Learning meaningful latent representations from nonlinear fMRI data remains a fundamental challenge in neuroimaging analysis. Traditional independent component analysis, widely used due to its ability to estimate interpretable functional brain networks, relies on a linear mixing assumption for latent sources, limiting its ability to capture the inherently nonlinear and complex organization of brain dynamics. More recently, deep representation learning methods have emerged as promising alternatives for modeling nonlinear latent structure. However, many of these approaches have been evaluated primarily on simulated datasets or natural image benchmarks, with comparatively limited validation on real-world neuroimaging data such as fMRI. In this work, we are motivated by the β-TCVAE (Total Correlation Variational Autoencoder), a refinement of the β-VAE framework for learning latent representations without introducing additional hyperparameters during training. We adapt and modify this model to fMRI data for nonlinear source disentanglement, aiming to separate mixed spatial and temporal brain signals into interpretable components. We show that the β-TCVAE framework can recover meaningful nonlinear spatial components with biological relevance, including well-established intrinsic connectivity networks such as the default mode network. Furthermore, we evaluate the learned representations using functional network connectivity, showing that the latent structure captures coherent and interpretable brain organization patterns. This study provides a pilot investigation that bridges nonlinear representation learning and fMRI analysis.

High dimensionality of dynamic functional connectivity (dFNC) data representation complicates clinical interpretation and biomarker discovery. We propose a new complementary analytical framework based on dynamic inter-network connectivity entropy (DICE) and its potential use as a biomarker of mental illness. Our framework shows that DICE features extend beyond patient-control discrimination, revealing distinct pathophysiological signatures and differential associations with symptom dimensions. Using resting-state fMRI data from 311 participants, 160 controls, 151 schizophrenia (SZ) patients, we identified 53 intrinsic networks, computed DICE and derived three families of DICE-based metrics: (i) entropy level and range, (ii) distributional shape and temporal organization, (iii) entropy-state repertoire and occupancy. These measures revealed a multidimensional signature of altered entropy dynamics in SZ: (1) elevated baseline entropy with reduced fluctuation magnitude and reduced entropy acceleration; (2) reduced temporal persistence of entropy excursions and entropy distributions closer to Gaussian; and (3) a narrowed repertoire of entropy states, prolonged time in near-baseline entropy configurations. The DICE-based metrics within the SZ group show different associations with symptom dimensions. Reduced fluctuation magnitude and acceleration were associated with greater PANSS general symptom severity (disturbance of volition and preoccupation). Reduced deviation from Gaussianity was associated with higher PANSS positive severity (delusions and hallucinations). Reduced temporal persistence was associated with multiple PANSS positive, negative, and general symptoms. Reduced entropy-state diversity and prolonged dwell time in near-baseline states were associated with depression and PANSS positive/general severity, respectively. The multidimensional pathophysiology revealed through the different entropy patterns may potentially guide biomarker development and personalized treatments.

Dynamic connectivity is central to understanding time-varying interactions between brain regions. Despite decades of methodological development, approaches to measuring dynamic connectivity remain fragmented, leading to inconsistent findings, limited comparability across studies, and difficulty attributing observed effects to computational choices. Here we introduce dynamic co-modulation (DyCoM), a compact operator-level framework that expresses dynamic connectivity estimators as compositions of a small set of fundamental signal processing operations. Using simulations and resting-state fMRI data, we show that DyCoM disentangles previously conflated findings by revealing that lower-order sensory and higher-order executive control neurobiological signatures, state-transition sensitivity, and medication-linked clinical associations arise from distinct operator choices within a single unified framework. Together, these results establish DyCoM as a unifying foundation for dynamic interaction analysis, revealing how differences in estimator design give rise to divergent biological interpretations and offering a principled, domain-agnostic framework for coherence, interpretability, and estimator development.

Functional MRI (fMRI) and structural MRI (sMRI) offer complementary insights into brain function and anatomy, but their integration for schizophrenia identification remains challenging due to modality heterogeneity. Many existing methods fall short of effective modeling of the interaction between two modalities. We propose CAMF, a Cross-Attentive Multi-modal Fusion framework that employs self-attention to capture intra-modal patterns and cross-attention to learn inter-modal relationships. In addition, we introduce the gradient-guided score-class activation map to enhance interpretability by highlighting salient features. Our approach significantly improves the accuracy in classifying schizophrenia, as demonstrated by the evaluation of multi-modal brain imaging datasets from four cohorts of schizophrenia studies. Furthermore, the model identifies functional networks and anatomical regions aligned with established biomarkers. CAMF provides an accurate and interpretable framework for multimodal brain imaging analysis, offering new insights into schizophrenia-related alterations.

Diagnostic MRI evaluation of temporal lobe epilepsy (TLE) depends on the subjective visual interpretation of MRI images. These interpretations could be enhanced by quantitative artificial intelligence (AI) support tools. Humans often make sequential and conditional decisions during their radiological interpretations, such as whether an abnormality is present and, if present, characterizing the abnormality. It is not known whether it is superior to train AI to treat every decision separately in a similar step-wise manner or to train a model holistically on all decisions simultaneously. Here, we analysed three large epilepsy MRI datasets [n = 3676, 2320 people with epilepsy and 1356 healthy controls (HC)] to perform two tasks: (i) establish the presence of a TLE pattern on MRI and (ii) determine TLE pattern lateralization. We compared Step-wise models that independently classify TLE versus HC and lateralize patients as left TLE (L-TLE) or right TLE (R-TLE), against a simultaneous model trained to distinguish all three classes in a single step. To do this, 3D volumetric T1-weighted images were input into an EfficientNetV2 model multiple times to ensure reproducibility of results. Class prediction, model classification confidence and saliency maps were output for interpretability. Step-wise models outperformed the Simultaneous model on both tasks (both Ps < 0.001), with an average ∼2.8% accuracy increase for discriminating HC from TLE and an average 12.7% accuracy increase for distinguishing L-TLE from R-TLE. For both the Step-wise and Simultaneous models, important features discriminating TLE from HC included the known TLE limbic pattern involving the hippocampus, parahippocampal cortical regions, cingulate cortex and lateral temporal regions. However, there was less concordance between the Step-wise and Simultaneous models for the L-TLE versus R-TLE task (all Fisher's Zs > 10.5, Ps < 0.001); the Step-wise model focused less on subcortical regions such as the thalamus and hippocampus and focused more on distributed cortical pathology. Across the two Step-wise models, 95.1% of TLE patients had accurate classifications in either HC versus TLE and/or L-TLE versus R-TLE tasks. These results included 69.6% of patients being both correctly labelled as TLE and lateralized, 13.9% being correctly labelled TLE but lateralized incorrectly and 11.6% being lateralized correctly but not detected as TLE. These findings provide evidence that diagnostic tasks with simpler, Step-wise AI models may enhance diagnostic performance and interpretability in clinical workflows. Future AI clinical support tools can leverage this step-wise approach in the early identification of TLE-related structural patterns, supporting timely diagnosis and treatment decisions.

Comparative mapping of functional and structural homologies across humans, small animals, and nonhuman primates has been extensively pursued due to its strong translational relevance. However, these experimental models possess inherent limitations in fully recapitulating the complexity of human cortical organization. The porcine model has recently emerged as a promising alternative, given its neuroanatomical and physiological similarities to the human brain. Despite these advantages, systematic cross-species characterization of functional and structural homologies between humans and pigs remains largely understudied. In the present study, we acquired resting-state functional MRI and diffusion MRI data from pigs and analyzed them alongside corresponding human datasets to investigate cross-species correspondence in large-scale brain organization. First, to enhance functional network alignment across species, group independent component analysis was performed separately within each species to identify intrinsic large-scale functional networks. Our results demonstrated that multiple canonical human resting-state networks are represented in the porcine brain, including sensorimotor, default mode, cerebellar, frontal, and central executive networks. Moreover, we observed significant cross-species concordance in intrinsic functional architecture across multiple distributed networks, both in spatial distribution and temporal patterns, indicating homologous large-scale brain organization between pigs and humans. Second, we conducted comparative structural analyses using tractography derived from diffusion MRI and color-encoded fractional anisotropy maps to examine white matter geometry in pigs and humans. Cross-species comparison revealed substantial similarities in major white matter pathways and their spatial organization, supporting structural correspondence at the level of tract geometry. Together, these findings underscore the translational value of the porcine model as a robust and neurobiologically relevant platform for investigating human brain function, structural organization, and related neurological disorders.

Background: Transcriptomic gene signatures are widely used to infer pathway activity and biological mechanism from bulk cancer expression data, yet current evaluation strategies primarily emphasize internal coherence, predictive performance, or scoring robustness. A quantitative framework for assessing how much signature variation remains independent of background expression structure has been lacking. Results: Unlike existing single-number signature-quality metrics such as Berglund uniqueness, residual-ratio auditing reports a trajectory across null-model richness: for each signature we compute the residual ratio resratio(k) = 1 - sum_{j=1}^{k}(mathbf{q}_j^top mathbf{h})^2 at progressively enriched expression-PC subspaces, together with an inverse-participation-ratio (IPR) concentration diagnostic that reports the effective number of axes absorbing each signature. Applied to a curated 17-entry benchmark, all 50 MSigDB Hallmark gene sets, and 1181 Reactome pathways across 8 TCGA cancer types (4462 samples), with external validation in METABRIC, the framework produces two complementary readouts. First, the curated panel is absorbed into the ExprPC50 subspace at residual ratios 18--43% below size-matched random 30-gene baselines in every cancer (curated mean resratio range 0.109--0.177 vs. random mean 0.18--0.288), providing the framework's central quantitative discrimination between biologically coherent signatures and arbitrary gene combinations. Second, within the curated panel the ExprPC50 residual ratio is negatively correlated with the top-5 absorption concentration in every cancer (Spearman rho from -0.59 in PRAD to -0.89 in SKCM, median -0.71; all 8 significant at p < 0.05, most at p < 10^-3); we report this correlation as a descriptive geometric property of the null-model coordinate system rather than as a biological law, because 1000 random 30-gene draws projected through the same top-50 expression-PC basis reproduce the same pan-cancer median rho (-0.73; Supplementary Table~ref{tab:S17}), and it is robust to compositional nuisance: after rebuilding the null basis as immune-PC1 oplus stromal-PC1 oplus proliferation-PC1 plus 47 residual PCs, the per-cancer rho becomes more negative rather than shallower (median -0.86; Supplementary Table~ref{tab:S18}), ruling out tumor purity, immune infiltrate, and stromal fraction as drivers of the pattern. Because absorption at ExprPC50 is a geometric property of how any signature direction sits in expression-PC space, tier-level distributional structure at this operating point is not separable beyond the low-vs-upper band split: a Kruskal--Wallis omnibus is significant (p = 4.9 x 10^-13), but pairwise Dunn's post-hoc tests show that Tiers~1, 4, and~5 are not separable (p_{\mathrm{BH}} > 0.2). The trajectory shape itself is empirically bootstrap-invariant: across 200 sample-level fixed-basis bootstrap resamples of the 17 curated entries in BRCA, the mean pairwise Pearson correlation of trajectory-shape vectors is 0.999, and individual cell-level 95% bootstrap CI half-widths at B = 1000 resamples are in the range 0.002--0.053. External replication in the METABRIC breast cancer cohort (n_{\text{samples}} = 1980, microarray) showed moderate-to-strong rank-ordering concordance with TCGA-BRCA across the 17 curated entries (Spearman rho = 0.72 on the 17-signature ordering, 95% Fisher-z CI 0.37--0.89, p = 0.001). Under an upper-bound sensitivity analysis, 45 of 50 Hallmark gene sets and 992 of 1181 Reactome pathways had ExprPC200 residual ratios below the mean of their size-matched random baselines---a descriptive statistic reflecting axis alignment under rich null models, not a failure rate. In causal DAG simulations (n_{mathrm{rep}} = 100 replicates), a signature driven entirely by a latent confounder retained resratio = 0.233 at ExprPC50, numerically comparable to Tier~1 validated drivers, so a single-point residual ratio cannot adjudicate confounder-independence. The framework's load-bearing signals are therefore the trajectory shape (statistically invariant under sample-level resampling) and the magnitude gap between the curated panel and its random 30-gene baseline (the curated-vs-random discrimination), read jointly---not the value of resratio at any single null-model dimensionality. Conclusions: Residual-ratio auditing provides an interpretable and practical framework for quantifying how much of a transcriptomic gene signature's variance remains orthogonal to a chosen background-expression model. The two statistically reliable quantities it reports are (i) the shape of the trajectory resratio(k) across null-model richness, which is bootstrap-invariant across sample-level resamples, and (ii) the magnitude gap between the curated panel's residual ratio and size-matched random 30-gene baselines at a fixed operating point, which is 18-43% in all 8 TCGA cancers and survives a purity-aware null-model construction. The negative correlation between resratio and the top-5 absorption concentration $c$ (curated-panel median rho = -0.71) is reproduced by random 30-gene sets under the same basis (random-draw median rho = -0.73) and is therefore best read as a descriptive geometric property of the null-model coordinate system rather than a biological discovery about curated signatures. Any single operating-point residual ratio carries materially wider cell-level uncertainty than the trajectory shape and cannot, on its own, adjudicate confounder-independence. The framework's outputs describe a signature's geometric relationship to modeled background expression structure and do not evaluate clinical utility: a signature with a low residual ratio may still be clinically valuable when that low value reflects alignment with a strong prognostic or actionable program such as proliferation, immune infiltration, or cell cycle, and the framework is not a substitute for calibrated prognostic or predictive classifiers. All findings are based on bulk RNA-seq (TCGA PanCancer Atlas, 8~cancer types) and microarray (METABRIC) data; transfer to single-cell, single-nucleus, or spatial transcriptomics is out of scope and not claimed. Used within this scope---reading the trajectory shape and the magnitude-gap signal jointly, rather than the value of resratio at any one k---the framework adds a complementary audit layer to existing pathway-scoring and experimental-validation workflows, and supports more calibrated interpretation, comparison, and reporting of transcriptomic gene signatures in cancer studies.

Schizophrenia is a multidimensional psychiatric disorder lacking a unifying systems-level framework. We introduce a cortical brain entropy architecture that characterizes the spatial and hierarchical organization of functional entropy across the cerebral cortex. In large-scale neuroimaging data, this architecture differentiates schizophrenia from health and captures multidimensional network variation. Entropy alterations follow structured cortical organization, revealing coordinated disruptions within association networks. These findings identify cortical brain entropy architecture as a compact systems-level marker of schizophrenia heterogeneity and establish a quantitative link between macroscale cortical entropy and clinical symptom dimensions.

Understanding the dynamic nature of brain connectivity is critical for elucidating neural processing, behavior, and brain disorders. Traditional approaches such as sliding-window correlation (SWC) characterize time-varying undirected associations but do not resolve directional interactions, limiting inference about time-resolved information flow in brain networks. We introduce sliding-window prediction correlation (SWpC), which embeds a directional linear time-invariant (LTI) model within each sliding window to estimate time-varying directed functional connectivity (FC). SWpC yields two complementary descriptors of directed interactions: a strength measure (prediction correlation) and a duration measure (window-wise duration of information transfer). Using concurrent local field potential (LFP) and fMRI BOLD recordings from rat somatosensory cortices, we demonstrate stable directionality estimates in both LFP band-limited power and BOLD. Using Human Connectome Project (HCP) motor task fMRI, SWpC detects significant task-evoked changes in directed FC strength and duration and shows higher sensitivity than SWC for identifying task-evoked connectivity differences. Finally, in post-concussion vestibular dysfunction (PCVD), SWpC reveals reproducible vestibular-multisensory brain-state shifts and improves healthy-control vs subacute patient (HC-ST) discrimination using state-derived features. Together, these results show that SWpC provides biologically interpretable, time-resolved directed connectivity patterns across multimodal validation and clinical application settings, supporting both basic and translational neuroscience.

Background Functional magnetic resonance imaging (fMRI) has been used to characterize functional brain networks in disorders ranging from depression to schizophrenia. Like fMRI, single photon emission computed tomography (SPECT) is a technique which captures information about neurally linked blood flow activity through radioactive tracers. While a few SPECT studies in schizophrenia populations have been conducted along with fMRI based studies, research on SPECT data for individual subject classification is limited. We used an independent component analysis (ICA) approach to estimate covarying SPECT networks from our prior study. Results were then fed as input to a classifier model to evaluate accuracy of individual diagnostic prediction. Methods 213 subjects (137 schizophrenia patients and 76 healthy controls) were used for the analysis. Classification input was based on loading parameters generated from spatially constrained ICA using a set of network priors derived from fMRI. Fifty-three SPECT components were estimated guided by the NeuroMark fMRI 1.0 template. We initially focused on a support vector machine (SVM) classifier given previous favorable fMRI-SVM results. We also evaluated performance of multiple classifiers post hoc. Results and conclusion Surprisingly, linear SVM performed worse compared to random forest, logistic regression, voting, and multilayer perceptron. Linear SVM had a greater number of false negatives compared to the other classifier approaches. By contrast, random forest and logistic regression performed the highest, with an 88% sensitivity and 61% specificity (random forest), and 87% sensitivity and 68% specificity (logistic regression). These results demonstrate that other approaches such as random forest and logistic regressions should be further investigated for future SPECT studies, and coupled with sc-ICA, provide an improved understanding of aberrant functional networks in schizophrenia.

Natural environments (e.g., forests, parks, neighborhood greenery) may support adolescent mental health, but whether such associations generalize across diverse U.S. populations remains unclear. The Adolescent Brain Cognitive Development (ABCD) Study provides an opportunity to test these associations in a large, demographically diverse national sample (>11,000 adolescents). We evaluated five baseline (age 9–10) nature measures from ABCD Release 5.1 using the PRIGSHARE (Preferred Reporting Items for Greenspace and Health Research) framework to assess accuracy, quality, and relevance. Two measures—percent tree canopy and park cover at the census tract level—were identified as high-quality and epidemiologically appropriate. Mental health was assessed using parent-reported outcomes from the Child Behavior Checklist (CBCL). Nature data were available for 11,164 participants (47.65% female). After covariate adjustment, adolescents living in areas with greater tree canopy demonstrated modestly better mental health, including lower internalizing (β = –0.03, 95% CI: –0.052, –0.015, p < .001) and externalizing (β = –0.02, 95% CI: –0.040, –0.003, p = .026) problems. Secondary analyses indicated modestly lower somatic complaints, social problems, depression, and anxiety symptoms. This study introduces a framework for evaluating linked-nature measures and offers preliminary evidence of weak beneficial associations with adolescent mental health across the United States.

Background Adolescence is a critical development stage, and adverse childhood experiences (ACEs) are important environmental factors influencing behavior. However, the relationship between ACEs, individual behavior, brain mechanisms, and the role of genetic risk remains underexplored. Methods We analyzed data from 9,796 participants aged 9–13 years from the Adolescent Brain Cognitive Development (ABCD) study. Sixty-seven ACE-related variables covering caregiving disruption, family functioning, neighborhood environment, and safety were examined. Sparse canonical correlation analysis (sCCA) and multiple sparse canonical correlation analysis (msCCA) identified multivariate ACE–behavior–brain patterns. Moderated mediation was used to test genetic risk modulation, and a cross-lagged panel model examined longitudinal associations. Results Two robust modes emerged. Mode 1 r eflected parental mental health difficulties linked to child mental health ( r = 0.576, P perm < 0.001). Mode 2 represented socioeconomic and family environment factors associated with cognitive performance ( r = -0.272, P perm < 0.008). Neuroimaging revealed convergent alterations in hippocampal and cerebellar volumes, white-matter integrity, and large-scale functional networks. Polygenic risk for psychiatric disorders significantly moderated ACE–brain–behavior pathways, with effects varying across ancestry groups. Longitudinal analyses showed that parental depression and ADHD strongly predicted youth psychiatric symptoms two years later. Conclusions Two distinct profiles of ACEs influence adolescent behavior through neuroimaging pathways, and these associations are further shaped by genetic risk. Findings advance mechanistic understanding of ACE-related vulnerability and highlight intervention targets for at-risk youth.

Individual brain parcellation is vital for understanding individual variability and advancing precision neuroscience. However, existing deep learning parcellation methods often lack principled seed selection and overlook neuroscientific priors such as inter-hemispheric homotopy. To address these limitations, we propose a submodular-homotopic atlas parcellation encoder (SHAPE) with contrastive learning. To address seed selection, SHAPE applies submodular optimization on supervoxels to identify homotopic seed pairs, ensuring cross-subject correspondence. To integrate neuroscientific priors, SHAPE introduces a homotopic contrastive learning strategy within a graph convolutional network, embedding inter-hemispheric symmetry into node representations. A competitive region growing algorithm is used to generate the final individualized parcellations. Results show that SHAPE outperforms existing methods in functional homogeneity, clustering validity, and sensitivity to individual variability across multiple parcellation scales. Ablation studies confirm the contributions of key components. Overall, SHAPE establishes a principled and neuroscientifically grounded framework for individualized brain parcellation, bridging the gap between data-driven learning and biological plausibility.

Hemodialysis (HD) is the predominant treatment for end-stage renal disease (ESRD). Despite the efficacy of HD, the neurobiological underpinnings underlying high-risk complications remain unclear. In this study, using unsupervised fusion of functional and structural MRI, we identified a longitudinally altered default mode network (DMN)-insula pattern in ESRD receiving HD over 1-year follow-up (n = 39). This pattern was associated with cognition, and its related genes were enriched in biological processes involving DNA damage and repair, energy metabolism, and cellular activation. The baseline DMN-insula pattern demonstrated potential predictive value for follow-up cognition in ESRD. More importantly, these brain-cognition associations were validated in independent high-risk complications cohorts, including major depressive disorder (n = 60), mild cognitive impairment (n = 291), and Alzheimer’s disease (n = 77) by extracting the corresponding brain features and assessing their correlations with cognition. Collectively, this study may help researchers better understand the underlying mechanisms of ESRD receiving HD from a multimodal neuroimaging and molecular perspective.

Major Depressive Disorder (MDD) involves large-scale brain network disruption at rest. Canonical zero-lag functional connectivity methods often miss temporal offsets (or "lags") in interactions. Lag-adjusted functional connectivity captures intrinsic neural timescales (INTs) and directional signaling, offering a more sensitive framework to characterize network-level alterations. Here, we applied the NeuroMark framework to resting-state scans from 235 MDD and 284 healthy controls, identifying 105 intrinsic connectivity networks (ICNs) and their time series. To enable sub-TR estimation, time series were upsampled to 100 ms resolution. Lag-adjusted connectivity was computed as the maximal cross-correlation for each ICN pair within a ±2s window sampled at 0.1s intervals. Group differences were assessed using a regression model. Significant differences emerged between groups (p<0.05). Specifically, MDD revealed hyperconnectivity in salience-sensorimotor and sensorimotor-temporoinsular networks, alongside hypoconnectivity in salience-higher cognitive temporal and frontal networks and temporoparietal-visual systems, indicating altered coordination among sensory, emotional, and cognitive processes. An exploration of the lags revealed a non-random bias in the temporal ordering of networks operating at different INTs. This was characterized by earlier relative cortical coupling in MDD, suggesting compressed inter-network timing. These findings underscore the utility of lag-adjusted approaches for detecting impaired neural coordination, beyond alterations in connectivity strength.

Single photon emission computed tomography (SPECT) is a highly specialized imaging modality that enables measurement of regional cerebral perfusion and, in particular, regional cerebral blood flow (rCBF). Recent technological advances have improved SPECT quantification and reliability, making it increasingly useful for studying rCBF abnormalities and perfusion-network alterations in psychiatric and neurological disorders. To characterize large-scale functional organization in SPECT data, data-driven decomposition methods such as independent component analysis (ICA) have been used to extract covarying perfusion patterns that map onto interpretable brain networks. Blind ICA provides a data-driven approach to estimate these networks without strong prior assumptions. More recently, a hybrid approach that leverages spatial priors to guide a spatially constrained ICA (sc-ICA) have been used to fully automate the ICA analysis while also providing participant-specific network estimates. While this has been reliably demonstrated in fMRI with the NeuroMark template, there is currently no comparable SPECT template. A SPECT template would enable automatic estimation of functional SPECT networks with participant-specific expressions that correspond across participants and studies. The current study introduces a new replicable NeuroMark SPECT template for estimating canonical perfusion covariance patterns (networks). We first identify replicable SPECT networks using blind ICA applied to two large sample SPECT datasets. We then demonstrate the use of the resulting template by applying sc-ICA to an independent schizophrenia dataset. In sum, this work presents and shares the first NeuroMark SPECT template and demonstrating its utility in an independent cohort, providing a scalable and robust framework for network-based analyses.

Reservoir expansion can improve online independent component analysis (ICA) under nonlinear mixing, yet top-n whitening may discard injected features. We formalize this bottleneck as reservoir subspace injection (RSI): injected features help only if they enter the retained eigenspace without displacing passthrough directions. RSI diagnostics (IER, SSO, ρ_x) identify a failure mode in our top-n setting: stronger injection increases IER but crowds out passthrough energy (ρ_x: 1.00→0.77), degrading SI-SDR by up to 2.2 dB. A guarded RSI controller preserves passthrough retention and recovers mean performance to within 0.1 dB of baseline 1/N scaling. With passthrough preserved, RE-OICA improves over vanilla online ICA by +1.7 dB under nonlinear mixing and achieves positive SI-SDR_sc on the tested super-Gaussian benchmark (+0.6 dB).

Schizophrenia is one of the most complex brain disorders, arising from multidimensional pathophysiological processes that span genetic vulnerability, neurotransmitter dysregulation, structural brain damage, and large-scale brain network dysfunction. Large-scale neuroimaging studies have consistently demonstrated the critical role of frontal brain regions in schizophrenia. Despite substantial progress, the precise localization of structural damage within these regions and the neurobiological mechanisms linking such alterations to disease pathology remain poorly understood. In this study, we included a total of 115 subjects from two sites of the B-SNIP dataset, comprising 60 healthy controls and 55 individuals with schizophrenia. We employed diffusion tensor imaging (DTI) to precisely characterize specific structural alterations in the frontal brain regions associated with schizophrenia. Our findings reveal significant microstructural abnormalities in the forceps minor, a major commissural white-matter tract that serves as a critical interhemispheric bridge between the bilateral frontal lobes. Network-level mapping further demonstrates that the forceps minor is closely integrated with large-scale brain networks, particularly the default mode network, and maintains strong structural connectivity with orbitofrontal regions-both of which are known to exhibit dysfunction in schizophrenia. Moreover, converging evidence suggests that the forceps minor plays an important role in the regulation of social behavior, a core domain of impairment in schizophrenia. Collectively, these findings identify the forceps minor as a promising structural imaging biomarker for schizophrenia and provide novel insights into the microstructural mechanisms underlying the disorder.