The research, led by Harrison Watters of the Czech Academy of Sciences and Weber State University, together with colleagues at Universiti Putra Malaysia and Georgia Tech, applied two sophisticated dynamic functional MRI analyses to the resting brains of people with high versus low scores on the Smartphone Addiction Scale, Malay version (SAS-M). Unlike traditional static functional connectivity, which averages relationships between brain regions across an entire scan, these methods capture how network relationships fluctuate, recur, and reorganize from second to second. The distinction matters because attention is inherently temporal, and phase information, the timing offset between activations of different networks, may reveal more about cognitive state than average activity ever can.

The first technique, functional connectivity-based attractor networks (fcANN), treats the brain as a Hopfield network whose activity settles into stable basins, or attractors, defined in a pretrained energy landscape learned from a large independent dataset. Each moment of a participant’s 13-minute resting-state scan was assigned to one of four canonical attractor states organized along two interpretable axes: internal versus external context, and action versus perception. The second technique, complex principal component analysis (cPCA), identifies quasi-periodic patterns (QPPs), infraslow swells of network activity lasting roughly 20 seconds, in which the default mode network (DMN) and the dorsal attention network (DAN) repeatedly fall into antiphase cycles. The DMN is associated with mind wandering and internal thought, while the DAN mediates goal-directed external attention, and the frontoparietal network (FPN) supplies executive control over both.

The dataset comprised 45 participants, five of whom were excluded for excessive head motion or registration failure, leaving 18 individuals with High SAS-M scores and 22 with Low SAS-M scores. All were right-handed, neurologically healthy, and screened with a previously validated SAS-M threshold above 98, a cutoff for elevated risk of problematic smartphone use. Critically, the authors note, the SAS-M is a self-report screening instrument rather than a clinical diagnostic, so the groups are best described as high and low scorers rather than addicted versus non-addicted individuals.

When the researchers examined attractor-state occupancy, the two groups looked remarkably similar. Neither the fraction of time spent in each state, nor dwell times, entropy, or switching rates differed significantly, although the high-scoring group showed a non-significant tendency to spend less time in externally oriented states (22.5 percent versus 27.6 percent) and to linger briefly longer in internal contexts. In other words, people with high smartphone addiction scores did not obviously inhabit a different set of brain states; they navigated the same landscape, but something about the fine texture of the activity within those states appeared different.

That something emerged most clearly from the QPP analyses, particularly around moments when the brain transitioned between internal and external attractor states. Immediately before and after these transitions, high-scoring participants showed significantly greater separation between DMN and DAN activity, an effect that survived false discovery rate correction at lags of one image volume (three seconds) on either side of the transition. Cluster-based permutation tests confirmed extended DMN-DAN differences spanning several seconds around the shift, as well as a DMN-FPN difference stretching from one second before the transition to five seconds after it. The pattern amounts to a form of default mode dominance: during periods of cognitive-state reorganization, the internally oriented network overwhelms the attention and control systems that should be reallocating resources.

The authors interpret this through the lens of attentional control theory, which holds that efficient shifting between internal and external focus depends on the limited resources of the DAN and FPN. Constant switching between salient smartphone stimuli, notifications, feeds, and apps, could plausibly place sustained demands on exactly these executive systems. The increased DMN dominance observed in high scorers may therefore reflect reduced efficiency in reallocating attention during shifts in cognitive context, though the researchers are careful to stress that this interpretation is speculative and cannot be directly inferred from resting-state data alone.

Corroborating the categorical results, continuous analyses of individual SAS-M scores found positive associations between smartphone addiction scores and DMN-FPN separation around external-to-internal transitions, with several Spearman correlations surviving multiple-comparison correction. Together with prior work showing that DMN-FPN decoupling during tasks tracks cognitive demand, the findings hint that heavy smartphone users may carry a task-like strain into their resting brains, their network dynamics resembling those of people under elevated cognitive load.

The study’s authors are candid about its limits. The sample of 40 is modest, several moderate effects did not survive correction, and the subject-level QPP estimation relied on relatively short scans, which can destabilize spatiotemporal estimates. The QPP component selection procedure, though applied identically and blind to group membership, partly built DMN-DAN opposition into the measure being tested. Self-report questionnaires are an imperfect proxy for actual usage, and the cross-sectional design rules out any claim about causality: it remains possible that these dynamics predispose people to problematic use rather than resulting from it, or that they represent the brain’s adaptive plasticity rather than damage, much as videogamers show altered but improved attentional performance.

Still, the paper makes a broader case that extends well beyond smartphone research. Because objective measures of smartphone use are rarely collected in neuroimaging datasets, heavy digital engagement may be quietly injecting heterogeneity into brain-imaging studies of all kinds. The authors propose that large consortia such as the Human Connectome Project and UK Biobank begin logging screen time and social media behavior alongside age, sleep, and psychiatric history, and they suggest that future studies combine longitudinal designs, objective usage metrics, and task paradigms to determine whether the transition-linked dynamics observed here are stable traits or reversible adaptations. As the authors conclude, understanding how digital environments reshape intrinsic brain dynamics will matter increasingly for both basic neuroscience and mental health, and the present findings, while preliminary, offer one of the first dynamic glimpses of what habitual smartphone engagement may be doing to the brain’s switching machinery.

Subject of Research: Dynamic resting-state fMRI of attention network coordination in high smartphone addiction scores

Article Title: Altered attention network dynamics in resting-state fMRI of individuals with high smartphone addiction scores

Article References: Watters, H., Rashid, A. A., Suppiah, S., LaGrow, T. J., Hlinka, J., & Keilholz, S. (2026). Altered attention network dynamics in resting-state fMRI of individuals with high smartphone addiction scores. Frontiers in Neuroscience, 20, Article 1948718. https://doi.org/10.3389/fnins.2026.1948718

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1948718

Keywords: smartphone addiction, resting-state fMRI, default mode network, dorsal attention network, frontoparietal network, quasi-periodic patterns, attractor states, dynamic functional connectivity, attentional control, neuroimaging, SAS-M, brain dynamics

The work, led by C. S. Smitha and S Sathiya of Annamalai University in Tamilnadu, India, together with T Soumya of the College of Engineering Muttathara in Thiruvananthapuram, Kerala, focuses on three of the most burdensome neurological conditions: Alzheimer’s disease, epilepsy, and Parkinson’s disease. Each of these disorders is marked by subtle, progressive changes in how brain regions activate and communicate with one another. Catching those changes early, and grading them accurately, matters enormously for clinical decision-making and personalized treatment planning, yet the task has stubbornly resisted automation because fMRI data are extremely high-dimensional, brain connectivity patterns shift from moment to moment, and no two brains are wired quite the same way.

The researchers’ answer is a dual-stream architecture, meaning the model processes the same input through two parallel computational pathways before merging what each has learned. The first stream, called the AdaBelief Boosted Contrastive DenseConvo net, or ABCD Net, is built on a 3D DenseNet backbone, a convolutional architecture in which each layer receives feature maps from all preceding layers, encouraging rich reuse of spatial information across the volumetric brain image. On top of that backbone, the team added Temporal Contrastive Learning, a technique that teaches the network to distinguish meaningful changes in neural activity over time from irrelevant fluctuations, and ConvLSTM fusion, which blends convolutional feature extraction with long short-term memory cells that can retain information across successive time frames.

Where the first stream asks what is happening inside the brain’s tissue, the second asks how the brain’s regions are talking to each other as the scan unfolds. This stream, named the Dynamic Slepian Gated Recurrent Transformer, or DynSGRT, constructs windowed representations of dynamic functional connectivity, the time-varying statistical relationships between activity in different brain areas. Rather than treating connectivity as a single static snapshot, the model slices the scanning session into windows and builds a graph for each one, with brain regions as nodes and their temporal correlations as edges. Slepian basis functions, which are optimally concentrated in both time and frequency, help represent these dynamic signals compactly, while gated recurrent units and attention mechanisms track how the connectivity patterns evolve.

Those windowed connectivity graphs are then handed to a third component, the Hierarchically differentiated temporal-evolving graph neural network, or HD-TEG Net. Graph neural networks are a class of deep learning models designed to operate on data structured as networks rather than grids, and they are a natural fit for brain connectivity, which is inherently a network problem. HD-TEG Net applies graph pooling, a method for progressively coarsening a graph into higher-level summaries, and temporal evolution modeling, which captures how the pooled graph representations change from one window to the next. In effect, the second stream learns a compressed story of how the brain’s communication network reorganizes itself over the course of a scan.

Bringing the two stories together is where the framework earns its multimodal label. The researchers use a Cross-modal dual-head attentive concatenation strategy, in which two attention heads weigh the contributions of the spatial-temporal stream and the connectivity stream before their features are concatenated into a single joint representation. This lets the classifier learn, for each case, whether structural activity patterns or connectivity dynamics carry more diagnostic signal, or how best to combine them. The fused representation is then classified by a Chicken swarm optimized DEformer, a dynamic embedding transformer whose training is guided by a bio-inspired optimization algorithm modeled on the hierarchical foraging search of chicken flocks. The optimization step is intended to improve generalization and help the model avoid poor local minima during training, a common pitfall for deep networks on limited medical datasets.

The evaluation design deserves particular attention, because inflated results are a chronic problem in machine learning applied to brain imaging. The team assembled a balanced benchmark of 800 subjects drawn from the NeuroVault and NITRC public repositories, with 200 subjects each in four groups: Alzheimer’s disease, Parkinson’s disease, epilepsy, and healthy normal controls. Data were split in a subject-independent 70:30 train-test ratio, meaning that no individual contributing training data ever appeared in the test set, and the pipeline was further validated with five-fold cross-validation. All headline results were computed on completely unseen testing subjects, a protocol designed to ensure that the reported performance reflects genuine diagnostic capability rather than memorization of individual scan signatures.

On that benchmark, the framework reached 99.36 percent accuracy with an AUC-ROC of 0.991, figures the authors report as significantly outperforming both baseline models and previously published approaches in the literature. An AUC-ROC near 0.99 means that, for almost any threshold a clinician might choose, the model separates diseased from healthy or distinguishes severity levels with near-perfect trade-offs between sensitivity and specificity. For disorders like Alzheimer’s and Parkinson’s, where severity grading can guide when to escalate therapy, and for epilepsy, where characterizing abnormal activity patterns informs surgical and pharmacological decisions, a reliable automated second opinion could meaningfully change clinical workflows.

Why might the dual-stream design matter so much? Neuroscientists have increasingly recognized that disease signatures live in both places: in the raw spatiotemporal patterns of neural activity and in the way connectivity between regions degrades or reorganizes as pathology advances. Alzheimer’s disease, for example, is associated with progressive disruption of large-scale network communication, while epilepsy involves abnormal synchrony that can emerge and dissipate on rapid timescales. A single-stream model sees only one side of that picture. By fusing a volumetric convolutional view of the brain with a graph-based view of its evolving functional architecture, the framework bridges what the authors describe as spatiotemporal and connectivity-based perspectives of brain function, extracting complementary evidence that neither view alone provides.

The study, published open access on 6 October 2026, arrives amid a broader surge of interest in applying graph neural networks, transformers, and hybrid bio-inspired optimizers to neuroimaging, and it suggests that the next generation of diagnostic tools may increasingly be multimodal even when the input is a single imaging modality. The authors declare no competing interests and report no external funding for the work. Significant hurdles remain before such systems reach the clinic, including validation on larger and more diverse populations, robustness to differences in scanner hardware and acquisition protocols, and prospective testing in real diagnostic settings. But the message of the study is clear: when a model is built to watch both what the brain does and how its parts converse, the faint fingerprints of neurological disease become far easier to read, and the long-sought bridge between fMRI research and bedside decision support moves a measurable step closer.

Subject of Research: Deep learning classification of neurological disease severity from fMRI using spatiotemporal and dynamic functional connectivity features

Article Title: Dual stream spatiotemporal feature learning framework for neurological disease severity level classification using fMRI data

Article References: Smitha, C. S., Sathiya, S., & Soumya, T. (2026). Dual stream spatiotemporal feature learning framework for neurological disease severity level classification using fMRI data. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02527-6

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02527-6

Keywords: fMRI, neurological disorders, Alzheimer's disease, Parkinson's disease, epilepsy, deep learning, dynamic functional connectivity, graph neural network, 3D DenseNet, transformer, disease severity classification, brain connectivity

Carotid stenosis refers to the narrowing of the major arteries in the neck that supply blood to the brain, most often caused by atherosclerotic plaque. When the narrowing exceeds fifty percent of the vessel diameter, the risk of ischemic cerebrovascular events rises sharply. But even in patients who have never experienced a transient ischemic attack or stroke, clinicians have long suspected that reduced or unstable perfusion may contribute to subtle cognitive impairment. Studying this silent phase is difficult precisely because patients feel well, and standard structural imaging often appears unremarkable. The Chinese team, led by Yiyun Zhang and corresponding author Lin Jiang, approached the problem with a pair of complementary analytical techniques that go beyond conventional, static pictures of brain function.

The first technique, dynamic functional connectivity (dFC), treats the brain not as a fixed wiring diagram but as a network whose links strengthen and weaken over seconds. The researchers used a sliding-window approach, chopping continuous resting-state fMRI recordings into short overlapping segments and computing a full connectivity matrix for each. Recurring patterns of connectivity, known as connectivity states, were then identified through clustering, allowing the team to derive temporal metrics such as how often the brain visits each state and how long it lingers there. The second technique, dynamic amplitude of low-frequency fluctuations (dALFF), quantifies the moment-to-moment intensity of spontaneous neural oscillations in each brain region. Crucially, the team computed dALFF not only in the conventional frequency band of 0.01 to 0.08 Hz but separately in two sub-bands: slow-5, spanning 0.01 to 0.027 Hz, and slow-4, spanning 0.027 to 0.073 Hz. This frequency-resolved strategy matters because different frequency bands are thought to reflect distinct physiological and neural processes, and vascular disease may affect them unequally.

Participants in the study were patients with unilateral moderate-to-severe carotid stenosis, graded at fifty percent or greater using criteria derived from the North American Symptomatic Carotid Endarterectomy Trial (NASCET), together with demographically matched healthy controls. All volunteers underwent comprehensive neuropsychological testing, including the mini-mental state examination (MMSE), the digit span test in its forward and backward forms, the Montreal Cognitive Assessment, and the Rey Auditory Verbal Learning Test, alongside the resting-state fMRI scanning session. Ethics approval was granted by the hospital’s ethics committee, and all participants provided written informed consent.

The results revealed a striking pattern. Across the conventional band and both sub-bands, ACS patients showed reduced dALFF compared with controls, meaning the amplitude of their spontaneous low-frequency brain activity was diminished. The affected regions were not random: they clustered within three major brain networks. The default mode network (DMN), which supports self-referential thought and memory consolidation; the frontoparietal network (FPN), the brain’s executive control system; and the sensorimotor network (SMN), which governs movement and bodily sensation, all showed frequency-dependent reductions. One region stood out for its consistency. The left triangular part of the inferior frontal gyrus, a hub for language and cognitive control, was abnormal in every frequency band examined, making it a potential marker of the earliest functional consequences of carotid narrowing.

The dynamic connectivity analysis painted an equally broad picture. ACS patients exhibited widespread alterations in connectivity involving frontal, parietal, and temporal cortical regions, as well as visual, limbic, and subcortical structures, including the right parahippocampal gyrus, the right insula, and the left caudate nucleus. These are not simply areas adjacent to the diseased artery; they span the entire brain, suggesting that chronic hemodynamic stress triggers a global reorganization of network dynamics rather than a localized deficit. The researchers interpret this as evidence that the brain compensates for compromised blood supply by shifting its patterns of coordination, a process that may carry a cognitive cost even when it succeeds in preserving basic function.

The relationship between these imaging abnormalities and cognition was more tentative. Several dALFF and dFC measures showed nominal associations with MMSE scores, digit span test performance, and forward and backward digit span results. These correlations, based on raw uncorrected p-values, hint at a link between altered brain dynamics and poorer attention, working memory, and global cognition. However, the authors are careful to note that none of these associations survived false discovery rate (FDR) correction, the statistical standard used to guard against false positives when many comparisons are made. The findings must therefore be considered preliminary. An additional sobering detail: after applying hemodynamic response function (HRF) correction, a procedure that accounts for the blurring influence of the blood-oxygenation signal on the underlying neural dynamics, no evidence of altered temporal state dynamics remained. Because carotid stenosis directly alters blood flow, disentangling neural change from vascular change is one of the central methodological challenges of the entire field, and the authors transparently report where that challenge limits interpretation.

Even so, the study’s conclusions carry weight for both researchers and clinicians. The demonstration that spontaneous local brain activity is altered in a frequency-dependent manner, while dynamic connectivity is reorganized across multiple networks, provides preliminary neuroimaging evidence for the pathological mechanisms that may underlie ACS-related cognitive decline. If brain dynamics begin to drift years before symptoms appear, then dynamic fMRI measures could eventually serve as early warning indicators, identifying which patients with silent carotid narrowing are most likely to benefit from aggressive management of vascular risk factors, or from revascularization procedures such as carotid endarterectomy or stenting. The work was supported by the National Natural Science Foundation of China and by grants from the Natural Science Foundation of Guizhou Province and the Zunyi Science and Technology Cooperation Project, and it emerges from a provincial innovation team dedicated to functional imaging and artificial intelligence applications.

The methodological toolkit itself represents a step forward for cerebrovascular neuroscience. Machine-learning classifiers mentioned in the study’s analytical framework, including linear and radial basis function support vector machines, random forests, and k-nearest neighbors models, evaluated with leave-one-out cross-validation and receiver operating characteristic analysis, reflect a growing ambition to translate dynamic imaging metrics into diagnostic tools. Whether dALFF reductions in the left inferior frontal gyrus or shifts in dFC state occupancy can ultimately classify patients with clinically useful accuracy will require larger, longitudinal cohorts. The present study’s sample, drawn from a single hospital and analyzed with uncorrected cognitive correlations, is best seen as a proof of concept rather than a definitive answer.

What makes the research resonate beyond the specialist literature is its implication for a remarkably common condition. Carotid atherosclerosis is widespread in aging populations, and many people carry significant narrowing without knowing it. The idea that the resting brain, scanned while a person simply lies still and thinks of nothing in particular, can betray the early consequences of that narrowing is both elegant and clinically provocative. It reframes asymptomatic carotid stenosis not as a dormant disease waiting to strike, but as an active process already imposing costs on brain function. Future work combining dynamic fMRI with direct perfusion measurements, longer follow-up, and stricter statistical correction will determine whether these network signatures can predict who will decline cognitively, and whether restoring blood flow can reverse them. For now, the Zunyi team’s results stand as an early, frequency-resolved portrait of a brain quietly adapting to a compromised blood supply, and a reminder that silence in the arteries is not always silence in the brain.