The research, led by Zhihan Jiang and Xinlin Wang of the Department of Psychosomatic Medicine at Shanghai Tongji Hospital, Tongji University School of Medicine, together with colleagues in Shanghai, Huzhou, and Guangzhou, addressed a question that carries enormous clinical weight. Depression frequently co-occurs with chronic pain, and this overlap is associated with worse clinical outcomes and treatment resistance. Understanding the brain-level signature of this comorbidity could ultimately help clinicians identify which patients are at greatest risk and design more targeted interventions.

The study recruited three groups of participants matched for age, gender, and education level: 35 depression patients with chronic pain, referred to as the DCP group; 39 depression patients without chronic pain, the DNCP group; and 45 healthy controls. Each participant underwent a resting-state functional magnetic resonance imaging scan, a technique that measures spontaneous brain activity while the participant lies still and performs no explicit task. This approach is particularly valuable in psychiatry because it captures the brain’s intrinsic functional organization rather than performance on any specific cognitive exercise. In addition to imaging, participants completed the Hamilton Depression Rating Scale, a standard clinical measure of depressive symptom severity, and the Visual Analog Scale of pain, which quantifies subjective pain intensity.

The final analysis included 34 patients in the depression-with-pain group, 38 patients with depression alone, and all 45 healthy controls. The researchers applied two complementary measures of local brain function. The first, the amplitude of low-frequency fluctuations, or ALFF, quantifies the intensity of spontaneous neural activity in a given region by measuring fluctuations in the blood oxygenation signal at very low frequencies, typically below 0.1 hertz. Regions with higher ALFF are generally interpreted as showing greater spontaneous neural activity. The second measure, regional homogeneity, or ReHo, assesses the degree to which activity in a given voxel is synchronized with its immediate neighbors, providing an index of local functional coherence. Because these two metrics probe different aspects of resting-state function, examining both in parallel gives a richer picture than either alone.

The comparison between the two patient groups produced a clear and internally consistent result. Compared with depressed patients who did not have chronic pain, those with comorbid chronic pain showed significantly decreased ALFF in two adjacent regions of the left temporal lobe: the left middle temporal gyrus and the left inferior temporal gyrus. These regions are traditionally associated with higher-order sensory processing and semantic and visual interpretation, but growing evidence implicates them in the cognitive and affective evaluation of bodily states. Reduced spontaneous activity in these areas suggests that when depression is accompanied by persistent pain, the temporal cortical machinery involved in integrating sensory and emotional information may be operating at a lower functional level.

The ReHo findings painted a more nuanced picture, revealing both decreases and increases in local synchronization depending on the region involved. Patients with depression and chronic pain showed lower local synchronization in the bilateral putamen and caudate, two core components of the striatum, the brain’s principal input structure for basal ganglia circuits. The striatum is central to reward processing, motivation, and habit formation, and its disruption is a well-established feature of depression. The new data suggest that when chronic pain is layered onto depression, this reward-related dysfunction extends into a measurable loss of local coordination within striatal circuits, potentially linking the blunted motivation and anhedonia of depression with the persistence of pain perception.

At the same time, the comorbid group displayed higher regional homogeneity in the left postcentral gyrus and the bilateral cuneus. The postcentral gyrus houses the primary somatosensory cortex, the region that processes tactile and bodily sensation, and heightened local synchrony there fits naturally with the clinical reality of chronic pain, in which somatosensory circuits become sensitized and persistently active. The cuneus, located in the occipital lobe and involved in visual processing, has also been repeatedly implicated in pain-related imaging studies, possibly reflecting attentional and perceptual amplification of bodily signals. Together, these increases point to a brain in which pain-processing regions are running hot even at rest, while reward and temporal integrative regions are running cold.

To probe whether the regions showing ALFF differences were also communicating abnormally with the rest of the brain, the researchers used the left middle temporal gyrus and left inferior temporal gyrus as seed regions for exploratory whole-brain functional connectivity analyses, which test how strongly activity in the seed correlates with activity elsewhere. Notably, no significant functional connectivity differences emerged between the comorbid and depression-only groups. This null result suggests that the neural alterations accompanying depression-chronic pain comorbidity may be primarily local in nature, involving the intensity and coherence of regional activity rather than long-range network integration. The authors framed these findings as providing further insight into the neural alterations associated with the comorbidity and as a foundation for future mechanistic investigations.

Statistical rigor was built into the analysis at multiple levels. Feature associations between brain measures and clinical scores were assessed using Pearson correlation analysis with Benjamini-Hochberg correction for multiple comparisons, a procedure that controls the false discovery rate and reduces the risk that apparently significant associations arise by chance alone. The demographic matching of the three groups on age, gender, and education helps ensure that the observed neural differences cannot be attributed to these basic variables, strengthening the inference that they relate specifically to the presence or absence of chronic pain in depressed patients.

The clinical implications of the study are significant. Chronic pain and depression are among the most common co-occurring conditions in medicine, and patients carrying both diagnoses frequently respond poorly to standard antidepressant treatment. By identifying specific neural markers, reduced spontaneous activity in temporal regions, desynchronized striatal function, and hyper-synchronized somatosensory and occipital activity, the study provides concrete biological targets that could guide future work on biomarkers, treatment stratification, and mechanism-based therapies. If the striatal findings, for example, prove reproducible in larger samples, they might motivate treatment approaches that jointly address reward dysfunction and pain sensitization rather than treating the two conditions in isolation.

The study also has important limitations inherent in its design. As a cross-sectional resting-state investigation with moderately sized groups, it cannot determine whether the observed brain differences are a cause or a consequence of the comorbidity, and the exploratory functional connectivity analyses yielded no group differences, leaving open the question of how these local alterations relate to broader network dynamics. The authors acknowledge that their results serve as a basis for future mechanistic investigations rather than a definitive explanation. Longitudinal studies tracking patients over time, along with multimodal imaging that combines functional measures with structural and neurochemical data, will be needed to clarify how these regional signatures emerge and whether they predict treatment response.

Funding for the work came from the National Key Research and Development Program of China, the Shanghai Shen-Kang Hospital Development Center Program, the National Natural Science Foundation of China, and the National Postdoctoral Program for Innovative Talents. The study was approved by the Shanghai Tongji Hospital Ethics Committee, and all participants provided informed consent before data collection. The design adhered to the principles of the Declaration of Helsinki, and the authors declared no competing interests.

For a field in which depression and chronic pain have often been studied separately despite their frequent co-occurrence, this research represents a meaningful step toward treating the intersection as a distinct clinical and neurobiological entity. The picture that emerges is one of a brain whose temporal integrative regions and striatal reward circuits are dampened, while its somatosensory and visual pain-processing regions are amplified, a pattern that may ultimately explain why patients with both conditions suffer more and respond less to existing treatments.

The clinical problem is consequential because delayed recognition of neonatal seizures can allow repeated abnormal electrical activity to continue before treatment begins. Newborn EEG is especially challenging to interpret: normal activity changes with developmental state, artifacts can resemble neurological events, and seizures may have limited visible clinical expression. The study notes that expert readers can miss roughly one in four events under standard monitoring conditions, consistent with the broader difficulty of visual interpretation reported in neonatal care. The researchers therefore focused not simply on detecting an ongoing seizure, but on classifying EEG segments as preictal or interictal, meaning sufficiently distant from a seizure to represent a non-seizure baseline. They defined preictal data as the four-minute interval before seizure onset and interictal data as periods more than five minutes from any seizure onset or offset. A 60-second guard interval and all ictal segments were excluded, preventing the two labels from overlapping. This produced a strongly imbalanced learning problem: interictal segments outnumbered preictal segments by approximately 6.48 to one.

The dataset contained about 5,800 hours of continuous, multichannel EEG from 79 term infants and 456 annotated seizure events. Signals were recorded through a 21-channel International 10–20 montage at 256 hertz, providing coverage across frontopolar, frontal, central, temporal, parietal, and occipital regions, along with auxiliary ECG and respiration channels. The researchers divided the recordings into overlapping 10-second epochs, generating 75,488 usable segments after preprocessing. Each channel was normalized separately within each recording to reduce differences in scale and signal drift, and flatline clips were set to zero. Crucially, the split was performed by patient rather than by individual epoch. Fifty-five infants were assigned to training, 12 to validation, and 12 to testing, so neighboring windows from the same recording could not appear in different partitions. The held-out test set contained 10,889 segments, including 9,376 interictal and 1,513 preictal examples. The authors also report a five-fold patient-level cross-validation analysis intended to test whether results depended too heavily on one division of the cohort.

The first stage of the model addresses a central limitation in medical AI: labeled seizure examples are scarce, while unlabeled monitoring data are abundant. Inspired by the SimCLR framework, the researchers used self-supervised contrastive pretraining primarily on interictal EEG. For each segment, the training process created two altered views and taught an encoder to produce similar representations for the paired versions while separating representations from other examples. The alterations were designed to mimic conditions encountered in clinical recordings, including Gaussian noise, temporal shifts, random channel dropout, pointwise masking, and amplitude scaling. A one-dimensional residual convolutional encoder transformed the 21-channel signals into a lower-dimensional representation. Its projection head produced a normalized 64-dimensional contrastive embedding. In this setting, the system did not need seizure labels to learn general features of neonatal EEG. According to the study, these pretrained representations improved downstream F1 scores by 8 to 12 percent compared with the relevant non-pretrained configurations, while the contrastive training loss fell below 0.1.

The second stage combines the learned representation with conventional signal-processing information before passing it to a spiking classifier. The pretrained module supplied 192 features: a 128-dimensional encoder output and a 64-dimensional contrastive projection. The researchers also calculated power spectral density with Welch’s method across five frequency bands—delta, theta, alpha, beta, and gamma—for each of the 21 electrodes. These 105 spectral measurements were compressed to 32 features, producing a 224-dimensional input. The classifier, called an attention-enhanced spiking neural network, used fully connected layers with batch normalization and dropout, followed by leaky integrate-and-fire neurons. These units accumulate input in a membrane-potential state, gradually lose that potential through leakage, and emit a binary spike when a threshold is reached. The network simulated this process over 50 timesteps, allowing it to represent temporal evolution rather than treating each input as a static vector. A surrogate gradient enabled backpropagation through the otherwise discontinuous spike-generation function, and average spike rates were used to produce probabilities for the preictal and interictal classes.

The architecture was trained with focal loss, which gives extra emphasis to difficult examples and the under-represented preictal class without discarding data through resampling. The model contained approximately one million parameters and exhibited reported spike sparsity of 15 to 20 percent, features the researchers associate with potential low-power, edge-device deployment. On the held-out test data, it identified 1,362 of 1,513 preictal segments, corresponding to the reported 90.02 percent recall, while missing 151. Its precision was 60.35 percent, yielding an F1 score of 72.25 percent; the macro-F1 score was 83.22 percent and the weighted F1 score was 91.14 percent. The confusion matrix included 8,481 true-negative classifications and 895 false positives. The precision-recall analysis produced an average precision of 0.76. These figures illustrate the trade-off at the heart of an early-warning system: prioritizing sensitivity can produce more alarms, some of which may not correspond to a genuinely approaching seizure. The authors describe the high recall as clinically attractive but acknowledge that false alarms could contribute to alarm fatigue.

Interpretability was built into the analysis rather than treated as an afterthought. The researchers applied Integrated Gradients, SHAP, LIME, saliency gradients, and attention profiling to preictal examples, then mapped the resulting attributions to the standard electrode layout. These methods answer related but different questions: which features change a prediction, which contribute globally, which matter for an individual example, where the output is most sensitive, and how the model’s internal weighting is distributed. Across the analysis, temporal regions accounted for approximately 40 percent of the reported contribution, central regions 25 percent, frontal regions 20 percent, parietal regions 10 percent, and occipital regions 5 percent. SHAP identified the T3 temporal-left channel, F8 frontal-right channel, and P4 parietal-right channel among the leading contributors. Temporal channels such as T3 and T4 and the central midline site Cz repeatedly ranked highly across attribution methods and sampled windows. The researchers say this pattern is qualitatively consistent with established descriptions of temporal and central-temporal involvement in neonatal seizure activity, but they emphasize that the explanations have not undergone formal validation by expert neurophysiologists.

The results suggest that combining representation learning, spectral information, and event-driven temporal modeling may help address the particular constraints of neonatal EEG, but substantial barriers remain before clinical use. The study was conducted offline on a single publicly available dataset, and performance on recordings from other hospitals, equipment, populations, and clinical workflows remains unknown. Fixed preictal windows may not represent the same biological process for every infant, motivating future adaptive or personalized definitions. Continuous explainability analysis could also be computationally demanding, even if the underlying classifier is compact. The authors propose further work involving model compression, lighter interpretability methods, multimodal information, streaming evaluation, and clinician-in-the-loop assessment. Ethical safeguards, patient privacy, and direct clinical oversight would be essential in any deployment. For now, the system is best understood as a research prototype: a promising attempt to forecast neonatal seizure-related activity while exposing the EEG regions and features behind its predictions, rather than as a replacement for specialist monitoring or medical judgment.

Contrastive pretraining is particularly relevant to neonatal EEG because the model can learn recurring structure from recordings that lack event annotations. By bringing augmented views of the same signal closer in representation space, the encoder is encouraged to retain features that remain stable despite modest shifts, noise, amplitude changes, or missing channels. This may improve robustness to routine recording imperfections, although the value of any augmentation depends on whether it preserves clinically meaningful seizure-related information. An alteration that is harmless for baseline EEG could potentially obscure a transient abnormality.

The spiking component provides a different form of temporal representation from the preceding convolutional encoder. A leaky integrate-and-fire unit carries a decaying internal state, so inputs separated in time can influence one another without requiring every signal value to be processed identically. The reported sparsity indicates that many potential spike operations are absent, which could reduce energy use on suitable neuromorphic hardware. It does not by itself establish faster or more efficient clinical operation, however, because total system cost also includes signal conditioning, feature extraction, memory access, and explanation generation.

Performance should also be interpreted at the level of clinical episodes rather than only short EEG windows. A high segment-level recall can arise when several neighboring epochs from one evolving event are correctly classified, while false positives distributed across long recordings may still create a burdensome alarm rate. Prospective testing would therefore need episode-level sensitivity, false alarms per monitoring hour, warning time, calibration, and stability across infants. Attribution maps can help investigate such behavior, but agreement among explanation methods is not proof that the highlighted electrodes represent a causal seizure mechanism. Their main immediate value is supporting model auditing and clinician review.

Subject of Research: Interpretable AI for forecasting neonatal seizures from EEG recordings

Article Title: A hybrid spiking neural network with contrastive pretraining for interpretable seizure forecasting using explainable AI

Article References: Selvaraj, J., Krishna, R., Gupta, A., & Guruviah, V. (2026). A hybrid spiking neural network with contrastive pretraining for interpretable seizure forecasting using explainable AI. Discover Informatics, 1(1), Article 8. https://doi.org/10.1007/s44564-026-00010-5

Image Credits: AI Generated

DOI: 10.1007/s44564-026-00010-5

Keywords: neonatal seizures, electroencephalography, seizure forecasting, spiking neural networks, contrastive learning, explainable AI, neuromorphic computing, neonatal intensive care, hybrid, spiking, neural, network

Cassia absus belongs to the Fabaceae family and grows in tropical and subtropical regions of Asia. Its seeds have a history of medicinal use and contain reported bioactive constituents including chaksine and isochaksine. That traditional and phytochemical background made the plant a candidate for a more systematic investigation of its extractable compounds. The researchers were not testing a finished medicine or demonstrating a treatment for disease; they were developing and optimizing a laboratory extraction process. The distinction matters because measurements such as antioxidant activity in a test tube do not establish clinical benefit. They do, however, help characterize extracts and identify whether a plant material may warrant further chemical, toxicological and pharmaceutical study.

The process began with seeds purchased from a local market in Lahore, Pakistan. The seeds were washed, dried, ground and passed through an 80-mesh sieve to produce a relatively uniform powder. The material was defatted by soaking it in n-hexane for three days, then dried at 40 degrees Celsius. For the greener extraction stage, the researchers prepared the deep eutectic solvent by combining choline chloride and glycerol in a 1:3 molar ratio. The mixture was heated at 80 degrees Celsius for two hours under vacuum until it formed a clear, colourless liquid. For extraction, the solvent was mixed with water in equal proportions. This water-containing system was then brought into contact with one gram of the prepared seed powder.

Ultrasound supplied the physical force intended to open the plant matrix. When a probe emits high-intensity sound into a liquid, microscopic bubbles can form, grow and collapse in a phenomenon known as acoustic cavitation. These rapid events can disturb cell walls, improve wetting and solvent penetration, and increase movement of dissolved molecules from the solid material into the surrounding liquid. The technique can therefore accelerate mass transfer compared with passive soaking. But more energy or more time is not automatically better. Excessive sonication can increase heating, alter fragile compounds or reduce the energy delivered efficiently through the liquid. The researchers monitored temperature and avoided excessive heat build-up while varying ultrasound amplitude, extraction time and solvent-to-feed ratio across a Box–Behnken experimental design.

The study evaluated four responses: total phenolic content, total flavonoid content, DPPH radical-scavenging activity and iron-chelating activity. Total phenolic content was expressed as milligrams of gallic acid equivalents per gram, while flavonoid content was reported using a rutin-equivalent calibration. DPPH testing measures how effectively an extract reduces a stable laboratory radical, producing an estimate of radical-scavenging capacity. The iron-chelation assay examined the ability of the extract to interfere with the reaction between ferrous ions and ferrozine, with lower colour formation corresponding to greater apparent chelation. These are widely used screening measurements, but they represent chemical behaviour under defined assay conditions rather than proof that the extract will neutralize radicals or regulate iron in a human body.

Seventeen experimental runs, including five centre points, were used to map how the three process variables affected the four responses. The results showed that no single run maximized every measurement. One run produced the highest total phenolic content at 44.18 milligrams of gallic acid equivalents per gram, another produced the highest flavonoid content at 29.26 milligrams of rutin equivalents per gram, a third reached 89.53 percent DPPH radical-scavenging activity, and a fourth recorded 90.31 percent iron-chelating activity. This divergence reflects the chemical complexity of extraction. Phenolics, flavonoids and other active constituents differ in polarity, solubility and stability, so conditions that release one group efficiently may not recover another group or preserve its activity. Maximizing one result could consequently produce an extract that performs poorly across the broader set of desired properties.

Response surface methodology was used first to fit second-order polynomial models describing linear, quadratic and interaction effects. The models were statistically significant for all four responses, with p-values below 0.0001 for phenolic and flavonoid content, 0.0011 for DPPH activity and 0.0024 for iron-chelating activity. Model R-squared values ranged from 0.9299 to 0.9933, indicating that the equations accounted for much of the variation within the tested design space. The solvent-to-feed ratio emerged as the strongest influence on phenolic and flavonoid recovery. Increasing solvent availability likely improved penetration and maintained a concentration gradient that favoured diffusion, but the negative quadratic terms showed that the benefit eventually levelled off or declined. Too much solvent could dilute the extract or reduce ultrasonic energy density.

The antioxidant response was more complicated. Extraction time had a significant negative effect on DPPH activity, suggesting that prolonged sonication may have degraded or structurally modified sensitive radical-scavenging compounds. Ultrasound amplitude and solvent-to-feed ratio also interacted, meaning their effects could not be interpreted independently. Iron-chelating activity was governed mainly by quadratic effects rather than simple increases or decreases in individual variables. The researchers then trained a feedforward artificial neural network using 70 percent of the experimental data for training, with 15 percent each reserved for validation and testing. The selected network used two hidden layers containing 20 and 10 neurons, with logsig and tansig activation functions. Its overall correlation values ranged from 0.97256 for iron-chelating activity to 0.99469 for total phenolic content, although the small dataset means these strong figures apply only within the investigated range and require confirmation with additional experiments.

TOPSIS provided the final decision framework by treating every experimental run as an alternative and all four responses as beneficial criteria. The data were normalized, given equal weights and compared with an ideal solution representing the best combined performance. Run 12 achieved the highest closeness coefficient, 0.7153, at 60 percent amplitude, six minutes and 20 millilitres per gram. Its measured results were 33.51 milligrams of gallic acid equivalents per gram of total phenolics, 27.99 milligrams of rutin equivalents per gram of flavonoids, 74.70 percent DPPH activity and 82.33 percent iron-chelating activity. It did not lead every individual category, but it offered the strongest overall balance. Compared with the lowest-ranked run, it had approximately 2.7 times more total phenolics, 27 percent higher flavonoid content and 59 percent higher DPPH activity, despite slightly lower iron-chelating activity.

The modelling comparison gave the neural network a modest advantage over response surface methodology. At the selected condition, the artificial neural network showed prediction errors of 1.15 percent for flavonoid content and 2.18 percent for DPPH activity, while TOPSIS was closest for total phenolic content with a 1.03 percent error. All models predicted iron-chelating activity with errors below 1 percent. Across the dataset, the neural network generally produced lower average absolute deviations and mean absolute percentage errors, particularly for the nonlinear antioxidant and chelation responses. The researchers also assessed the method with the ComplexMoGAPI green analytical metric, which gave an overall score of 81. The favourable score reflected the use of a choline chloride–glycerol and water system, room-temperature extraction, short sonication and avoidance of more hazardous conventional solvents. However, the reported E-factor was 60, and extraction yield remained below 70 percent, showing that waste and solvent efficiency still need improvement.

The findings position the method as a promising laboratory framework rather than an industrially validated product. Natural deep eutectic solvents can be tuned by changing their components and proportions, and their low volatility and biodegradability are attractive for natural-product processing. Yet solvent recovery, viscosity, long-term stability, compound identification and scale-up must be addressed before commercial adoption. The researchers recommend compound-level characterization, stability testing, toxicity evaluation and pilot-scale validation. Future work could also examine whether the solvent can be reused, whether lower solvent volumes can maintain performance and which specific molecules account for the measured activities. For now, the study demonstrates how acoustic cavitation and data-driven optimization can turn a traditional plant resource into a more systematically studied extraction target, while also showing that a greener label does not eliminate the need to measure waste, validate predictions and test biological claims carefully.

An important consideration is that the reported response values are operational measurements tied to the extraction and assay protocols. Total phenolic and flavonoid results depend on the calibration standards used, while DPPH and iron-chelation values summarize reactions in controlled chemical systems. They can therefore be useful for comparing extraction conditions without identifying which individual seed constituents produced the response. Chemical profiling would be needed to connect the optimized process with specific molecules such as the reported Cassia absus alkaloids or other extract components.

The optimization also illustrates why process conditions should be treated as a defined operating window rather than a universal recipe. The selected settings were derived from a Box–Behnken design covering particular amplitude, time and solvent-to-feed ranges, with a 50:50 NaDES–water extraction mixture and pretreated seed powder. Performance outside those conditions cannot be inferred from the model alone. Changes in particle characteristics, solvent composition, equipment geometry or temperature control could alter cavitation and mass transfer. Independent confirmation using new batches of seeds, expanded chemical characterization and scale-relevant equipment would help establish how reproducible the balance identified by TOPSIS is.

Subject of Research: AI-guided ultrasound extraction of Cassia absus seed phytochemicals using a natural deep eutectic solvent

Article Title: Artificial neural network and TOPSIS guided ultrasound extraction of Cassia absus L. seed phytochemicals using a natural deep eutectic solvent

Article References: Khalid, N. U. A., Iftikhar, H., Ahmed, D., & Mushtaq, M. (2026). Artificial neural network and TOPSIS guided ultrasound extraction of Cassia absus L. seed phytochemicals using a natural deep eutectic solvent. Discover Green Chemistry, 1(1), Article 26. https://doi.org/10.1007/s44509-026-00031-1

Image Credits: AI Generated

DOI: 10.1007/s44509-026-00031-1

Keywords: Cassia absus, green extraction, natural deep eutectic solvents, ultrasound extraction, artificial neural networks, TOPSIS, antioxidants, phytochemicals, Artificial, neural, network, guided

Magnetic resonance imaging (MRI) is central to this emerging field because it allows researchers to examine brain structure and function in living patients without exposing them to ionizing radiation. Structural MRI can measure regional brain volume, cortical thickness and the condition of white matter, the communication network formed by nerve fibers. Functional MRI, by tracking changes in blood oxygenation, provides indirect measures of neural activity and identifies alterations in connectivity between brain regions. Diffusion tensor imaging offers a closer look at the organization of white matter by mapping the movement of water along nerve fibers, while magnetic resonance spectroscopy can detect changes in brain chemistry, including metabolites associated with energy use, inflammation and neuronal injury.

Cardiovascular disease provides one of the clearest examples of the brain-body connection. Long-term hypertension can damage small blood vessels that supply the brain, while atherosclerosis may reduce the flexibility and efficiency of larger arteries. Heart failure can also limit cerebral blood flow and alter the delivery of oxygen and nutrients. MRI studies have associated these conditions with white matter hyperintensities, reductions in gray matter volume and changes in cerebral perfusion. Such abnormalities may appear before obvious memory loss or other cognitive symptoms, raising the possibility that neuroimaging could identify cardiovascular-related brain injury at an earlier stage. The growing concept of the “cardiac brain” emphasizes that protecting the heart may also help preserve cognition.

Metabolic health is similarly reflected in the brain. Diabetes exposes tissues to prolonged hyperglycemia, insulin resistance and vascular stress, while obesity can promote chronic low-grade inflammation and hormonal disruption. Together, these processes may accelerate biological brain aging and increase the risk of cognitive decline and dementia. Imaging research has linked type 2 diabetes with smaller hippocampal volumes, compromised white matter integrity and disrupted functional communication among networks involved in memory, attention and executive control. These changes suggest that the effects of metabolic disease extend beyond the blood vessels and may directly influence the brain’s ability to maintain and repair its neural circuits.

The immune system represents another major route of communication between the body and the brain. In autoimmune diseases such as systemic lupus erythematosus and rheumatoid arthritis, persistent inflammation can affect the vascular system and alter the permeability of the blood-brain barrier. This barrier, formed by specialized blood vessels and supporting cells, normally limits the entry of potentially harmful substances into neural tissue. When its protective function is weakened, inflammatory molecules and immune cells may contribute to neuroinflammation, altered neurotransmission and neuronal dysfunction. Multiple sclerosis illustrates a more direct immune attack on the central nervous system, with imaging revealing lesions and changes in brain volume that may develop alongside, or sometimes precede, clinical disability.

Infectious diseases have made the systemic vulnerability of the brain especially visible. During and after COVID-19, neuroimaging studies reported changes in brain structure, connectivity and metabolism in some patients, including individuals who experienced persistent symptoms after the acute infection had resolved. The biological mechanisms remain under investigation and may involve inflammation, vascular injury, impaired oxygen delivery, immune dysregulation or indirect effects of severe illness. Comparable concerns have emerged after other viral and bacterial infections, reinforcing the idea that pathogens do not need to invade brain tissue directly to affect neurological function. Systemic infection can create a biological environment capable of altering the brain over extended periods.

The relationship also works in the opposite direction. Brain disorders can influence peripheral organs through the autonomic nervous system, which regulates heart rate, blood pressure, digestion and immune activity. Endocrine pathways involving stress hormones can modify metabolism and immune responses, while inflammatory signals generated in the body can feed back into the brain and affect mood, cognition and behavior. This bidirectional communication helps explain why neurological and psychiatric conditions are frequently accompanied by cardiovascular, gastrointestinal or immune disturbances. It also provides a framework for understanding psychoneuroimmunology, a field focused on the interactions among psychological processes, the nervous system and immune function.

The clinical potential of these discoveries is substantial. Imaging biomarkers could help detect hidden brain involvement in patients being treated for hypertension, diabetes, autoimmune disease or infection. They might also assist physicians in estimating disease risk, selecting therapies and monitoring whether an intervention is protecting neural tissue. Conversely, treatments aimed at systemic inflammation, vascular health or metabolic regulation could potentially improve outcomes in brain disorders. The survey emphasizes that a patient’s neurological condition should not be assessed independently from the health of the heart, blood vessels, immune system and metabolism.

The next phase of brain-body research will depend on combining multiple forms of evidence rather than relying on a single scan or biological measurement. Integrating structural MRI, fMRI, diffusion imaging, spectroscopy, blood-based biomarkers and clinical data may reveal patterns that are invisible to any one technique. Machine-learning systems could help identify complex imaging signatures associated with disease and predict which patients are most likely to develop cognitive complications. However, large longitudinal studies will be essential to determine whether imaging changes are causes, consequences or early indicators of systemic disease. By tracing these relationships over time, researchers hope to transform brain-body imaging from a descriptive tool into a practical system for earlier diagnosis, individualized treatment and prevention.

Subject of Research: Not applicable

Article Title: Brain-body interactions in systemic diseases: a survey from an imaging perspective

Web References: https://doi.org/10.1007/s11684-026-1229-8

Image Credits: Higher Education Press

Keywords: brain-body interactions, magnetic resonance imaging, systemic diseases, neuroimaging, cardiovascular disease, diabetes, autoimmune disease, neuroinflammation, COVID-19, brain health