Most recent paper

BMC Med. 2025 Dec 3;23(1):675. doi: 10.1186/s12916-025-04556-3.

ABSTRACT

BACKGROUND: Bulimia nervosa (BN) is a severe psychiatric disorder characterized by dysregulated eating behaviors and impaired cognitive-emotional control. Despite increasing recognition of brain network dysfunction in BN, the interplay between structural connectivity (SC) and functional connectivity (FC), termed SC-FC coupling, remains poorly understood. This study aimed to comprehensively characterize SC-FC coupling alterations in BN using multimodal neuroimaging and to evaluate the predictive value for disordered eating behaviors.

METHODS: This study enrolled 79 patients with BN and 69 healthy controls who underwent high-resolution structural magnetic resonance imaging (MRI), diffusion tensor imaging (DTI), and resting-state functional MRI (rs-fMRI). Functional and structural connectomes were constructed using the Schaefer-400 atlas. SC-FC coupling was quantified using eight biologically grounded similarity and communication metrics. A multivariate linear modeling framework was applied to estimate region-specific coupling profiles. Group comparisons and ridge regression-based leave-one-out cross-validation were used to identify altered coupling and predict symptom severity.

RESULTS: The global topological properties of the SC and FC networks were preserved in BN. However, patients exhibited significantly reduced degree centrality and nodal efficiency in the inferior frontal gyrus within the FC network. SC-FC coupling, quantified using the matching index (MI), showed widespread regional alterations in BN, particularly within the default mode, control, and attention networks. Seventeen brain parcels demonstrated significant group differences in MI-based coupling (false discovery rate (FDR)-corrected, p < 0.05), with both hypercoupling and hypocoupling observed. Findings were parcellation-robust (Glasser-360 replication; Dice = 0.93 vs. Schaefer-400). Moreover, coupling features moderately predicted binge-eating frequency (r = 0.24, p < 0.001), but not questionnaire-based emotional or behavioral scores.

CONCLUSIONS: In BN, macroscale white-matter organization is preserved, yet focal prefrontal functional decentralization and widespread, parcellation-robust SC-FC coupling changes invisible to single-modality analyses were observed. Multidimensional SC-FC coupling provides a sensitive neurobiological marker that explains clinically relevant variance in binge-eating behavior, highlighting its potential as a target for personalized diagnosis and intervention in BN.

PMID:41340129 | DOI:10.1186/s12916-025-04556-3

Addict Biol. 2025 Dec;30(12):e70105. doi: 10.1111/adb.70105.

ABSTRACT

Cannabis use disorder (CUD) affects ~22-million people globally and is characterised by difficulties in cutting down and quitting use, but the underlying neurobiology remains unclear. We examined resting-state functional connectivity (rsFC) between regions of interest (ROIs) of the addiction neurocircuitry and the rest of the brain in 65 individuals with moderate-to-severe CUD who reported attempts to cut down or quit, compared to 42 controls, and explored the association between rsFC and cannabis exposure and related problems, to elucidate potential drivers of rsFC alterations. The CUD group showed greater rsFC than controls between ROIs implicated in reward processing and habitual substance use (i.e., nucleus accumbens, putamen and pallidum) and occipito/parietal areas implicated in salience processing and disinhibition. Putamen-occipital rsFC correlated with levels of problematic cannabis use and depression symptoms. CUD appears to show neuroadaptations of the addiction neurocircuitry, previously demonstrated in other substance use disorders.

PMID:41339716 | DOI:10.1111/adb.70105

Zhonghua Nei Ke Za Zhi. 2025 Dec 1;64(12):1226-1234. doi: 10.3760/cma.j.cn112138-20250707-00395.

ABSTRACT

Objective: To investigate the relationship between alterations in dynamic functional connectivity (dFC) and spatial navigation abilities in individuals with subjective cognitive decline (SCD) across different Traditional Chinese Medicine (TCM) constitutions. Methods: Seventy-five participants with SCD, comprising 34 individuals with balanced constitutions and 41 individuals with biased constitutions, were recruited from the Affiliated Drum Tower Hospital of Nanjing University Medical School between August 2022 and January 2025. The participants underwent TCM constitution assessment, spatial navigation ability testing, and neuropsychological scale evaluation. Additionally, each participant was assessed using 3.0 T resting-state functional magnetic resonance imaging (rs-fMRI) and high-resolution T1-weighted imaging scans. Based on prior research, 20 spatial navigation-related regions of interest (ROIs) were defined. Afterwards, rs-fMRI time series were segmented using a sliding time window approach before calculating the dFC within the spatial navigation brain network. Results: Compared to the balanced constitution group, the biased constitution SCD group showed significantly lower scores on the Mini-Mental State Examination (MMSE) (z=-3.05, P=0.002) and the Auditory Verbal Learning Test (AVLT) measures: immediate recall (z=-2.12, P=0.035), short-delay recall (z=-2.22, P=0.026), long-delay recall (z=-2.88, P=0.004), cued recall (z=-2.91, P=0.004), and recognition (z=-2.20, P=0.028). They also exhibited significantly higher average error distances in ego-allocentric navigation (z=-2.28, P=0.023), egocentric navigation (z=-2.31, P=0.021), and delayed navigation (z=-2.02, P=0.043). Participants with SCD who had a biased constitution also demonstrated significantly reduced dFC between the left parahippocampal gyrus (PHG) and left prefrontal cortex (PFC) (t=2.43), right precuneus and right retrosplenial cortex (RSC) (t=2.96), and left inferior parietal lobule (IPL) and left hippocampus (t=2.42) (all P<0.05, Bonferroni-corrected). Conversely, the dFC was significantly increased between the right PHG and left PFC (t=-2.29, P<0.05, Bonferroni-corrected). Significant correlations were also found in participants with SCD who had biased constitutions: the dFC between the left PHG and left PFC positively correlated with the egocentric navigation average total error (r=0.34, P=0.030) and negatively correlated with the visuospatial memory cognitive domain (r=-0.35, P=0.026); the dFC between the left IPL and left hippocampus negatively correlated with the egocentric navigation average total error (r=-0.32, P=0.043); and the dFC between the right PHG and left PFC positively correlated with the delayed navigation average total error (r=0.33, P=0.037). The area under the ROC curve for the combined differences in cognitive assessments, spatial navigation behavior, and navigation-related brain network dFC was 0.966 in predicting biased constitution versus balanced constitution in participants with SCD. Conclusions: Individuals with SCD and biased constitutions demonstrated poorer spatial navigation ability, possibly due to altered dFC within the spatial navigation brain network. Furthermore, the integrated model based on spatial navigation behaviors and dFC exhibited a high predictive value in distinguishing between individuals with SCD who had balanced and biased constitutions.

PMID:41338558 | DOI:10.3760/cma.j.cn112138-20250707-00395

Sci Data. 2025 Dec 3;12(1):1893. doi: 10.1038/s41597-025-06183-2.

ABSTRACT

A central challenge in speech perception is the lack of a one-to-one mapping between acoustic patterns and linguistic interpretations. This is often resolved through intrinsic normalization, where acoustic cues mutually influence each other's categorization. Notably, segmental (e.g., consonants, vowels) and suprasegmental (e.g., tone) features overlap temporally during speech perception, giving rise to complex interactions across linguistic and acoustic levels. However, the neural basis of these interactions remains underexplored due to a lack of integrated neuroimaging datasets designed for this purpose. This dataset presents a multimodal neuroimaging resource comprising structural MRI (sMRI), resting-state fMRI (rs-fMRI), categorization task-based fMRI, diffusion MRI (dMRI), and behavioral data from 28 participants (14 females, mean age 20.79 ± 1.52 years). Each participant completed two separate two-alternative forced-choice categorization tasks using 7 × 7 consonant-tone and vowel-tone continua. This resource is uniquely valuable for its explicit design to capture interactions between segmental and suprasegmental features, enabling researchers to explore neural representations, functional connectivity, and white matter correlates of speech normalization processes.

PMID:41339578 | DOI:10.1038/s41597-025-06183-2

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-4. doi: 10.1109/EMBC58623.2025.11253471.

ABSTRACT

The use of functional magnetic resonance imaging (fMRI) to map brain activity through functional network connectivity (FNC) has become a focal point in research. Most studies focus on static or dynamic FNC between predefined spatial network nodes, neglecting the possibility of time-varying dynamics within these spatial networks. While recent methods estimate voxel-level spatial dynamic networks, no approach has explored FNC between these spatial dynamic networks. In this study, we propose a novel method for examining FNC within spatially dynamic brain networks using human resting-state fMRI (rsfMRI) data. This method enables the calculation of network-specific FNC across (localized) voxel subsets. We applied this technique to the baseline dataset of 100 participants from the large-scale Adolescent Brain and Cognitive Development (ABCD) study. We first show our voxel-based FNC approach successfully replicates traditional static FNC results, demonstrating similar significant modularity in both the static FNC (sFNC) and global voxel FNC (GvFNC) matrices. The key advancement of our approach, however, lies in its ability to investigate local FNC within different voxel subsets. The findings reveal a reduction in anticorrelations within the average local voxel FNC (LvFNC) as the voxel inclusion rate decreases.

PMID:41337285 | DOI:10.1109/EMBC58623.2025.11253471

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11253918.

ABSTRACT

Anomic aphasia is a subtype of aphasia, characterized by impaired naming functions while other language abilities remain relatively intact. However, due to the relatively mild symptoms, patients with anomic aphasia are often prone to misdiagnosis or underdiagnosis, which may delay treatment and intervention. This study employed resting-state functional magnetic resonance imaging (rs-fMRI) and machine learning techniques to classify anomic aphasia and differentiate it from post-stroke non-aphasic subjects, while also investigating the neural mechanisms underlying its manifestation. Brain imaging analysis techniques, including fractional amplitude of low-frequency fluctuations (fALFF), regional homogeneity (ReHo), and the Laterality Index (LI), were used to analyze data from 95 subjects to reveal significant differences in brain activity between anomic aphasia subjects and post-stroke non-aphasic subjects. Subsequently, these imaging-derived features were used to train and validate machine learning classifiers. Among the classifiers tested, the Multilayer Perceptron (MLP) achieved an accuracy of 94.74% in distinguishing between the two groups. Collectively, our findings highlight the potential of automated methods based on neuroimaging and machine learning in assisting clinicians to enhance diagnostic efficiency for anomic aphasia, enabling the early detection of symptoms and timely intervention.Clinical Relevance- Subjects with anomic aphasia exhibited increased rightward activation in regions such as the superior frontal gyrus and inferior frontal gyrus, coupled with reduced activation in the inferior parietal lobule and superior temporal gyrus.

PMID:41337239 | DOI:10.1109/EMBC58623.2025.11253918

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-4. doi: 10.1109/EMBC58623.2025.11254083.

ABSTRACT

Alzheimer's disease (AD) is a neurodegenerative disorder that causes cognitive decline, and early detection remains a challenge. Resting-state functional MRI (rs-fMRI) has shown potential for identifying early AD signs by analyzing brain connectivity. In this study, we propose a Hierarchical GCN-Transformer Network (HGTNet) for brain age prediction using rs-fMRI data. Through experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, we demonstrate that, compared to traditional machine learning and deep learning methods, the combination of Graph Convolutional Networks (GCN) and Transformer architecture enhances the model's ability to capture complex brain interactions. Our model's more accurate brain age predictions provide a valuable step in identifying early neurodegenerative changes, aiding in the better intervention and management of Alzheimer's disease.

PMID:41337145 | DOI:10.1109/EMBC58623.2025.11254083

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11252645.

ABSTRACT

Resting-state functional MRI (rs-fMRI) data analysis provides essential insights into early neurodevelopment through longitudinal assessment of functional connectivity (FC) patterns in infant brains, which may help uncover critical biomarkers for developmental monitoring. However, due to challenges in acquiring high-quality functional MRI (fMRI) data in infants, such as strong motion artifacts, short scan durations, and participant compliance, longitudinal FC of infants remain scarce, which significantly hampers the capacity to systematically investigate early functional brain development. To address this challenge, we propose MAD-Net, a novel diffusion model that predicts longitudinal FC from morphometric features derived from structural MRI (sMRI). Our framework integrates classifier-free guidance with a cross-modal attention mechanism, enabling the dynamic fusion of morphometric features and developmental age constraints during the diffusion process. A shared triplet encoder learns robust feature representations from longitudinal data, while a U-Net-based architecture ensures precise conditioning on individual morphometry and target age. We evaluate MAD-Net on 386 longitudinal infant fMRI scans and demonstrate its superior performance in FC prediction compared to state-of-the-art methods. By integrating diffusion-based learning, structural priors, and age-dependent constraints, MAD-Net represents a significant advancement in neuroimaging-based functional network reconstruction. The code is available at https://github.com/IPMI-NWU/MAD-Net.

PMID:41336914 | DOI:10.1109/EMBC58623.2025.11252645

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-7. doi: 10.1109/EMBC58623.2025.11251603.

ABSTRACT

Alzheimer's Disease (AD) is characterized by progressive functional network disruptions that precede cognitive decline, yet traditional functional connectivity analyses often fail to capture transient network instabilities critical for early diagnosis. This study investigates the role of Quasi-Periodic Patterns (QPPs) in identifying disease-related connectivity changes across longitudinal stable disease stages (sNC, sMCI, sDAT) and transitioning (uNC, pMCI) AD cohorts using resting-state fMRI data from the Alzheimer's Disease Neuroimaging Initiative. By integrating QPP occurrences with intrinsic connectivity networks (ICNs), we assessed network integrity across disease stages, with statistical significance evaluated using the Kruskal-Wallis test and Dunn's test for post-hoc analysis. Results revealed a progressive decline in functional connectivity integrity, with early impairments in subcortical and executive function networks in stable groups, followed by widespread disconnection in higher cognition, sensorimotor, and visual networks at later stages. Transitioning AD groups exhibited earlier disruptions in visual and cerebellar networks, suggesting their potential as early biomarkers for disease onset. The occurrence of QPPs decreased significantly with disease progression, indicating an increase in functional disconnection. These findings highlight the synergy between QPPs and ICNs as a dynamic and sensitive biomarker framework for AD progression. Future research should further explore this integration within multimodal imaging and clinical diagnostic frameworks to enhance early detection and intervention strategies.

PMID:41336827 | DOI:10.1109/EMBC58623.2025.11251603

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-4. doi: 10.1109/EMBC58623.2025.11252770.

ABSTRACT

In this study, we introduce Dynamic Inter-Modality Source Coupling (dIMSC), an extension of our earlier Inter-Modality Source Coupling (IMSC) method. While IMSC evaluated the coupling between source-based morphometry (SBM) from structural MRI (sMRI) and static functional network connectivity (sFNC) from resting-state fMRI (rs-fMRI), dIMSC incorporates the temporal dimension by linking SBM with dynamic functional network connectivity (dFNC). Using data from the Adolescent Brain Cognitive Development (ABCD) study, we applied dIMSC to examine brain connectivity and evaluate sex differences in children aged 9-11. Our analysis revealed significant sex-specific patterns: males exhibited stronger positive coupling in the putamen and hippocampus, while females showed stronger coupling in the superior parietal lobule and anterior cingulate cortex. On average, 27.12% of timecourses exhibited positive coupling, 46.63% neutral coupling, and 26.25% negative coupling, reflecting a balanced alignment between structural and functional features. Sex differences were also observed in neutral and negative coupling groups, with males demonstrating stronger coupling in the caudate and middle cingulate gyrus, and females in the cerebellum and inferior parietal lobule. These findings suggest distinct developmental trajectories in brain network organization between sexes, potentially reflecting sex-specific adaptations in functional integration and compensatory mechanisms. The dIMSC method advances our earlier work by enabling time-sensitive analysis of brain structure-function coupling, providing a powerful framework for investigating neurodevelopmental processes and their implications for cognitive and behavioral outcomes.Clinical RelevanceThis study identifies sex-specific patterns in brain connectivity during childhood, offering insights that could inform sex-tailored diagnostic and therapeutic approaches for neurodevelopmental disorders.

PMID:41336665 | DOI:10.1109/EMBC58623.2025.11252770

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11253394.

ABSTRACT

Understanding the relationship between cognition and intrinsic brain activity through purely data-driven approaches remains a significant challenge in neuroscience. Resting-state functional magnetic resonance imaging (rs-fMRI) offers a non-invasive method to monitor regional neural activity, providing a rich and complex spatiotemporal data structure. Deep learning has shown promise in capturing these intricate representations. However, the limited availability of large datasets, particularly for disease-specific groups such as Alzheimer's Disease (AD), constrains the generalizability of deep learning models. In this study, we focus on multivariate time series forecasting of independent component networks derived from rs-fMRI as a form of data augmentation, using both a conventional LSTM-based model and the novel Transformer-based BrainLM model. We assess their utility in AD classification, demonstrating how generative forecasting enhances classification performance. Post-hoc interpretation of BrainLM reveals class-specific brain network sensitivities associated with AD.

PMID:41336569 | DOI:10.1109/EMBC58623.2025.11253394

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-7. doi: 10.1109/EMBC58623.2025.11251833.

ABSTRACT

Visual hallucinations can severely impact the quality of life of affected individuals and are linked to greater disease severity in psychosis. To facilitate the detection of imaging biomarkers of visual hallucinations, we developed an automated pipeline to compare and evaluate feature extraction and classification methods using resting-state functional MRI scans from individuals with and without visual hallucinations. Five common functional connectivity features were assessed in this study: Regional Homogeneity, Voxel-Mirrored Homotopic Connectivity, Amplitude of Low Frequency Fluctuations, Fractional Amplitude of Low Frequency Fluctuations, and Eigenvector Centrality Mapping. We further evaluated the use of Pearson correlation in feature selection with different cutoff-values and employed a linear support vector machine for classification. The pipeline was validated on a dataset of 45 individuals, including people with psychosis and healthy controls. The model performance was evaluated based on the classification accuracy, sensitivity, specificity, as well as the interpretability of the feature weights. The code for the created pipeline is publicly available: https://github.com/LEO-UMCG/Visual_Hallucinations_Classification.Clinical relevance- Classification of patients based on biomarkers holds potential for complementing clinical measures, predicting future cases, and guiding personalized treatment of schizophrenia. The comparison of feature types and identification of imaging-based biomarkers in this study provide valuable insights for future research on VH classification and the underlying mechanisms of visual hallucinations.

PMID:41336506 | DOI:10.1109/EMBC58623.2025.11251833

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-7. doi: 10.1109/EMBC58623.2025.11253094.

ABSTRACT

Autism Spectrum Disorder (ASD) is a complicated neurodevelopmental condition with numerous symptoms, making accurate diagnosis and the identification of reliable biomarkers particularly challenging. Recent advances in deep neural networks using connectivity features derived from resting-state functional magnetic resonance imaging have greatly extended our understanding of ASD and improved its diagnostic accuracy. However, most existing methods primarily focus on pairwise connections, limiting their ability to capture higher-order interactions in brain networks and resulting in suboptimal predictive performance. In this paper, to enhance the learning of higher-order relationships and improve model interpretability, we propose an Adaptive Hypergraph Contrastive Learning (AHCL) framework for ASD classification. Specifically, AHCL employs a trainable masking mechanism to adaptively estimate latent hyperedges, allowing the generation of two hypergraph views with distinct topological structures. Additionally, AHCL incorporates low-rank loss to improve the compactness of intra-class samples, effectively addressing the limitation of traditional contrastive learning in distinguishing negative samples. By jointly optimizing view similarity loss and contrastive loss, the framework ensures semantic consistency across views while enhancing topological differences, leading to robust and noise-resistant feature representations with minimal information redundancy. Experimental results demonstrate that AHCL outperforms competing methods in ASD classification. Furthermore, it identifies disease-related connections and regions, providing valuable insights into ASD and offering potential techniques for more precise and interpretable diagnostic strategies.

PMID:41336148 | DOI:10.1109/EMBC58623.2025.11253094

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11253059.

ABSTRACT

Early infancy is a crucial period for brain development, during which fundamental functional and structural frameworks are established. Understanding the maturation of large-scale brain networks during this stage is essential for characterizing normative neurodevelopment and identifying potential deviations linked to neurodevelopmental disorders. In this study, we investigated developmental changes in the spatial organization of functional brain networks in infants using a longitudinal resting-state fMRI dataset comprising 137 scans from 74 low-likelihood developing infants aged 0-6 months. We applied independent component analysis to extract large-scale brain networks and utilized advanced spatial metrics, including network-averaged spatial similarity (NASS) to assess alignment with group-level patterns, network strength to quantify neural engagement based on voxel intensities, and network size to examine spatial distribution. Our findings reveal significant age-related increases in NASS across multiple networks, indicating greater consistency in functional organization over time. Additionally, most networks demonstrated increased network strength, reflecting heightened neural involvement, while network size exhibited distinct developmental trajectories, with some networks expanding and others remaining stable. These results highlight the dynamic evolution of functional brain architecture during early infancy, providing critical insights into neurodevelopmental processes.Clinical Relevance- This study provides critical insights into early brain network development, which is essential for identifying biomarkers of neurodevelopmental disorders such as autism and schizophrenia. By mapping typical maturation patterns using advanced spatial metrics, our findings offer a foundation for early detection of atypical development. Deviations in network organization and strength could serve as early indicators, supporting neuroimaging-based screening and intervention strategies to optimize neurodevelopmental outcomes.

PMID:41336104 | DOI:10.1109/EMBC58623.2025.11253059

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11251640.

ABSTRACT

Brain tumors significantly disrupt brain network organization, yet the temporal dynamics of network reorganization following surgical intervention remain poorly understood. This study investigated longitudinal changes in functional brain network properties across pre-surgical, post-surgical, and follow-up time points in glioma patients. Using graph theory analysis of resting-state functional magnetic resonance imaging (fMRI) data, we examined whole-brain network metrics as well as the connections involving perilesional and contralesional regions. Results revealed significant alterations in network topology over time, with distinct patterns of reorganization in perilesional and contralesional regions, suggesting mechanisms of plasticity and recovery in brain network architecture following tumor resection.Clinical Relevance-These findings have significant implications for surgical planning and post-operative care, suggesting the need for therapeutic approaches that consider both local and distant network effects. The demonstrated importance of contralesional adaptation particularly warrants attention in rehabilitation strategies, potentially opening new avenues for targeted interventions in recovery.

PMID:41336078 | DOI:10.1109/EMBC58623.2025.11251640

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11252604.

ABSTRACT

We present a novel topological framework for analyzing functional brain signals using time-frequency analysis. By integrating persistent homology with time-frequency representations, we capture multi-scale topological features that characterize the dynamic behavior of brain activity. This approach identifies 0D (connected components) and 1D (loops) topological structures in the signal's time-frequency domain, enabling robust extraction of features invariant to noise and temporal misalignments. The proposed method is demonstrated on resting-state functional magnetic resonance imaging (fMRI) data, showcasing its ability to discern critical topological patterns and provide insights into functional connectivity. This topological approach opens new avenues for analyzing complex brain signals, offering potential applications in neuroscience and clinical diagnostics.

PMID:41335918 | DOI:10.1109/EMBC58623.2025.11252604

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-4. doi: 10.1109/EMBC58623.2025.11253645.

ABSTRACT

In neuroimaging-based Alzheimer's Disease (AD) diagnosis, effectively integrating structural MRI (sMRI) and resting-state functional MRI (rs-fMRI) data while preserving clinical interpretability remains a significant challenge. To address this issue, we propose a novel transformer-based framework that unifies heterogeneous imaging features into coherent region-level representations. Our approach uniquely leverages prior anatomical knowledge to guide attention toward AD-relevant regions while employing a learnable mapping mechanism that transforms sMRI spatial features into biologically meaningful regional representations. We implement a consistency constraint to ensure optimal alignment between structural and functional coupling across modalities, followed by a Bayesian fusion strategy to integrate these aligned multi-modal features. Through comprehensive evaluation on the ADNI dataset, our method demonstrates not only superior diagnostic accuracy compared to existing state-of-the-art approaches but also provides clinically interpretable insights into AD-related brain connectivity patterns. This work represents a significant advancement in multi-modal neuroimaging analysis for AD diagnosis, successfully combining enhanced diagnostic performance with clinical interpretability.

PMID:41335802 | DOI:10.1109/EMBC58623.2025.11253645

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-5. doi: 10.1109/EMBC58623.2025.11254461.

ABSTRACT

Tools for aiding in the diagnosis of Autism Spectrum Disorder (ASD) using machine learning (ML) and resting-state rs-fMRI (rs-fMRI) must encompass different phases such as data collection, preprocessing, feature extraction, model training, and validation. Many studies rely on a single preprocessing pipeline or use preprocessed data, which might not be optimal for the task at hand. This study investigates the impact of rs-fMRI preprocessing on the performance of ML models for ASD diagnosis. Using a subset of the Autism Brain Imaging Data Exchange (ABIDE) dataset, 72 subjects were preprocessed with 108 different configurations, and features were extracted to train 13 ML classifiers. Results indicate that preprocessing choices significantly influence model accuracy, with the best configurations achieving up to 95.83% accuracy. However, generalization tests on an extended dataset revealed a substantial performance drop, highlighting challenges in model robustness. Findings emphasize the need for adaptive preprocessing strategies and gender-balanced datasets to improve ASD classification reliability.

PMID:41335785 | DOI:10.1109/EMBC58623.2025.11254461

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-6. doi: 10.1109/EMBC58623.2025.11254951.

ABSTRACT

Functional near-infrared spectroscopy (fNIRS) is an optical imaging modality which, similar to fMRI, measures cerebral hemodynamics associated with neural activity. It has several advantages over fMRI, including low cost, portability, compatibility with metal or electrical medical implants, and ease of integration with electroencephalography (EEG) and transcranial magnetic stimulation (TMS). However, fNIRS signal contains a number of confounding components. Physiological noises due to superficial absorption by the scalp and skull are present in all fNIRS data. Additionally, low-frequency oscillations of respiration, cardiac pulse and movements all obscure the underlying cerebral hemodynamic signals. Our previous work has developed an automatic processing pipeline that effectively removes these physiological noise components from data during voluntary tasks (e.g., a motor task) and an endogenous state (e.g., awake resting) [1], [2]. However, to date it has not been known if the noises behave similarly in recordings involving an externally injected stimulus such as TMS. Therefore, in a unique setup of concurrent fNIRS, EEG and TMS (fNET), this study examined the spatial and temporal profiles of fNIRS data and noises during motor, single pulse and repetitive TMS. Specifically, we compared the multichannel short separation recordings with the regularly distanced long separation data in a whole-head montage. The results showed that superficial fluctuations indeed were present in the TMS-concurrent fNIRS recordings and that the noise components behaved similarly across motor task, single pulse and repetitive TMS at individuals' alpha frequency, which warrants removal of such physiological noises.Clinical Relevance- Compared to fMRI, fNIRS offers a much less expensive alternative for measuring cortical hemodynamics. Importantly, fNIRS can provide a clinic accessible options for concurrent measurement with TMS when TMS is given as treatment to patients with depression or other neurological disorders. Our findings indicate that fNIRS data acquired during concurrent TMS are contaminated by superficial fluctuations and that careful removal of these physiological noises from fNIRS data is critical in obtaining accurate images of cerebral activity with fNIRS.

PMID:41335688 | DOI:10.1109/EMBC58623.2025.11254951

Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-7. doi: 10.1109/EMBC58623.2025.11254960.

ABSTRACT

Diagnosing Parkinson's disease (PD) is particularly challenging due to the intricate and variable nature of its biomarkers, which span motor and non-motor symptoms, differ across individuals, and evolve over time. While machine learning has been used to automate this process, most studies focus on limited biomarkers due to dataset constraints. This study introduces a Federated Learning (FL) framework that integrates electroencephalography (EEG) and resting-state functional magnetic resonance imaging (rs-fMRI) data for improved PD classification. Unlike traditional fusion-based studies integrating multiple biomarkers from the same subject group, the Federated Learning framework processes EEG and fMRI data separately from distinct subject groups. Client nodes treat these as independent datasets and utilize convolutional neural networks (CNNs), Graph CNNs, and ResNet-18 models for analysis. A central server then aggregates insights, simulating a diagnostic center to evaluate the relevance of additional biomarkers for enhanced PD detection utilizing support vector machines (SVM) and federated dynamic model aggregation (Fed-Dyn). Additionally, gender-specific evaluations suggest that male-exclusive models outperform female models in biomarker representation. The study underscores the necessity of demographic-aware frameworks and optimized fusion techniques for early-stage PD detection.Clinical Relevance- This study significantly enhances early PD diagnosis by integrating EEG and fMRI biomarkers through Federated Learning (FL), offering a more comprehensive view of neurodegenerative changes while preserving patient privacy. By addressing gender-specific biomarker differences and tailoring models to diverse patient profiles and disease stages, it supports precision medicine and equitable healthcare. The advanced fusion techniques improve diagnostic accuracy in terms of ROC-AUC score, aiding clinical decision-making and enabling scalable telemedicine solutions. Beyond PD, the framework holds potential for broader neurodegenerative research and sets benchmarks for biomarker-based diagnostics, paving the way for impactful advancements in precision neurology.

PMID:41335685 | DOI:10.1109/EMBC58623.2025.11254960