies a significant underrepresentation of works that simultaneously emphasize multimodality, explainability, and methodological rigor. By adhering to both PRISMA and Kitchenham's guidelines, this review ensures transparency and replicability in evidence synthesis. Compared to existing reviews, our work uniquely focuses on the intersection of multimodal integration and explainability within a systematically validated framework. The review concludes with recommendations for future research aimed at developing robust, interpretable, and clinically relevant AI models for AD.

Explainability is crucial for establishing user trust in Artificial Intelligence (AI), particularly within safety-critical domains such as Air Traffic Management (ATM) and Air Traffic Control (ATC). This study empirically investigates the effects of Explainable AI (XAI), specifically HeatMap-based visual explanations, on cognitive workload, user acceptance, and intention to use AI-driven decision-support systems among Air Traffic Control Officers (ATCOs). Despite significant theoretical advancements in the broader XAI domain, empirical evidence addressing the specific impact of visual explanations on human-AI interactions in safety-critical environments like ATC remains limited. To address these critical gaps, an experimental comparison was conducted between explainable (HeatMap) and non-explainable (BlackBox) AI conditions, involving two user groups: expert and student ATCOs. Both objective neurophysiological measures (Electroencephalography) and subjective questionnaires were employed to capture comprehensive user responses. Key findings revealed that the presence of visual explanations significantly reduced cognitive workload and enhanced users' willingness to adopt the AI system, regardless of participants' level of expertise. However, explicit perceptions of AI's impact on work performance were predominantly influenced by expertise, with less experienced controllers reporting a greater perceived impact than their expert counterparts. By combining objective neurometrics with subjective user assessments, this research advances methodological rigor in evaluating human-AI interactions and highlights the importance of tailored, user-centric explanations. These findings directly contribute to practical guidelines for designing cognitively compatible and trustworthy AI tools in ATC, providing nuanced insights for targeted training and deployment strategies based on user expertise.

Purpose: Wearable EEG systems are increasingly used for brain-computer interface (BCI) applications beyond controlled laboratory environments. However, there is still limited evidence on their reliability in real-world cognitive monitoring, especially for deriving robust mental-state indicators. This study investigates the signal quality, computational stability, and neurometric consistency of two widely used consumer-grade EEG devices (Emotiv EPOC X and Muse S) compared to a validated research-grade system (Mindtooth Touch) during naturalistic tasks relevant to passive BCIs and brain-machine intelligence.

Method: Twenty-four participants completed a multimodal protocol including video observation, multitasking under varying cognitive loads, and a simulated driving task. Each participant used all three EEG systems in a counterbalanced order to avoid any bias induced by the order. Signal quality was assessed through artefact analysis and Power Spectral Density (PSD) stability. Neurometrics, i.e., metrics related to specific mental and emotional states that can be extracted from EEG signal processing (workload, attention, vigilance, and approach-withdrawal) were extracted and compared across devices, conditions, and subjective reports of effort and comfort.

Finding: The research grade system demonstrated higher signal stability, fewer artefacts, and more consistent neurometric responses to cognitive variations, with high significant correlation with subjective measures. Post-processing improved data continuity in consumer devices, but neurometrics remained less sensitive to task demands and less aligned with subjective ratings. Each device reflected different trade-offs between data quality, usability, and cost.

Conclusion: Research-grade systems remain more reliable for passive BCI applications requiring high-resolution cognitive state monitoring. Nevertheless, consumer-grade headsets may still be appropriate for exploratory studies or non-critical applications. This work highlights key trade-offs between signal quality, usability, and application goals, contributing to the broader integration of wearable neurotechnologies into brain-machine intelligence frameworks.

Background/objective: Predicting neurological outcomes in comatose cardiac arrest survivors remains challenging. Diffusion tensor imaging (DTI) offers potential as an objective biomarker of white matter injury, but its prognostic value needs further validation. We aimed to investigate the predictive value of DTI-derived metrics for six-month neurological outcomes in comatose cardiac arrest patients.

Methods: This prospective study enrolled 28 comatose cardiac arrest patients (mean age 54.36 ± 3.01 years; 71% male) and 28 age-/sex-matched healthy controls (HCs). All participants underwent 3T brain MRI (median 4 days post-arrest). DTI parameters (fractional anisotropy [FA], mean diffusivity [MD], axial diffusivity [AD], radial diffusivity [RD]) were analyzed using Tract-based spatial statistics (TBSS) and ROI approaches based on white matter atlas. Neurological outcome was assessed at six months using the modified Rankin Scale (good outcome: mRS 0-2; poor outcome: mRS 3-5). Statistical analyses included voxel-wise comparisons and ROC curve analysis for predictive performance.

Results: Compared to HCs, patients showed widespread reductions in FA, MD, AD, and RD (TFCE-corrected p < 0.05). Patients with poor outcomes (n = 18) exhibited significantly lower DTI metrics than those with good outcomes (n = 10) across most white matter tracts. The combination of whole-brain FA and RD demonstrated exceptional prognostic accuracy for good outcome (AUC = 0.984; 95% CI 0.925-1.000; sensitivity 92%, specificity 97.7%), significantly outperforming clinical variables and individual DTI parameters. ROI analysis identified specific tracts (e.g., right cingulum hippocampus, right uncinate fasciculus) with high predictive values. Ventricular fibrillation as initial rhythm was more frequent in the group with good outcomes.

Conclusions: DTI metrics, particularly the combination of FA and RD, provided outstanding early prediction of good six-month neurological outcomes after cardiac arrest, surpassing traditional biomarkers. These findings supported integrating DTI into multimodal prognostic models to guide clinical decisions and prevent premature withdrawal of life-sustaining therapy.

Primary angle-closure glaucoma (PACG), an irreversible blinding disease characterized by retinal ganglion cell damage and optic nerve atrophy, exerts significant effects on brain functional networks. Using resting-state functional magnetic resonance imaging (rs-fMRI) data from 34 PACG patients and 34 matched healthy controls (HCs), we extracted four types of connectivity features-voxel-wise static functional connectivity (FC), dynamic functional connectivity (dFC), effective connectivity (EC), and dynamic effective connectivity (dEC)-via the AAL90 (Automated Anatomical Labeling 90) atlas following preprocessing. Elastic net feature selection was applied independently to each connectivity type to retain the top 10% most discriminative features. We evaluated the classification performance of ten machine learning models using individual feature types as well as their combined features, with the FC-based logistic regression (LR) model achieving optimal diagnostic efficacy (accuracy = 0.92, AUC = 0.96). SHapley Additive exPlanations (SHAP) of the model identified 20 critical connections, revealing abnormal patterns at both the region of interest (ROI)-level and network-level within brain networks such as the visual network (VSN), dorsal attention network (DAN), and sensorimotor network (SMN). Statistical group comparisons validated reduced connectivity (e.g., VSN-SMN, VSN-DAN) and enhanced DAN-thalamus connectivity in patients, while voxel-wise analyses of key regions confirmed diminished connectivity to visual areas. The results provide insights into how machine learning can be effectively employed to detect PACG-specific brain network disruptions and highlight potential neuroimaging biomarkers.

Alzheimer's disease (AD), a significant public health challenge, requires accurate early diagnosis to improve patient outcomes. Vision Transformers (ViTs) and Convolutional Vision Transformers (CViTs) have emerged as powerful Deep Learning architectures for this task. Following PRISMA guidelines, this systematic review analyzes 68 studies selected from 564 publications (2021-2025) across five major databases: Scopus, Web of Science, ScienceDirect, IEEE Xplore, and PubMed. We introduce novel taxonomies to systematically categorize these works by model architecture, data modality, fusion strategy, and diagnostic objective. Our analysis reveals key trends, such as the rise of hybrid CViT frameworks, and critical gaps, including a limited focus on Mild Cognitive Impairment-to-AD progression. Critically, we also assess practical implementation details, revealing widespread challenges in algorithmic reproducibility. The discussion culminates in a forward-looking analysis of Large Vision Models and proposes future directions emphasizing the need for robust multimodal integration, lightweight transformer designs, and Explainable AI to advance AD research and bridge the critical gap between high-performance modeling and clinical applicability.

Objective: Brain functional connectivity alterations have been observed in cardiovascular diseases (CVDs), but the causality between brain resting-state functional connectivity networks and CVDs remains undetermined. We aimed to investigate the bidirectional causality between brain network connectivity and CVDs using Mendelian randomization (MR) analysis.

Methods: Using genome-wide association study (GWAS) data from the UK Biobank (n = 34,691), we conducted bidirectional two-sample MR analyses between 191 resting-state functional MRI phenotypes and four major CVDs: hypertension, atrial fibrillation (AF), heart failure (HF), and coronary artery disease (CAD). Sensitivity analyses, including MR-Egger regression and weighted median methods, were conducted to ensure the robustness of causal estimates and to test for potential pleiotropy.

Results: For hypertension, four networks showed negative causal associations (ORs 0.882-0.904), primarily involving motor, subcortical-cerebellar, default mode, and visual networks. In AF, we observed both increased connectivity in salience and default mode networks (ORs 1.157-1.288) and decreased connectivity in visual-motor networks (OR 0.790). For HF, three networks showed significant associations: decreased connectivity in visual and temporal networks (ORs 0.791-0.804) and increased connectivity in motor networks (OR 1.352). CAD was associated with increased connectivity in both default mode and central executive networks (ORs 1.145-1.147). These relationships remained robust after multiple sensitivity analyses.

Conclusion: Our findings reveal novel bidirectional causal relationships between specific brain functional networks and CVDs, with distinct patterns of network involvement for different CVDs suggesting disease-specific mechanisms in the cardio-cerebral axis. These findings identify potential neuroimaging biomarkers for early detection and monitoring of cardiovascular diseases.

Partial differential equations (PDEs) model various physical phenomena, such as electromagnetic fields and fluid mechanics. Methods such as sparse identification of nonlinear dynamics (SINDy) and PDE-Net 2.0 have been developed to identify and model PDEs on the basis of data via sparse optimization and deep neural networks, respectively. While PDE models are less commonly applied to fMRI data, they have the potential to uncover hidden connections and essential components in brain activity. Using the ADHD200 dataset, we applied canonical independent component analysis (CanICA) and uniform manifold approximation (UMAP) for dimensionality reduction of fMRI data. We then used sparse ridge regression to identify PDEs from the reduced data, and applied significant PDE features for classification achieving high accuracy in distinguishing individuals with attention deficit hyperactivity disorder (ADHD). This study demonstrates a novel approach for extracting meaningful features from fMRI data for neurological disorder analysis to understand the role of oxygen transport (delivery & consumption) in the brain during neural activity, which is relevant for studying intracranial pathologies.

Detecting Brain Tumors is essential in medical imaging, as early and accurate diagnosis significantly improves treatment decisions and patient outcomes. Convolutional Neural Networks have demonstrated high efficiency in this domain, but their lack of interpretability remains a significant drawback for clinical adoption. This study explores the integration of Explainability techniques to enhance transparency in CNN-based classification and improve model performance through advanced optimization strategies. The primary research question addressed is how to improve the accuracy, generalization, and interpretability of CNNs for brain tumor Detection. While previous studies have demonstrated the effectiveness of deep learning for tumor detections, challenges such as class imbalance and overfitting of CNNs persist. To bridge this gap, we employ different dynamic learning rate modifiers, perform architectural enhancements, and apply XAI techniques, including Grad-CAM and LIME. Our experiments are conducted on three publicly available multiclass tumor datasets to ensure the generalizability of the proposed approach. Among the tested architectures, the enhanced ResNet model consistently outperformed others across all datasets, achieving the highest test accuracy, ranging from 99.36% to 99.65%. The techniques such as unfreezing layers, integrating various blocks, pooling, and dropout layers enhanced feature refinement and reduced overfitting. By incorporating XAI, we improve model interpretability, ensuring that clinically relevant regions in MRI scans are highlighted. These advancements contribute to highly reliable AI-assisted diagnostics, addressing significant challenges in medical image classification.