Aug 2026· Medical Image Analysis· Vol 114, pp.
104242
· 0 citations· 44 references
Medicine
TL;DR
Results show that ENCORE consistently improves inter-subject correspondence of fine-grained connectivity features and yields higher accuracy in structural pathway localization compared with existing surface-based registration methods.
Abstract
Brain networks are typically represented by adjacency matrices, where each node corresponds to a brain region. In traditional brain network analysis, nodes are assumed to be matched across individuals, but the methods used for node matching often overlook the underlying connectivity information. This oversight can result in inaccurate node alignment, leading to inflated edge variability. To overcome this challenge, we propose a novel framework for registering high-resolution continuous connectivity (ConCon), defined as a continuous function on a product manifold space - specifically, the cortical surface - capturing structural connectivity between all pairs of cortical points. Leveraging ConCon, we formulate an optimal diffeomorphism problem to align both connectivity profiles and cortical surfaces simultaneously. We introduce an efficient algorithm to solve this problem and validate our approach using data from the Human Connectome Project (HCP). Results show that ENCORE consistently improves inter-subject correspondence of fine-grained connectivity features and yields higher accuracy in structural pathway localization compared with existing surface-based registration methods.
Accurate alignment of hippocampal anatomy across individuals remains challenging due to complex and highly variable folding patterns that are not well captured by conventional volumetric approaches. HippUnfold introduced a surface-based representation of the hippocampus, but key components—including coordinate estimation and inter-subject correspondence—were defined in the volumetric domain, making them susceptible to topological errors and interpolation artifacts. Here, we introduce a surface-intrinsic formulation of hippocampal unfolding in which geometry, intrinsic coordinates, and correspondence are defined directly on subject-specific surface manifolds. Intrinsic anterior–posterior and proximal–distal coordinates are computed by solving Laplace equations on the surface, and correspondence is established through surface-based resampling in unfolded space, replacing inverse volumetric warping. Relative to the original HippUnfold approach, this formulation improves test–retest consistency, subject identifiability, and mesh quality, while better preserving subject-specific gyral and sulcal morphology. Surface representations show reduced distortion between folded and unfolded spaces and eliminate misplaced or outlier vertices associated with volumetric warping. These improvements translate to enhanced sensitivity in a clinical application, improving lateralization of temporal lobe epilepsy. These results demonstrate that a surface-intrinsic formulation provides a principled and robust foundation for hippocampal unfolding, enabling topology-preserving alignment and more accurate characterization of inter-individual variability in health and disease.
J. DeKraker, D. Bansal, M. Snyder et al.· bioRxiv· 0 citations
The emergence of foundation models has revolutionized neuroimaging analysis, offering universal representations for decoding complex brain functions. However, existing fMRI-based foundation models predominantly rely on standard group-level atlases neglecting inter-subject heterogeneity. They often operate on voxel-wise time series with inherent low signal-to-noise ratios (SNR) and employ single-objective pre-training strategies that fail to capture fine-grained local structures and global discriminative representations. To bridge these gaps, we present BrainNetGFM, a novel graph-based foundation model for brain network construction that integrates individualized geometry with joint self-supervised learning, pre-trained on a massive dataset of 44,476 samples. First, we introduce an individualized brain atlas construction method utilizing region growing on functional priors to ensure precise alignment of subject-specific functional topology. Brain network graphs integrating functional connectivity strength with individualized spatial geometry is constructed. Second, we propose a joint self-supervised learning (SSL) framework that synergizes generative and contrastive paradigms. Specifically, we combine graph masked autoencoders (GMAE) to reconstruct masked node and edge features for learning fine-grained local structures, alongside multi-level graph contrastive learning (GCL) spanning both node and graph levels to enforce global discriminative representations. Extensive experiments on downstream tasks including the diagnosis of 8 brain disorders, age and cognition prediction, demonstrate that BrainNetGFM outperforms 9 state-of-the-art methods. Moreover, BrainNetGFM exhibits superior interpretability, identifying neurologically meaningful network biomarkers for clinical analysis. The code is publicly available at https://github.com/BrainLabA/BrainNetGFM.
Chunzhi Zhao, T. Adalı, Jing Sui et al.· Proceedings of the 32nd ACM...· 0 citations
Objective. Characterizing the relationship between structural connectivity (SC) and functional connectivity (FC) is a central problem in multimodal neuroimaging. SC-to-FC prediction accuracy is often used to summarize this relationship, but a high score may largely reflect population-level regularities shared across subjects and does not reveal whether regional, network-level, or subject-specific correspondence is preserved. We therefore developed a framework that makes these different levels of SC–FC correspondence explicit and directly testable. Approach. For each subject, we constructed separate spectral representations of SC and FC and estimated an explicit linear map between them. Signal transfer was evaluated against matched controls that disrupted node identity, within-network organization, or subject pairing, allowing aggregate transfer to be distinguished from more specific forms of correspondence. A training-derived reference map was further used to define the functional component expected under a reference SC–FC relationship and the deviation from that expectation for downstream aging and disease-related analyses. Main results. Across four independent datasets, all primary matched-control gaps were positive after false-discovery-rate correction, although their magnitudes varied across cohorts. Strong aggregate transfer did not necessarily imply strong subject-level correspondence. Compared with external baselines, the proposed framework more consistently preserved regional identity, within-network organization, and correct subject pairing. In aging analyses, reduced alignment with a young-reference SC–FC relationship was robust in CamCAN but not significant in NKI, whereas empirical FC magnitude showed no comparable decline. In Alzheimer’s Disease Neuroimaging Initiative, reference-conditioned markers showed exploratory disease-related signals, but the associated uncertainty did not support definitive diagnostic or incremental-value claims. Significance. The proposed framework separates the ability to transfer SC-derived signals into FC space from the specificity of the correspondence being preserved. It provides an interpretable basis for testing population-level, network-level, and subject-sensitive SC–FC relationships and for constructing reference-conditioned individual deviation measures without treating prediction accuracy as direct evidence of biological constraint.
Zhengyuan Lyu, Zhiwei Song, Si-Yue Fu et al.· Journal of Neural Engineerin...· 0 citations
This work introduces an alternative criterion based on the persistent topological cycles in which each node participates---a measure of mesoscale integration that captures features beyond local connectivity---and demonstrates that persistent topology captures information about brain network control that scalar energy summaries miss.
Carter Sale, Marco Coraggio, Mengsen Zhang et al.· 0 citations
Functional connectome analysis examines brain-region interactions to understand and identify disorders such as autism spectrum disorder and Alzheimer's disease. Existing methods typically use GNNs and Transformers to model the full functional connectivity matrix. However, processing tens of thousands of connections introduces redundancy and noise, increases computational cost, and limits connection-level interpretability. This raises a central question: do we really need complex interaction modeling, or is identifying a small set of disease-relevant connectivity patterns sufficient? To answer this question, we propose BrainLinear, a lightweight geometry-aware framework for mining disease-discriminative connectome patterns. BrainLinear first maps each functional connectivity matrix to a shared tangent space centered at the Fr\'echet mean of the training set, capturing subject-specific deviations while respecting matrix geometry. It then scores each ROI-pair tangent direction by its classification contribution and disease--control difference, retaining Top-$K$ directions as a compact representation. Finally, a shallow multilayer perceptron performs classification on the selected representation. Experiments on ABIDE and ADNI show that BrainLinear matches or exceeds strong GNN and Transformer baselines at a fraction of their cost: it improves AUC and ACC over the best baseline for each metric by up to $3.54$ and $1.39$ percentage points, while reducing runtime and peak GPU memory by $84.0\%$ and $68.4\%$ relative to the closest baseline in AUC. The selected directions are directionally consistent with between-group displacements and organized across major functional systems, supporting connection-level interpretation.
Sijing Wu, Dongyuan Li, Miaoting Huang et al.· 0 citations
Accurate inter-individual alignment of human cerebral cortex is challenging because of the high variability of human cortical folding patterns and the regionally non-uniform and inconsistent spatial relationships across individuals of cortical folds versus the functional networks and cortical areas that we wish to study. To achieve precise alignment across individuals, an algorithm must use multi-modal neuroimaging features related to cortical areas and functional networks, ideally as inputs to surface-based registration, but at least during training if only folding patterns will be available during inference, so that it can learn where folding patterns are trustworthy and learn a spatially non-uniform regularization function that reflects the true gamut of human inter-individual variability in cortical organization. Additionally, an algorithm ideally will be capable of denoising its own input registration features to avoid overfitting to noise, enabling reproducible registration in test-retest data, and will not require precise hand tuning of input regularization parameters. To address these challenges, we developed Deep-learning Registration Using U-Net with Multimodal Supervision (DRUMS), a novel framework for multi-modal cortical surface registration. DRUMS applies its deep-learning approach to cortical surface registration using multi-resolution spheres, a well-validated approach used in other registration algorithms such as Multi-modal Surface Matching (MSM) (Robinson, et al., 2014; Robinson, et al., 2018). It also includes the same physically inspired strain energy regularization that we pioneered for MSM and the same precise barycentric interpolation on spherical surface meshes. DRUMS has a three-stage architecture: (1) multiscale feature extraction, (2) multiscale feature integration, and (3) deformation field generation. The framework’s multimodal design supports flexible integration of diverse imaging modalities, enabling registration using either folding features or multi-modal features as inputs with independent control over supervision during training (e.g., training DRUMS with folding inputs and multi-modal supervision to learn which folds best correlate with multi-modal features). DRUMS outperforms Multi-modal Surface Matching (MSM), the current state-of-the-art method used in the Human Connectome Project (HCP) pipelines, over a wide range of input regularization settings in both registration accuracy and test-retest reproducibility of registration results for both supervised folding registrations and multi-modal registrations across all tested modalities, including held out modalities. DRUMS further learns a biologically plausible spatially non-uniform regularization function, with registration induced distortion correctly matched to known human inter-individual cortical variability. These results suggest that DRUMS should replace MSM in the HCP Pipelines and position DRUMS as a versatile and reliable tool for cortical surface registration in neuroimaging research.
Unknown authors· bioRxiv· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
