Research themes

Sheaf Neural Networks (SNNs) generalize Graph Neural Networks (GNNs) by replacing scalar node signals with stalk-valued signals and by using restriction maps to measure compatibility across edges. Unlike standard graph diffusion, which encourages neighboring node features to become similar, sheaf diffusion promotes consistency through the restriction maps and can therefore model more general relationships between neighboring nodes. However, existing sheaf neural architectures mainly operate at a fixed graph resolution and do not provide a principled pooling mechanism for building hierarchical representations. In this paper, we introduce Hierarchical Sheaf Pool (HiSP), a sheaf-aware pooling framework based on local spectral coarsening. Given a partition of the graph, HiSP constructs each coarse stalk by projecting fine stalk-valued features onto the low-frequency eigenmodes of the cluster-internal sheaf Laplacian. These local modes define a cochain-level prolongation map, which allows the fine sheaf energy to be represented on the coarse space through a Galerkin operator. We further analyze the approximation induced by coarsening by separating truncation loss, due to discarded local modes, from realization loss, due to representing the projected operator as a coarse sheaf. Finally, we implement HiSP as a GNN pooling layer compatible with SNNs and provide a PyG implementation supporting batching, lifted sheaf Laplacians, and hierarchical architectures.

A prominent theory in cognitive science suggests that concepts in the brain are organized as high-dimensional vectors, with semantic meaning captured by directions and relative angles in this space. Brain decoding is the effort of reconstructing or retrieving stimuli (or their representations) from neural activity and involves finding a function that approximates how the brain represents concepts. This motivates the investigation of contrastive objectives as biologically plausible candidates to reverse the brain loss function. In this work, we study how functional MRI (fMRI) activity can generally be mapped with the embedding spaces of foundation models in vision, language, and audio. Although neural computations are highly non-linear at the microscale, fMRI measurements average signals across space and time, further smoothed by noise, effectively linearizing the observable representation. Consistent with these views, our experiments across multiple datasets demonstrate that linear contrastive decoders consistently outperform ridge regression and standard non-linear alternatives, and that these results generalize across images, text, and sound. These findings indicate that decoding gains arise more from the choice of training objective than from architectural complexity, pointing to contrastive-linear models as a principled strategy for brain decoding.

Music is a universal phenomenon that influences human experiences across cultures. We investigate whether music can be decoded from human brain activity measured with fMRI, by modeling mappings between neural data and latent representations of musical stimuli. Our approach integrates functional and anatomical alignment techniques to facilitate cross-subject decoding. Starting from the GTZan fMRI dataset, where five participants listened to 540 musical tracks from 10 genres, we used the CLAP model to extract latent representations of the musical stimuli and developed voxel-wise encoding models to identify brain regions responsive to these stimuli, by applying a threshold to the correlation between predicted and actual brain activity. Our decoding pipeline, primarily retrieval-based, employs a linear map to project back brain activity to the corresponding CLAP features. This enables us to retrieve the musical stimuli most similar to those that originated the fMRI data. Our results demonstrate state-of-the-art identification accuracy, outperforming existing approaches.

Recent progress in visual brain decoding from fMRI has been enabled by large-scale datasets such as the Natural Scenes Dataset (NSD) and powerful diffusion-based generative models. While current pipelines are primarily optimized for perception, their performance under mental-imagery remains less well understood. In this work, we study how a state-of-the-art (SOTA) perception decoder (DynaDiff) can be adapted to reconstruct imagined content from the Imagery-NSD benchmark. We propose a latent functional alignment approach that maps imagery-evoked activity into the pretrained model's conditioning space, while keeping the remaining components frozen. To mitigate the limited amount of matched imagery-perception supervision, we further introduce a retrieval-based augmentation strategy that selects semantically related NSD perception trials. Across four subjects, latent functional alignment consistently improves high-level semantic reconstruction metrics relative to the frozen pretrained baseline and a voxel-space ridge alignment baseline, and enables above-chance decoding from multiple cortical regions. These results suggest that semantic structure learned from perception can be leveraged to stabilize and improve visual imagery decoding under out-of-distribution conditions.

This paper presents a novel approach towards creating a foundational model for aligning neural data and visual stimuli across multimodal representationsof brain activity by leveraging contrastive learning. We used electroencephalography (EEG), magnetoencephalography (MEG), and functional magnetic resonance imaging (fMRI) data. Our framework's capabilities are demonstrated through three key experiments: decoding visual information from neural data, encoding images into neural representations, and converting between neural modalities. The results highlight the model's ability to accurately capture semantic information across different brain imaging techniques, illustrating its potential in decoding, encoding, and modality conversion tasks.

Quantifying the volume of distribution (VT) in Positron Emission Tomography (PET) is widely considered the gold standard for assessing tracer binding. However, this process requires an accurate estimation of the tracer's input function (IF) obtained through arterial sampling and metabolite correction-procedures that are both invasive and technically demanding. To overcome these limitations, we introduce a neural network-based framework for estimating the IF directly from [11C]PBR28 dynamic PET data, exploring generalisability across datasets and scanners. The framework employs a patched variational autoencoder (pVAE) for dimensionality reduction, generating IFs with uncertainty (NNIF-dPET), and computes VT from the mean output signal. Additionally, we evaluate two alternative methods: NNIF-IDIF, which derives IFs from image-derived input functions, and NNIF-unBlood, which uses uncorrected blood signals as input. NNIF-dPET achieves accuracy comparable to true arterial IFs while outperforming IDIF-based methods, suggesting that latent space representations can effectively approximate whole-blood activity for parent plasma input function estimation, rather than relying on pre-selected voxels.Clinical Relevance-This approach highlights the potential for scalable, non-invasive PET quantification across diverse clinical settings.

Background/Objectives: Parkinson's disease (PD) is characterized by progressive neurodegeneration affecting both motor and non-motor functions. Identifying early alterations in PD patients before the onset of dopaminergic therapy is crucial for understanding disease progression and developing targeted interventions. This study aimed to investigate early changes in the putamen and thalamus in de novo PD patients using diffusion tensor imaging (DTI) compared to healthy controls. Methods: Thirty-one de novo PD patients and thirty-three healthy controls underwent DTI scanning. Tract-based spatial statistics were used to compare fractional anisotropy (FA) values between groups. Results: De novo PD patients exhibited significantly lower FA values in the right thalamus compared to controls, suggesting alterations in neuronal integrity or fiber degeneration in the early stages of the disease. However, no significant differences were demonstrated for FA values in the putamen between groups. Conclusions: We demonstrated that the FA value in the right thalamus was lower in PD compared with healthy controls. These findings highlight the potential of DTI as a non-invasive tool for detecting early neural changes in PD patients. Further studies would be helpful to assess the clinical utility of serial FA measurements of the subcortical gray matter in objective quantification of disease progression and monitoring of the therapeutic response.

Childhood maltreatment is a key risk factor for conduct disorder (CD), and the "ecophenotype hypothesis" suggests that maltreatment-related versus non-maltreatment-related CD are neurobiologically distinct. This may explain inconsistent findings in previous structural connectivity studies of CD. We tested this hypothesis by comparing youth with CD with (CD/+) versus without (CD/-) childhood physical or sexual abuse in white-matter microstructure. Diffusion tensor imaging data were collected from 100 CD and 169 control participants aged 9-18 years. Using Tract-Based Spatial Statistics, we compared the CD and control groups in fractional anisotropy, and axial, radial and mean diffusivity, then compared the CD/+ (n = 39) and CD/- (n = 61) subgroups and controls. The combined CD group had higher fractional anisotropy in the corpus callosum than controls. When divided by abuse history, only the CD/- subgroup exhibited higher corpus callosum fractional anisotropy than controls; the CD/+ subgroup did not differ from controls. Comparing the CD subgroups, the CD/+ subgroup displayed higher superior longitudinal fasciculus axial diffusivity than the CD/- subgroup. Notably, sex-stratified analyses yielded different findings in all-male and all-female samples. Findings support the ecophenotype hypothesis, demonstrating microstructural differences between the CD/+ and CD/- subgroups and emphasizing the importance of considering abuse/maltreatment (and sex) in future studies.

BACKGROUND: Statins appear to be useful in patients with acute ischemic stroke. Our aim was to evaluate the association between premorbid statin treatment and CT perfusion characteristics of acute ischemic stroke. METHODS: A retrospective analysis of patients with acute stroke secondary to occlusion of large vessels in the anterior circulation was performed to assess collateral flow, ischemic core volume, and ischemic penumbra using CT angiography and CT perfusion maps. Fisher's exact test was used to compare baseline characteristics of patients in the two groups. The Wilcoxon rank-sum test for independent groups was used to compare all variables obtained for the two different groups with and without statin use. RESULTS: We identified 61 patients, including 29 treated with statins and 32 not treated with statins before stroke onset matched by age, gender, and vascular risk factors except for hypercholesterolemia. The statin group showed lower National Institutes of health Stroke Scale scores at onset (14 ± 6.1 vs. 16 ± 4.5; p = 0.04) and lower volumes of brain tissue characterized by impaired cerebral blood flow (CBF), cerebral blood volume (CBV), and Tmax9.5-25s; otherwise, no statistically significant difference was found in the volume of the Tmax16-25s between the two groups. CONCLUSIONS: Premorbid statin treatment is associated with a favorable imaging condition of acute ischemic stroke in terms of ischemic core and ischemic penumbra volume.

INTRODUCTION: We propose a novel approach for the non-invasive quantification of dynamic PET imaging data, focusing on the arterial input function (AIF) without the need for invasive arterial cannulation. METHODS: Our method utilizes a combination of three-dimensional depth-wise separable convolutional layers and a physically informed deep neural network to incorporatea priori knowledge about the AIF's functional form and shape, enabling precise predictions of the concentrations of [11C]PBR28 in whole blood and the free tracer in metabolite-corrected plasma. RESULTS: We found a robust linear correlation between our model's predicted AIF curves and those obtained through traditional, invasive measurements. We achieved an average cross-validated Pearson correlation of 0.86 for whole blood and 0.89 for parent plasma curves. Moreover, our method's ability to estimate the volumes of distribution across several key brain regions - without significant differences between the use of predicted versus actual AIFs in a two-tissue compartmental model - successfully captures the intrinsic variability related to sex, the binding affinity of the translocator protein (18 kDa), and age. CONCLUSIONS: These results not only validate our method's accuracy and reliability but also establish a foundation for a streamlined, non-invasive approach to dynamic PET data quantification. By offering a precise and less invasive alternative to traditional quantification methods, our technique holds significant promise for expanding the applicability of PET imaging across a wider range of tracers, thereby enhancing its utility in both clinical research and diagnostic settings.

BACKGROUND: Focused ultrasound (FUS) combined with microbubbles enables transient and noninvasive blood-brain barrier (BBB) opening, facilitating targeted drug delivery. However, accurate treatment planning remains difficult due to inter-patient anatomical variability and the common assumption in simulations that brain tissues behave like water. PURPOSE: To develop and evaluate MODFUS ( M O D e l t o p r e c i s e l y d e s i g n F o c u s e d U l t r a S o u n d e x p e r i m e n t s ), an in silico acoustic simulation framework that integrates a high-resolution anatomical head model to quantify the impact of intracranial tissue heterogeneity and probe alignment on transmitted acoustic pressure, supporting treatment planning for BBB opening. METHODS: MODFUS integrates an anatomical head model comprising 115 tissue types and Computed Tomography (CT)-derived skull properties. FUS simulations were conducted using a 250 kHz single-element transducer at 13 distinct skull entry locations. Two modeling approaches were evaluated: a classical model, in which the skull was embedded in water and brain tissues were homogenized as water, and an heterogeneous anatomical model. To further assess robustness, an additional set of 50 simulations introduced controlled perturbations in probe positioning, consisting of angular deviations ( ± 18 ∘ ) and translational offsets ( ± 7 mm) relative to the reference configuration. Model- and configuration-dependent differences were quantified using peak positive pressure (PPP), peak negative pressure (PNP), and potential therapeutic volume. Statistical significance was assessed using Wilcoxon rank-sum or Wilcoxon signed-rank tests ( α = 0.05). For Classical vs Realistic models, Wilcoxon rank-sum tests were applied to PPP, PNP, and BBB exposure volume, and Levene's test assessed variance differences across 13 positions. Multiple testing was controlled using the Holm-Bonferroni procedure ( α = 0.05) across all tests. Effect sizes were quantified using Cohen's d with 95% confidence intervals. RESULTS: Compared to the classical "skull+water" benchmark model, the heterogeneous model predicted up to 11% lower PPP, slightly lower PNP ( ∼ -5%), and 35% smaller potential BBB exposure volumes across the 13 paired sonication positions. After Bonferroni correction, Paired statistical testing (Wilcoxon signed-rank, two-sided) showed significant differences for PPP (p = 0.0270) and potential BBB exposure volume (p = 0.0015), while differences in PNP were not statistically significant (p = 0.8286). Levene's test for variance confirmed significant heteroscedasticity for PPP (p = 0.027) and PNP (p = 3.2 × 10), but not for BBB exposure volume (p = 0.2680). Cohen's d effect sizes indicated a large positive effect for PPP, a small negative effect for PNP, and a very large positive effect for BBB exposure volume. CONCLUSIONS: MODFUS demonstrates the influence of incorporating detailed tissue heterogeneity on simulation outcomes, including pressure distribution and potential BBB exposure volume. These results highlight the importance of realistic soft tissue modeling and stereotaxic probe alignment for safe and effective FUS treatment planning. The study serves as a preliminary proof-of-concept. Future studies incorporating in vivo experiments will be required to quantify the accuracy of this approach.

The CROSSBRAIN EU project aims to address the heterogeneous nature of brain pathologies by developing wireless implantable microbots (µBots, planned dimensions 100 × 100 × 100 μm3) for highly localized neuromodulation. These devices are designed to precisely modulate brain activity with minimal invasiveness, enabling targeted resolution of specific spatiotemporal events, capabilities not currently achieved by existing neuromodulation technologies. A crucial step involves visualizing and ensuring the optimal placement of the µBots in the brain tissue, to study their functionality after implantation. In this preliminary ex vivo study, we used non-functional µBot silicon (Si) dummies matching the lateral dimensions of the intended µBots, with reduced thickness (100 × 100 × 50 μm3) to simplify fabrication and handling. Due to the intrinsic MRI incompatibility of the µBot platform, encompassing both the dummies used in this study and the future functional devices under development, and the limitations of standard histological approaches in reliably identifying and preserving the implant site during processing, we developed an integrated imaging workflow combining 2D and 3D techniques. While standard histological methods and tissue clearing presented substantial limitations in preserving the position of the dummies within the brain tissue, combining histological techniques with 3D X-ray tomography provided a robust strategy. In particular, synchrotron radiation-based X-ray Phase Contrast Tomography (XPCT), with its intrinsic high contrast and resolution, enabled detailed visualization of dummies within the surrounding vascular and cellular architecture. In contrast, conventional micro-Computed Tomography (micro-CT), although more widely accessible, enabled non-destructive guidance for targeted sectioning. Importantly, and in line with the scope of a Brief Research Report, this study presents a preliminary but technically robust investigation conducted within the CROSSBRAIN project, aimed at identifying and establishing an optimized imaging strategy for the visualization of implanted µBots in brain tissue. This methodological framework is intended as an initial step toward future in vivo studies, in which the validated imaging pipeline will be applied to track both dummy and functional devices and to enable subsequent evaluation of foreign body response under physiologically relevant conditions. This ex vivo workflow therefore provides the essential technical foundation for such future investigations and supports the clear positioning of this work as a feasibility and optimization study. This approach could be particularly valuable for new generations of implantable technologies incompatible with MRI and could support future development of personalized neuromodulation therapies by enabling precise device localization and structural tissue assessment.

In everyday behaviour, the ability to stop an already initiated action is critical for ensuring both your safety and that of others; for example, when stopping a reaching movement towards a hot stove-top after realising it is hot. Neuroscientific evidence points towards the critical role of several regions in the right prefrontal cortex in the coordination and execution of this response inhibition-specifically the right inferior frontal gyrus (rIFG) and the right dorsolateral prefrontal cortex (rDLPFC). The present study investigated the effects of different transcranial magnetic stimulation (TMS) protocols on stop-signal task (SST) performance. We hypothesized that TMS over one or both of these areas would be detrimental to performance. However, contrary to our hypothesis, TMS significantly facilitated performance regardless of the stimulation condition. We applied both frequentist and Bayesian methods to assess the robustness of these effects, revealing consistent reductions in stop-signal reaction time (SSRT) across active conditions. Our results add to the growing body of results that suggest TMS effects may not be as straight-forward as usually assumed and that so-called "inhibitory protocols" can facilitate performance. This result could be explained by a shift in the signal-to-noise ratio depending on the pre-activation of the area. Put differently, TMS may have primed task-related activity in the target areas to a level that was optimal for task performance. Alternatively, the observed effect may reflect an (over)compensation by other parts of the network or disruption of competing resources. Future studies may provide further support for these hypotheses.

Neuromodulation techniques have emerged as transformative tools for treating several neurological and psychiatric disorders, offering alternatives to traditional pharmacological approaches often hindered by the blood-brain barrier and off-target effects. While conventional modalities like deep brain stimulation, transcranial magnetic stimulation, and optogenetics have shown promise, they each face limitations in invasiveness, spatial resolution, or clinical applicability. In recent years, low-intensity ultrasound has gained attention as a noninvasive, deep-penetrating modality capable of modulating brain circuits with millimeter-sized spatial precision. This review explores the synergistic integration of ultrasound with engineered nanosystems to achieve multimodal neuromodulation such as electrical, mechanical, optical, and chemical via acoustic energy. We describe how nanoscale materials can transduce ultrasound into localized bioelectric signals, mechanical stress, light emission, or controlled drug release. These interactions enable precise, on-demand stimulation or inhibition of neuronal activity, including in deep brain regions. Experimental studies have demonstrated neuromodulatory effects across a variety of models, with applications ranging from optogenetics and drug delivery to behavioral modulation in rodents and primates. The review concludes with a critical assessment of the translational challenges such as nanoparticle delivery, biocompatibility, long-term clearance, and safety thresholds for human use while outlining promising strategies like cell-based delivery, biodegradable materials, and closed-loop control systems. These innovations highlight the potential of ultrasound-assisted nanosystems as transformative tools for precise neuromodulation in both experimental neuroscience and clinical applications.

Demyelinating diseases, such as multiple sclerosis, result from the progressive loss of myelin sheaths in the central and peripheral nervous systems, leading to impaired neural conduction and disability. Current disease-modifying therapies focus on immunosuppression to limit inflammation but fail to restore lost myelin. This lack of regenerative capacity underscores the need for strategies that actively promote remyelination. Recent advances highlight neuromodulation, and in particular low-intensity ultrasound (US), as a promising approach to stimulate both neuronal activity and glial responses essential for myelin repair. Ultrasound noninvasively promotes remyelination through complementary mechanisms: indirectly, by enhancing activity-dependent myelination via neuronal firing, and directly, by exerting mechanical bioeffects on oligodendrocyte precursor cells, oligodendrocytes, astrocytes, microglia, and Schwann cells. Experimental studies show US activation of key signaling cascades (PI3K/Akt, MAPK/ERK, NF-κB, TGF-β1), promoting oligodendrocyte survival, differentiation, and myelin repair, alongside microglial polarization, astrocytic neurotrophic support, and functional recovery in central and peripheral models. Converging data from neuromodulation research indicate that activation of cholinergic and noradrenergic circuits-such as those engaged by vagus nerve stimulation-can enhance OPC differentiation, attenuate neuroinflammation, and support remyelination, raising the possibility that ultrasound-based stimulation of these pathways may synergistically amplify regenerative outcomes while avoiding the need for implanted devices. Ultrasound holds transformative potential for central nervous system repair and can also promote regenerative processes in the peripheral nervous system. Cutting-edge ultrasound technologies enable noninvasive penetration of the skull, precise modulation of deep brain circuits with millimeter accuracy, and fine temporal control without inducing systemic side effects. When combined with advanced imaging techniques (e.g., MR-guided US), ultrasound achieves increasingly effective therapeutic outcomes by enhancing beam-targeting precision and enabling real-time monitoring. Moreover, emerging approaches such as sonogenetics and magneto-acoustic stimulation further expand its specificity and therapeutic potential. Collectively, current evidence establishes therapeutic ultrasound as a transformative, noninvasive strategy for treating demyelinating diseases.

OBJECTIVE: The objective of this study is to present a report from the Big Multiple Sclerosis Data (BMSD) statistics workshop (Bari - Italy, June 2023) which focused on advanced statistical approaches for real-world data (RWD) analyses in multiple sclerosis (MS). The report emphasises the application of these approaches in predicting individual treatment response, assessing comparative effectiveness and safety of therapies and their sequences, and harmonizing data for large-scale federated analyses. METHODS: The BMSD network, comprising five national registries and the international MSBase database (> 350,000 total patients), convened in June 2023 in Bari (Italy) to review methodological advances in RWD analysis. Experts discussed strengths, limitations, and regulatory implications of frequentist, Bayesian, and machine learning (ML) approaches, with case studies on treatment response modelling, comparative effectiveness, safety surveillance, and Common Data Model (CDM)-based federated learning. RESULTS: Bayesian and ML techniques, integrated with causal inference frameworks, can improve personalized predictions of treatment benefit and risk by using high-dimensional longitudinal data. Propensity score-based methods and marginal structural models remain essential for minimizing confounding in comparative analyses, but require rigorous diagnostics and sensitivity analyses. Adoption of a CDM facilitates harmonization of heterogeneous datasets, while federated learning enables privacy-preserving, multi-jurisdictional collaboration. Together, these innovations address key challenges in studying treatment sequences, rare adverse events, and underrepresented patient groups. CONCLUSIONS: This workshop report highlights how advanced statistical and computational methodologies enhance the robustness, interpretability, and regulatory relevance of MS RWD studies. By promoting the integration of complementary statistical and computational approaches within harmonized data infrastructures, the BMSD network is positioned to accelerate the translation of real-world evidence into precision medicine for MS.

Global Navigation Satellite System (GNSS) meteorology has emerged as a valuable tool for atmospheric monitoring, providing high-resolution, near-real-time data that can significantly improve precipitation nowcasting. This study aims to enhance short-term precipitation forecasting by integrating GNSS-derived Precipitable Water Vapor (PWV)-a key indicator of atmospheric moisture-with traditional meteorological observations. A novel two-step machine learning framework is proposed that combines a Random Forest (RF) model and a Long Short-Term Memory (LSTM) neural network. The RF model first estimates current precipitation based on PWV, surface weather parameters, and auxiliary atmospheric variables. Then, the LSTM network leverages temporal dependencies within the data to predict precipitation for the subsequent hour. This hybrid method capitalizes on the RF's ability to model complex nonlinear relationships and the LSTM's strength in handling time series data. The results demonstrate that the proposed approach improves forecasting accuracy, particularly during extreme weather events such as intense rainfall and thunderstorms, outperforming conventional models. By integrating GNSS meteorology with advanced machine learning techniques, this study offers a promising tool for meteorological services, early warning systems, and disaster risk management. The findings highlight the potential of GNSS-based nowcasting for real-time decision-making in weather-sensitive applications.

Successful learning often emerges through social interaction: what are the neural and behavioral systems that support this process? This ecological, multimodal study combines functional near-infrared spectroscopy hyperscanning with detailed behavioral and physiological measures in 27 unconstrained social learning interactions. Learning was supported by teacher-learner interbrain synchrony (interpersonal neural synchrony), over regions important for mutual understanding (temporoparietal junction) and communicative coordination (ventral premotor cortex). Joint attention and mutual gaze modulated the interpersonal neural synchrony-learning association in oppositive ways, motivating a dual-process model: during knowledge-building phases, learning is supported by informational uptake dynamic, with high joint attention, low interpersonal neural synchrony in regions for mutual understanding (temporoparietal junction) and coordination (right ventral premotor cortex), and high interpersonal neural synchrony in language-related areas (left ventral premotor cortex). In contrast, during moments of mutual grounding, learning is supported by high mutual gaze and high interpersonal neural synchrony over temporoparietal junction. Cross-brain general linear modeling revealed asymmetric neural dependencies linked to speaking and teaching roles in the left-hemisphere language network. These effects remained after controlling for nodding, gaze, and breathing, indicating that interpersonal neural synchrony reflects true social-cognitive alignment beyond sensorimotor coupling. Taken together, this study shows that successful learning arises from coordinated and nonlinear brain-body dynamics and positions interpersonal neural synchrony as a marker of mutual prediction during communicative social interaction.

Social neuroscience research investigates the neural basis of social cognition, which encompasses various cognitive abilities involved in processing social information, such as empathy. Empathy is a key function driving human socialization and can be defined as the ability to perceive others' feelings. When studying the neural basis of this process, recent research has highlighted the limitations of the traditional approach, which focuses on the study of the brain activity of a single subject engaged in a social interaction. To gain a more comprehensive understanding of social behavior, neuroscience research has increasingly adopted the hyperscanning technique that allows the simultaneous recording of neural activity in two or more interacting individuals. By considering interacting participants as a unique system, hyperscanning provides a deeper insight into the dynamic and reciprocal nature of social interactions, offering a more accurate representation of social behavior, including empathy. In the present study, we investigated empathy (i.e., the emotional responses elicited by positive and negative events) modulated by familiarity (friends or strangers) in two participants engaged in a computer-based game, while EEG data were simultaneously collected in both participants. EEG-hyperscanning data were analyzed using spectral analysis in the source domain and the results showed that both participants exhibited overlapping activations in brain regions related to the processing of negative events (insula). In addition, the agent observing the game manifested activations in areas associated with social cognition and typically involved during the evaluation of emotional and psychological states of others such as the Medial Prefrontal Cortex and the temporal lobes.Clinical RelevanceThe present study can help us to better understand the underlying social deficits that characterize psychopathological conditions by providing insights into the neurocognitive process of empathy and social evaluations systems.

Artificial Intelligence (AI) applications and Machine Learning (ML) methods have gained much attention in recent years for their ability to automatically detect patterns in data without being explicitly taught rules. Specific features characterise the ECGs of patients with Brugada Syndrome (BrS); however, there is still ambiguity regarding the correct diagnosis of BrS and its differentiation from other pathologies. This work presents an application of Echo State Networks (ESN) in the Recurrent Neural Networks (RNN) class for diagnosing BrS from the ECG time series. 12-lead ECGs were obtained from patients with a definite clinical diagnosis of spontaneous BrS Type 1 pattern (Group A), patients who underwent provocative pharmacological testing to induce BrS type 1 pattern, which resulted in positive (Group B) or negative (Group C), and control subjects (Group D). One extracted beat in the V2 lead was used as input, and the dataset was used to train and evaluate the ESN model using a double cross-validation approach. ESN performance was compared with that of 4 cardiologists trained in electrophysiology. The model performance was assessed in the dataset, with a correct global diagnosis observed in 91.5 % of cases compared to clinicians (88.0 %). High specificity (94.5 %), sensitivity (87.0 %) and AUC (94.7 %) for BrS recognition by ESN were observed in Groups A + B vs. C + D. Our results show that this ML model can discriminate Type 1 BrS ECGs with high accuracy comparable to expert clinicians. Future availability of larger datasets may improve the model performance and increase the potential of the ESN as a clinical support system tool for daily clinical practice.

OBJECTIVE: To explore the relationship between dopaminergic denervation and motor impairment in two de novo Parkinson's disease (PD) cohorts. METHODS: n = 249 PD patients from Parkinson's Progression Markers Initiative (PPMI) and n = 84 from an external clinical cohort. Clustering analysis stratified dopaminergic denervation, measured with 123I-FP-CIT-SPECT, and motor impairment into mild [D and M] and severe [D+ and M+]. Differences in terms of biomarkers and clinical progression were assessed across subgroups. Causal mediation analysis evaluated the effect of co-pathology on the relationship between subgroups and cognitive decline. RESULTS: Four subgroups were identified. Two subgroups showed concordant profiles: the severe dopaminergic and motor impairment subgroup [D+/M+] exhibited poorer memory performance, pathological Aβ1-42, as well as higher longitudinal Levodopa equivalent daily dose (LEDD) values and faster progression of motor disability; the mild dopaminergic and motor deficits [D/M] subgroup displayed a benign clinical profile and stable disease progression. Two subgroups exhibited dopaminergic and motor severity mismatch: the mild dopaminergic but severe motor impairment [D/M+] subgroup showed severe and rapidly progressive rigidity. CSF Aβ1-42 levels mediated the association between D+/M+ and cognitive decline in patients who were cognitively preserved at onset, accounting for 13% of the total effect. The external cohort supported the malignancy of D+/M+ and the presence of rigidity in D/M+. INTERPRETATION: Concordant severe impairment reflects a malignant profile linked to Aβ-related cognitive decline, while mild concordant cases show stable progression. Mismatch subgroups display distinct clinical patterns, underscoring the value of integrating imaging and motor features for early disease stratification.

OBJECTIVE: To deepen the underlying pathophysiology of gait impairment in early-stage Parkinson's disease (PD) through a multimodal assessment, including MR-based brainstem morphometry and EEG functional connectivity (FC). METHODS: 73 PD patients and 60 healthy controls (HC) were included. The MR Parkinsonism Index (MRPI) was calculated for each subject. FC was measured using HD-EEG in θ-α-β-low-γ-high-γ bands. Partial Least Squares Path Modeling (PLS-PM) was used to explore the causal relationships between MRPI, FC, and clinical data. RESULTS: MRPI was higher in PD than HC (p = 0.001) and correlated with MDS-UPDRS-III gait subscore (r = 0.54,p < 0.001). α-FC was significantly lower in PD than HC (t = -5.4,p < 0.001). α-FC in PD was negatively correlated with both MRPI (r = -0.31,p = 0.007) and gait subscore (r = -0.31,p = 0.007). PLS-PM revealed MRPI's negative direct effect on α-FC (-0.59,p < 0.001) and significant positive total effect on gait performance (0.53,p < 0.001), partially mediated by α-FC (0.19, p = 0.006). Additionally, α-FC had a direct negative impact on gait performance (-0.33,p = 0.004). CONCLUSIONS: The effect of MRPI on gait performance, partially mediated by α-FC, highlights a dual pattern in which brainstem structural alterations show an association with gait impairment both directly and indirectly through their relationship with cortical functional disruptions. SIGNIFICANCE: These findings support a multimodal framework for understanding and addressing gait disorders in PD, potentially opening avenues for personalized treatment.

The creation of synthetic medical data that truly captures the statistical distribution of real-world patient information, while simultaneously protecting individual privacy, remains a formidable challenge for the clinical and scientific community. This challenge is especially pronounced in nuclear medicine research, where rigorous data sharing is hindered by tight regulations and ethical considerations. In this study, we introduce a multimodal deep learning model designed to reconstruct (and thereby enable future synthesis of) 123I-FPCIT SPECT (DaT) scans by leveraging both the target DaT and co-registered T1-weighted MRI scans. Through extensive experimentation on a large Parkinson's Progression Markers Initiative (PPMI) dataset, comprising healthy controls, Parkinson's disease (PD) patients, and individuals without imaging evidence of dopaminergic deficits (SWEDD), we demonstrate that the proposed framework yields DaT images that strongly preserve clinical signal distributions. Our findings show minimal intensity discrepancies, unbiased contrast-to-noise ratios, and robust region-based analyses across pathological and demographic subgroups, highlighting the feasibility of this approach for large-scale data augmentation in neurodegenerative research.Clinical Relevance-By enabling the reconstruction of DaT scans from multi-contrast inputs, this framework has the potential to enhance future generation of synthetic data that are applicable to early PD detection, disease progression studies, and model training in clinical scenarios where data are often limited, and privacy constraints are stringent.

INTRODUCTION: Neuroinflammation, a pathophysiological process involved in numerous disorders, is typically imaged using [11C]PBR28 (or TSPO) PET. However, this technique is limited by high costs and ionizing radiation, restricting its widespread clinical use. MRI, a more accessible alternative, is commonly used for structural or functional imaging, but when used using traditional approaches has limited sensitivity to specific molecular processes. This study aims to develop a deep learning model to generate TSPO PET images from structural MRI data collected in human subjects. METHODS: A total of 204 scans, from participants with knee osteoarthritis (n = 15 scanned once, 15 scanned twice, 14 scanned three times), back pain (n = 40 scanned twice, 3 scanned three times), and healthy controls (n = 28, scanned once), underwent simultaneous 3 T MRI and [11C]PBR28 TSPO PET scans. A 3D U-Net model was trained on 80% of these PET-MRI pairs and validated using 5-fold cross-validation. The model's accuracy in reconstructed PET from MRI only was assessed using various intensity and noise metrics. RESULTS: The model achieved a low voxel-wise mean squared error (0.0033 ± 0.0010) across all folds and a median contrast-to-noise ratio of 0.0640 ± 0.2500 when comparing true to reconstructed PET images. The synthesized PET images accurately replicated the spatial patterns observed in the original PET data. Additionally, the reconstruction accuracy was maintained even after spatial normalization. DISCUSSION: This study demonstrates that deep learning can accurately synthesize TSPO PET images from conventional, T1-weighted MRI. This approach could enable low-cost, noninvasive neuroinflammation imaging, expanding the clinical applicability of this imaging method.

BACKGROUND AND PURPOSE: In multiple sclerosis, the choroid plexus is thought to promote and sustain the disease immunopathologic inflammatory process. However, its association with cortical pathology and disease progression is still uncertain. We aimed to characterize choroid plexus enlargement and evolution in MS and its relationship with imaging markers of compartmentalized inflammation and clinical outcome. MATERIALS AND METHODS: One hundred patients with MS (74 relapsing-remitting [RRMS]; 26 secondary-progressive MS [SPMS] and 41 matched healthy controls (HC) underwent 3T MRI to measure brain volumes and 7T MRI for cortical and WM lesions, including paramagnetic rim lesion (PRL) segmentation. Choroid plexus volumes of the lateral ventricles were manually edited from FreeSurfer segmentation by 1 neurologist and normalized by total intracranial volume to estimate the choroid plexus ratio (CPR). The Expanded Disability Status Scale was assessed at baseline and longitudinally in 71 patients, 43 of whom had a follow-up MRI at 1.5 years. An extreme gradient boosting algorithm estimated the importance of CPR, alongside other imaging markers, in predicting disease phenotype (RRMS versus SPMS) and progression independent of relapse activity. RESULTS: Relative to HC, CPR was higher in patients (11.66 [3.21] versus 14.98 [4.37] × 10-4, P = .001), both in RRMS and SPMS (P = .007, P = .003), and there was a mean ∼4% increase per year in CPR at follow-up, though it was not significant[ (P = .14). Choroid plexus enlargement correlated with greater PRL and cortical lesion volumes (r = 0.35, P < .001). CPR did not discriminate MS phenotype but predicted progression independent of relapse activity alongside CSF subarachnoid space volume, PRL, and cortical lesion volumes (median [median absolute deviation] area under the curve 0.71 [0.12], accuracy 0.74 [0.10], sensitivity 0.81 [0.12], and specificity of 0.52 [0.22]). CONCLUSIONS: In MS, a CPR increase relates to imaging markers of compartmentalized disease activity including cortical lesions and PRLs and is a critical predictor of disease progression. Our findings could provide the rationale for implementing CPR estimation for prognosis prediction in MS.

Functional Magnetic Resonance Imaging (fMRI) serves as a unique non-invasive tool for investigating brain function by analyzing blood oxygenation level-dependent (BOLD) series. These signals result from the complex interplay between deterministic and stochastic components underpinning biological brain activity. In this context, the quantification of the stochastic component, here defined as brain noise, is challenging without making assumptions on the deterministic dynamics. Leveraging on Approximate Entropy, in this study we present a methodological framework aimed to estimate intrinsic stochastic brain dynamics through fMRI data analysis without making assumption on the deterministic model. We estimated brain noise from fMRI series of 200 participants from the publicly available Cam-CAN dataset, aiming to quantify the amount of stochastic dynamics in different brain regions. Moreover, we hypothesize that a functional relationship exists between intrinsic brain noise and subject's age. Results indicate that a significant part - approximately 18% to 60% - of the fMRI signal power can be attributed to the intrinsic stochastic dynamics within the brain, and a linear augmentation is reported in association with the maturation process. These findings underscore the physiological importance of characterizing neural noise and its unique distributions across various brain regions.

Functional magnetic resonance imaging (fMRI) is a powerful non-invasive method for studying brain function by analyzing blood oxygenation level-dependent (BOLD) signals. These signals arise from intricate interplays of deterministic and stochastic biological elements. Quantifying the stochastic part is challenging due to its reliance on assumptions about the deterministic segment. We present a methodological framework to estimate intrinsic stochastic brain dynamics in fMRI data without assuming deterministic dynamics. Our approach utilizes Approximate Entropy and its behavior in noisy series to identify and characterize dynamical noise in unobservable fMRI dynamics. Applied to extensive fMRI datasets (645 Cam-CAN, 1086 Human Connectome Project subjects), we explore lifelong maturation of intrinsic brain noise. Findings indicate 10% to 60% of fMRI signal power is due to intrinsic stochastic brain elements, varying by age. These components demonstrate a physiological role of neural noise which shows a distinct distributions across brain regions and increase linearly during maturation.

BACKGROUND: The autonomic response to transcutaneous auricular vagus nerve stimulation (taVNS) has been linked to the engagement of brainstem circuitry modulating autonomic outflow. However, the physiological mechanisms supporting such efferent vagal responses are not well understood, particularly in humans. HYPOTHESIS: We present a paradigm for estimating directional brain-heart interactions in response to taVNS. We propose that our approach is able to identify causal links between the activity of brainstem nuclei involved in autonomic control and cardiovagal outflow. METHODS: We adopt an approach based on a recent reformulation of Granger causality that includes permutation-based, nonparametric statistics. The method is applied to ultrahigh field (7T) functional magnetic resonance imaging (fMRI) data collected on healthy subjects during taVNS. RESULTS: Our framework identified taVNS-evoked functional brainstem responses with superior sensitivity compared to prior conventional approaches, confirming causal links between taVNS stimulation and fMRI response in the nucleus tractus solitarii (NTS). Furthermore, our causal approach elucidated potential mechanisms by which information is relayed between brainstem nuclei and cardiovagal, i.e., high-frequency heart rate variability, in response to taVNS. Our findings revealed that key brainstem nuclei, known from animal models to be involved in cardiovascular control, exert a causal influence on taVNS-induced cardiovagal outflow in humans. CONCLUSION: Our causal approach allowed us to noninvasively evaluate directional interactions between fMRI BOLD signals from brainstem nuclei and cardiovagal outflow.

Efficient strategies to promote microvascularization in vascular tissue engineering, a central priority in regenerative medicine, are still scarce; nano- and micro-sized aggregates and spheres or beads harboring primitive microvascular beds are promising methods in vascular tissue engineering. Capillaries are the smallest type and in numerous blood vessels, which are distributed densely in cardiovascular system. To mimic this microvascular network, specific cell components and proangiogenic factors are required. Herein, advanced biofabrication methods in microvascular engineering, including extrusion-based and droplet-based bioprinting, Kenzan, and biogripper approaches, are deliberated with emphasis on the newest works in prevascular nano- and micro-sized aggregates and microspheres/microbeads.

Major surgeries can result in high rates of adverse postoperative events. Reliable prediction of which patient might be at risk for such events may help guide peri- and postoperative care. We show how archiving and mining of intraoperative hemodynamic data in orthotopic liver transplantation (OLT) can aid in the prediction of postoperative 180-day mortality and acute renal failure (ARF), improving upon predictions that rely on preoperative information only. From 101 patient records, we extracted 15 preoperative features from clinical records and 41 features from intraoperative hemodynamic signals. We used logistic regression with leave-one-out cross-validation to predict outcomes, and incorporated methods to limit potential model instabilities from feature multicollinearity. Using only preoperative features, mortality prediction achieved an area under the receiver operating characteristic curve (AUC) of 0.53 (95% CI: 0.44-0.78). By using intraoperative features, performance improved significantly to 0.82 (95% CI: 0.56-0.91, P = 0.001). Similarly, including intraoperative features (AUC = 0.82; 95% CI: 0.66-0.94) in ARF prediction improved performance over preoperative features (AUC = 0.72; 95% CI: 0.50-0.85), though not significantly (P = 0.32). We conclude that inclusion of intraoperative hemodynamic features significantly improves prediction of postoperative events in OLT. Features strongly associated with occurrence of both outcomes included greater intraoperative central venous pressure and greater transfusion volumes.

This study proposes an automated approach to radiotherapy treatment planning by integrating a reinforcement-learning-style iterative framework with a multimodal Large Language Model (LLM). We specifically investigate the problem of Beam Angle Optimization, a high-dimensional and non-convex subproblem of Treatment Planning. Our system employs GPT-4V to select candidate beam angles and analyze three-dimensional dose distributions generated by Monte Carlo simulations within the MatRAD environment. Iterative plan refinement is guided by a reward function that encourages target dose conformity and penalizes excessive dose to organs at risk. We incorporate exploration-exploitation principles to strike a balance between investigating diverse action proposals and refining promising solutions. Experimental results on prostate cancer cases demonstrate that our LLM-based framework offers superior performance compared to random beam selection and can outperform the quality of deep reinforcement learning baselines, indicating the potential for LLMs to assist in complex radiotherapy treatment planning tasks.Clinical relevance-This approach is designed to alleviate the significant effort of manual treatment planning by assisting medical physicists in exploring beam configurations and systematically refining plans to improve dose coverage and protect healthy tissues.

Brain metastases (BM), along with primary central nervous system lymphomas and glioblastomas, represent the majority of malignant brain tumors encountered in clinical neuro-oncology, driving a need for advanced imaging techniques and post-processing methods to improve their characterization and treatment monitoring. In particular, stereotactic radiosurgery (SRS), a cornerstone treatment for BM, delivers high-dose, focused radiation (>20 Gy) to target lesions with minimal impact on surrounding tissues. Despite its efficacy, radiation-induced effects such as early radiation effects (ERE) and adverse radiation effects (ARE) complicate diagnosis and management, with ARE occurring in up to 30% of patients, often presenting as ring-enhancing T2/FLAIR hyperintensities. To address these challenges, we aimed to compare standard radiomics-based machine learning approaches with pretrained generative models for assessing ERE in BM lesions. A cohort of 21 patients for a total of 35 lesions (17 treatment-naïve and 18 post-SRS +/- combination therapy) who underwent multiparametric 18F-FPIA PET/MRI was analyzed. The study investigated: 1) Multiparametric analysis of PET and MRI diffusion/perfusion parameters (ADC, Ktrans, CBF, K1, vt); 2) MRI-based radiomics; 3) static PET radiomics; 4) Dynomics; 5) a combination of PET and MRI radiomics; and 6) low-level embeddings from a pretrained generative diffusion model applied to full T1, static PET, and their combination. Using manually contoured lesion masks for analyses 1-5 and lesion-free embeddings for analysis 6, multiple classifiers (SVM, XGBoost, Linear regressor) were applied after feature standardization and principal component analysis (retaining 90% variance). Fivefold cross-validation demonstrated comparable performances across radiomic approaches (Accuracy: 71.95±0.05%, AUC: 0.72±0.05%), while the pretrained generative model achieved significantly higher performance (Accuracy: 83.82±0.01%, AUC: 0.83±0.01%) without requiring lesion segmentation in assessing ERE in BM lesions.Clinical Relevance-This study shows the potential of generative models to streamline and enhance the assessment of early radiation effects in parenchymal metastatic lesions without need of lesion segmentation.

Radiotherapy is a crucial cancer treatment that demands precise planning to balance tumor eradication and preservation of healthy tissue. Traditional treatment planning (TP) is iterative, time-consuming, and reliant on human expertise, which can potentially introduce variability and inefficiency. We propose a novel framework to transform a large multimodal foundation model (MLM) into an action model for TP using a few-shot reinforcement learning (RL) approach. Our method leverages the MLM's extensive pre-existing knowledge of physics, radiation, and anatomy, enhancing it through a few-shot learning process. This allows the model to iteratively improve treatment plans using a Monte Carlo simulator. Our results demonstrate that this method outperforms conventional RL-based approaches in both quality and efficiency, achieving higher reward scores and more optimal dose distributions in simulations on prostate cancer data. This proof-of-concept suggests a promising direction for integrating advanced AI models into clinical workflows, potentially enhancing the speed, quality, and standardization of radiotherapy treatment planning.

BACKGROUND: Computed tomography scans are widely used in everyday medical practice due to speed, image reliability, and detectability of a wide range of pathologies. Each scan exposes the patient to a radiation dose, and performing a fast estimation of the effective dose (E) is an important step for radiological safety. The aim of this work is to estimate E from patient and CT acquisition parameters in the absence of a dose-tracking software exploiting machine learning. METHODS: In total, 69,037 CT acquisitions were collected with the dose-tracking software (DTS) available at our institution. E calculated by DTS was chosen as the target value for prediction. Different machine learning algorithms were selected, optimizing parameters to achieve the best performance for each algorithm. Effective dose was also estimated using DLP and k-factors, and with multiple linear regression. Mean absolute error (MAE, mean absolute percentage error (MAPE), and R2 were used to evaluate predictions in the test set and in an external dataset of 3800 acquisitions. RESULTS: The random forest regressor (MAE: 0.416 mSv; MAPE: 7%; and R2: 0.98) showed best performances over the neural network and the support vector machine. However, all three machine learning algorithms outperformed effective dose estimation using k-factors (MAE: 2.06; MAPE: 26%) or multiple linear regression (MAE: 0.98; MAPE: 44.4%). The random forest regressor on the external dataset showed an MAE of 0.215 mSv and an MAPE of 7.1%. CONCLUSIONS: Our work demonstrated that machine learning models trained with data calculated by a dose-tracking software can provide good estimates of the effective dose just from patient and scanner parameters, without the need for a Monte Carlo approach.

OBJECTIVE: Although conduct problems (CPs) are continuously distributed, little is known about how dimensional measures of CPs map onto brain structure. Therefore a large sample was used to comprehensively assess associations between dimensionally measured CPs and brain structure. METHOD: T1-weighted structural brain magnetic resonance imaging scans from 14,160 youths (5-21 years old, 46.2% female) across 18 international case-control, community-based, and population-based cohorts were preprocessed using ENIGMA-standardized protocols. Regression models examined associations between CPs and cortical thickness, surface area, and subcortical volumes, adjusting for age, sex, and intracranial volume. Moderation by sex, age, and callous-unemotional traits was also investigated. RESULTS: Widespread but small (β = -0.02 to -0.07) negative associations were observed between CPs and surface area (total surface area, 23/34 regions), cortical thickness (average thickness, 15/34 regions), and amygdalar and hippocampal volumes. Sex was a key moderator, with many surface area associations limited to boys and some thickness associations limited to girls. Some associations were stronger in younger children and at lower levels of callous-unemotional traits. The impact of adjusting for IQ and other psychopathology varied by outcome (eg, most surface area findings survived IQ adjustment, whereas cortical thickness associations did not). CONCLUSION: CPs were associated with subtle, yet widespread, alterations in brain structure. Findings overlapped with differences observed in categorically measured conduct disorder, but novel associations with cortical thickness were identified. This provides further evidence that neuroanatomical differences are not limited to youth with clinically elevated CPs. Our findings have potential implications for neurocognitive models of CPs as they extend beyond the regions highlighted in these models. STUDY REGISTRATION INFORMATION: Investigating dimensional relationships between conduct problems and brain structure: an ENIGMA mega-analysis; https://osf.io/nzj3r/.

BACKGROUND: Externalizing and internalizing disorders are common in youth but are often studied separately, preventing researchers from identifying shared (i.e., transdiagnostic) alterations in brain structure. Using data from the ENIGMA (Enhancing Neuro Imaging Genetics through Meta Analysis) Consortium, we conducted a mega-analysis to identify shared and distinct cortical and subcortical brain alterations across internalizing (anxiety disorders and depression) and externalizing (attention-deficit/hyperactivity disorder [ADHD] and conduct disorder [CD]) disorders in youth. METHODS: 3D T1-weighted magnetic resonance imaging data from youths (age range 4-21 years) with anxiety disorders (n = 1044), depression (n = 504), ADHD (n = 1317), and CD (n = 1172) along with healthy control participants (n = 4743) were analyzed. We assessed group differences in regional cortical thickness, surface area (SA), and subcortical volume using linear models, adjusted for site, age, and sex, as well as total intracranial volume in the SA and subcortical volume models. RESULTS: We observed transdiagnostic associations, with both internalizing and externalizing disorders characterized by lower SA in the insula, entorhinal cortex, and middle temporal gyrus and lower amygdala volume (Cohen's ds = -0.07 to -0.24) as well as total SA and intracranial volume (ds = -0.11 to -0.25). Externalizing-specific reductions in SA were observed in frontoparietal regions (ds = -0.08 to -0.13), but no internalizing-specific associations were identified. Disorder-specific alterations were identified for ADHD, CD, and anxiety disorders but not depression. CONCLUSIONS: Both common and disorder-specific alterations were identified, with regions involved in salience attribution and emotion processing implicated across internalizing and externalizing disorders. These novel findings can guide future research targeting common biological processes across youth psychiatric disorders as well as features unique to individual disorders.

Our aim is to evaluate the effect of a structured stress reduction intervention based on mindfulness during pregnancy on the maternal brain. We report a secondary analysis of IMPACT BCN, a randomized clinical trial including pregnant women randomly allocated to 8-week Mindfulness-Based Stress Reduction (n = 41) or usual care (without any intervention, n = 35). Maternal magnetic resonance (MR) was performed in the third trimester, cluster-wise analysis was used to assess cortical morphometric differences, and proton magnetic resonance spectroscopy (1H-MRS) to evaluate the metabolic characteristics. Mindfulness status was evaluated using the Five Facet Mindfulness Questionnaire (FFMQ). Results showed that participants from Stress reduction group had significantly larger surface areas in the right superior frontal region as compared to the Usual care group (90%CI: 0.023-0.029, p = 0.03). The1H-MRS revealed that Stress reduction group participants, had higher concentrations of myo-inositol (adjusted mean difference D 0.37 mol/L, 95%CI 0.05-0.69) as compared to Usual care. Participants who had high mindfulness on FFMQ facets of non-judgmental (D 358.5 mm2, 95%CI 53.5-663.6) and non-reactivity (D 362.3 mm2, 95%CI 18.8-705.7) had larger right superior frontal area. In conclusion, Mindfulness-Based Stress Reduction program during pregnancy has a significant effect on maternal brain structure and is associated with metabolite concentration changes.

Neuroimaging studies suggest that resilience to adversity is linked to reduced emotional reactivity or enhanced emotion regulation. However, such studies are scarce and mainly use adult samples and categorical definitions of resilience. Using a novel, data-driven approach to define resilience dimensionally, based on cumulative adversity exposure across childhood and psychopathology, we investigated associations between resilience and brain activation during facial emotion processing in youth. We also tested for sex differences in the relationship between resilience and brain activation. fMRI data were acquired from 208 youths (aged 9-18 years; Mean age = 13.28), while viewing angry, fearful, and neutral faces. Whole-brain analyses were performed, followed by region-of-interest analyses focusing on the amygdala, hippocampus, and prefrontal cortex. Resilience was positively correlated with bilateral inferior frontal gyrus responses to fearful (versus neutral) faces, and negatively correlated with right superior temporal gyrus, left hippocampal, and right inferior frontal gyrus responses to neutral faces (versus fixation). Sex-by-resilience interactions were observed in the medial prefrontal cortex: males showed positive, while females showed negative, associations between resilience and brain activation, though these results did not survive correction for multiple comparisons. These findings provide further evidence that resilience in youth is associated with enhanced emotion regulation at a neural level.

BACKGROUND: Functional magnetic resonance imaging studies of conduct disorder (CD) have mostly been limited to males. Here, we examined whether male and female youths with CD showed similar or distinct alterations in brain responses to emotional faces, using a large sample of male and female youths with CD. We also investigated the influence of callous-unemotional (CU) traits. METHODS: Brain responses to angry, fearful, and neutral faces were assessed in 161 youths with CD (74 female) and 241 typically developing (TD) youths (139 female) ages 9 to 18 years. Categorical analyses tested for diagnosis effects (CD vs. TD and CD with high levels of CU traits [CD/HCU] vs. low levels of CU traits [CD/LCU] vs. TD) and sex × diagnosis interactions. RESULTS: When processing faces in general (all faces vs. baseline), youths with CD exhibited lower amygdala responses compared with TD youths, which seemed to be driven by the CD/HCU subgroup. Sex × CU subgroup interactions were identified in the amygdala (CD/LCU females TD males) and anterior insula (CD/HCU females > CD/LCU females; CD/HCU males < CD/LCU males). CONCLUSIONS: The findings for males support an influential neurocognitive model of CD. However, the association between CU traits and brain response to facial expressions differed in females and males with CD, suggesting distinct pathophysiological processes.

Endogenous retroviruses (ERVs) are genetic elements derived from a process of germline infection by exogenous retroviruses. Some ERVs have been co-opted for physiological functions, and their activation has been associated with complex diseases, including Autism Spectrum Disorder (ASD). We have already demonstrated an abnormal expression of ERVs in the BTBR T + tf/J (BTBR) mouse model of ASD during intrauterine life till adulthood. Thus, starting from the assumptions that ERVs may contribute to the derailment of neurodevelopment and that ASD has fetal origins as a consequence of adverse intrauterine conditions, the present study aims to characterize the transcriptional activity of selected ERVs (MusD, IAP, Syn-A, Syn-B, ARC and GLN), LINE-1, inflammatory mediators (IL-6, IL-10, IL-11 CXCL-1) at the maternal-fetal interface and in dissected embryos from BTBR mice. Our results highlight the deregulation of ERVs and inflammatory mediators at the maternal-fetal interface, and in cephalic and non-cephalic embryonic tissues from BTBR compared to C57BL/6 J. Several correlations among ERV expression levels emerged in different tissues from C57BL/6 J mice while, in BTBR mice, no correlations were found, suggesting that in this model, the acquisition of autistic-like traits might be linked to the dysregulation of ERV activity occurring during intra-uterine life.

Resistance to first-line chemotherapies and crizotinib in anaplastic large cell lymphoma (ALCL) represents a significant challenge, often leading to a dismal outcome. Despite recent advancements, the dissection of the intrinsic and extrinsic molecular alterations underlying crizotinib resistance in ALCL is still poorly understood. Here, we transcriptionally unraveled the bidirectional interplay between anaplastic lymphoma kinase (ALK)-driven ALCL (ALK+ ALCL) and stromal cells in the presence of crizotinib at bulk and single-cell levels and identified that the microenvironment provides prosurvival signals leading to crizotinib persistence in ALK+ ALCL. We detected increased B-cell lymphoma 2 (BCL2) expression and downregulation of pathways related to apoptosis in crizotinib-persister ALK+ ALCL cells. Furthermore, we predicted in silico the ligand-receptor interactions between tumoral and stromal cells, supporting their contribution to ALCL pathogenesis mainly participating in the adhesion/membrane transport, triggering receptors, and promoting activation and microenvironment stimulation in lymphoma cells. Finally, we explored the effect of crizotinib in combination with BH3 mimetics. Pharmacologic and genetic ablation of anti-apoptotic targets displayed a significant synergistic effect with crizotinib, overcoming the stroma-mediated protection of lymphoma cells on drug treatment. Thus, BCL2/B-cell lymphoma-extra large (BCL-XL) targeting is synthetic lethal with crizotinib exposure in ALK+ ALCL and represents an intrinsic- and extrinsic-mediated targetable vulnerability in lymphoma cells challenged with crizotinib. Our data support the evaluation of BCL2 targeting in crizotinib-based regimens in the management of patients with ALK+ ALCL.

The intricate interplay between epithelial and fibroblast cells within the tumor microenvironment plays a crucial role in driving triple-negative breast cancer progression. This crosstalk involves the exchange of various signaling molecules, including growth factors, cytokines, extracellular matrix components, and extracellular vesicles. Recently, we demonstrated that triple-negative breast cancer extracellular vesicles carry and release a specific combination of miRs, including miR-185-5p, miR-652-5p, and miR-1246 (from here on, referred as combo-miRs), into normal fibroblasts, effectively reprogramming them into cancer-associated fibroblasts. Here, we show that the conditioned medium from the fibroblasts activated by combo-miRs exerts a pro-tumorigenic effect on epithelial cells, enhancing the viability and migratory potential while driving increased invasiveness in patient-derived breast cancer organoids. A proteomic analysis of conditioned medium from combo-miRs activated fibroblasts revealed 76 significantly upregulated secreted proteins compared to control. Bioinformatic analysis identified the transcriptional factor PATZ1 as a potential regulator of the 12 most highly upregulated proteins. Consistently, in-silico predictions and in vitro experiments confirmed that PATZ1 is a direct target of miR-185-5p and miR-652-5p. The downregulation of PATZ1 by these miRNAs led to increased levels of the secreted proteins in the conditioned medium from combo-miRs activated fibroblasts. Furthermore, the conditioned medium from PATZ1-knockout mesenchymal embryonic fibroblasts and normal fibroblasts with silenced PATZ1 similarly enhanced the migratory potential of MCF10A cells, further supporting the critical role of PATZ1 in regulating tumor-promoting mechanisms. These findings provide valuable insights into the dynamics of the TME in TNBC, highlighting combo-miRs and PATZ1 as promising targets for future therapeutic interventions.

Programmed DNA double-strand break (DSB) formation is a crucial feature of meiosis in most organisms. DSBs initiate recombination-mediated linking of homologous chromosomes, which enables correct chromosome segregation in meiosis. DSBs are generated on chromosome axes by heterooligomeric focal clusters of DSB-factors. Whereas DNA-driven protein condensation is thought to assemble the DSB-machinery, its targeting to chromosome axes is poorly understood. We uncover in mice that efficient biogenesis of DSB-machinery clusters requires seeding by axial IHO1 platforms. Both IHO1 phosphorylation and formation of axial IHO1 platforms are diminished by chemical inhibition of DBF4-dependent kinase (DDK), suggesting that DDK contributes to the control of the axial DSB-machinery. Furthermore, we show that axial IHO1 platforms are based on an interaction between IHO1 and the chromosomal axis component HORMAD1. IHO1-HORMAD1-mediated seeding of the DSB-machinery on axes ensures sufficiency of DSBs for efficient pairing of homologous chromosomes. Without IHO1-HORMAD1 interaction, residual DSBs depend on ANKRD31, which enhances both the seeding and the growth of DSB-machinery clusters. Thus, recombination initiation is ensured by complementary pathways that differentially support seeding and growth of DSB-machinery clusters, thereby synergistically enabling DSB-machinery condensation on chromosomal axes.

There is growing consensus among neuroscientists that neural circuits critical for survival are the result of genomic decompression processes. We introduce SynaptoGen, a novel computational framework--member of the Connectome Models family--bringing synthetic biological intelligence closer, facilitating neural biological agent development through precise genetic control of synaptogenesis. SynaptoGen is the first model of its kind offering mechanistic explanation of synaptic multiplicity based on genetic expression and protein interaction probabilities. The framework connects genetic factors through a differentiable function, working as a neural network where synaptic weights equal average numbers of synapses between neurons, multiplied by conductance, derived from genetic profiles. Differentiability enables gradient-based optimization, allowing generation of genetic expression patterns producing pre-wired biological agents for specific tasks. Validation in simulated synaptogenesis scenarios shows agents successfully solving four reinforcement learning benchmarks, consistently surpassing control baselines. Despite gaps in biological realism requiring mitigation, this framework has potential to accelerate synthetic biological intelligence research.

Nanoparticles (NPs) are a wide class of materials currently used in several industrial and biomedical applications. Due to their small size (1-100 nm), NPs can easily enter the human body, inducing tissue damage. NP toxicity depends on physical and chemical NP properties (e.g., size, charge and surface area) in ways and magnitudes that are still unknown. We assess the average as well as the individual importance of NP atomic descriptors, along with chemical properties and experimental conditions, in determining cytotoxicity endpoints for several nanomaterials. We employ a multicenter cytotoxicity nanomaterial database (12 different materials with first and second dimensions ranging between 2.70 and 81.2 nm and between 4.10 and 4048 nm, respectively). We develop a regressor model based on extreme gradient boosting with hyperparameter optimization. We employ Shapley additive explanations to obtain good cytotoxicity prediction performance. Model performances are quantified as statistically significant Spearman correlations between the true and predicted values, ranging from 0.5 to 0.7. Our results show that i) size in situ and surface areas larger than 200 nm and 50 m2/g, respectively, ii) primary particles smaller than 20 nm; iii) irregular (i.e., not spherical) shapes and iv) positive Z-potentials contribute the most to the prediction of NP cytotoxicity, especially if lactate dehydrogenase (LDH) assays are employed for short experimental times. These results were moderately stable across toxicity endpoints, although some degree of variability emerged across dose quantification methods, confirming the complexity of nano-bio interactions and the need for large, systematic experimental characterization to reach a safer-by-design approach.

While associations between exposure to air pollutants and increased morbidity and mortality are well established, few rigorous studies on this issue are available. The aim of the current study is to implement a new approach to the spatial analysis of mortality and morbidity, based on testing for the presence of the same association in other areas of similar size. Additionally, we perform a case study in Val d'Agri (VA), an area of Basilicata Region, Southern Italy, where oil and natural gas extraction began in 1998. In order to examine the spatial distribution of morbidity and mortality in the region of interest, Hospital discharge (2001-2013) and mortality (2003-2014) rates for the main environment-related diseases were calculated. In addition, a comparison between the period 1980-1998 and the period 1999-2014 was performed for cardiovascular disease mortality. For the period under study, a neutral scenario emerged for cancer and respiratory diseases, where we found no differences in morbidity and mortality as compared to the national benchmark. In some cases significantly lower values (as compared to the nation-wide benchmark) were found. Conversely, a slight excess in morbidity and mortality (as compared to the nation-wide benchmark) emerged for cardiovascular diseases. Still, this excess was common to a number of municipalities in the surroundings of VA, and appeared to be already present in 1980. Higher rates of cardiovascular diseases, lower rates of neoplastic disorders no differences in mortality for respiratory causes (as compared to the nation-wide benchmark) were found in multiple areas of the region, and were therefore not specific to VA. In summary, our data do not support the hypothesis of a role of industrial activities related to oil extraction in VA in determining mortality and morbidity patterns and trends.

There is evidence of adverse health impacts from human exposure to particulate air pollution, including increased rates of respiratory and cardiovascular illness, hospitalizations, and pre-mature mortality. Most recent hypotheses assign an important role to ultrafine particles (UFP) (<0.1 μm) and to associated transition metals (in particular Fe). In a large city like Rome, where many active people spend more than one hour per day in private or public transportation, it may be important to evaluate the level of exposure to harmful pollutants which occurs during urban travelling. In this context, the aim of this work was to examine the relative contribution of different transport modes to total daily exposure. We performed experimental measurements during both morning and evening traffic peak hours throughout the winter season (December 2013-March 2014), for a total of 98 trips. Our results suggest that the lowest UFP exposures are experienced by underground train commuters, with an average number concentration of 14 134 cm-3, and are largely a reflection of the routes being at greater distance from vehicular traffic. Motorcyclists experienced significantly higher average concentrations (73 168 cm-3) than all other exposure classes, and this is most likely a result of the presence of high-concentration and short-duration peaks which do not occur when the same routes are traveled by car. UFP concentrations in subway train environments were found to be comparable to urban background levels. Still, in underground trains we found the highest values of PM10 mass concentration with a maximum value of 422 μg/m3. PM10 concentration in trains was found to be four and two times higher than what was measured in car and motorbike trips, respectively. Transport mode contribution to total integrated UFP daily exposure was found to be 16.3%-20.9% while travelling by car, 28.7% for motorbike trips, and 8.7% for subway trips. Due to lower exposure times, commuting by car and motorbike is comparable to other daily activities in terms of exposure. Our data can provide relevant information for transport decision-making and increase environmental awareness in the hope that the information about inhaled pollutants can translate into a more rational approach to urban travelling.

Recently, interest for the potential impact of consumer-relevant engineered nanoparticles on pregnancy has dramatically increased. This study investigates whether inhaled silver nanoparticles (AgNPs) reach and cross mouse placental barrier and induce adverse effects. Apart from their relevance for the growing use in consumer products and biomedical applications, AgNPs are selected since they can be unequivocally identified in tissues. Pregnant mouse females are exposed during the first 15 days of gestation by nose-only inhalation to a freshly produced aerosol of 18-20 nm AgNPs for either 1 or 4 h, at a particle number concentration of 3.80 × 107 part./cm-3 and at a mass concentration of 640 μg/m³. AgNPs are identified and quantitated in maternal tissues, placentas and foetuses by transmission electron microscopy coupled with energy-dispersive X-ray spectroscopy and single-particle inductively coupled plasma mass spectrometry. Inhalation of AgNPs results in increased number of resorbed foetuses associated with reduced oestrogen plasma levels, in the 4 h/day exposed mothers. Increased expression of pregnancy-relevant inflammatory cytokines is also detected in the placentas of both groups. These results prove that NPs are able to reach and cross the mouse placenta and suggest that precaution should be taken with respect to acute exposure to nanoparticles during pregnancy.

Objective.Brain-to-text systems have recently achieved impressive performance when trained on single-participant data, but remain limited by uninvestigated cross-subject generalization.Approach.We present the first neural-to-phoneme decoder trained jointly on the two largest intracortical speech datasets (Willettet al2023Nature6201031-6; Cardet al2024New Engl. J. Med.391609-18), introducing day- and dataset-specific affine transforms to align neural activity into a shared space. Additionally, a hierarchical GRU decoder with intermediate CTC supervision and feedback connections is designed to address the conditional-independence assumption of standard CTC loss.Main results.Our model matches or outperforms within-subject baselines while being trained across participants, and adapts to unseen subjects using only a linear transform or brief fine-tuning. On an independent inner-speech dataset (Kunzet al2025Cell1884658-4673.e17), our approach shows some initial evidence of generalization, by training only subject-, day-specific transforms.Significance.These results demonstrate the feasibility of cross-subject pretraining as a promising direction toward more scalable speech Brain Computer Interfaces.

Music is a universal phenomenon that influences human experiences across cultures. We investigate whether music can be decoded from human brain activity measured with fMRI, by modeling mappings between neural data and latent representations of musical stimuli. Our approach integrates functional and anatomical alignment techniques to facilitate cross-subject decoding. Starting from the GTZan fMRI dataset, where five participants listened to 540 musical tracks from 10 genres, we used the CLAP model to extract latent representations of the musical stimuli and developed voxel-wise encoding models to identify brain regions responsive to these stimuli, by applying a threshold to the correlation between predicted and actual brain activity. Our decoding pipeline, primarily retrieval-based, employs a linear map to project back brain activity to the corresponding CLAP features. This enables us to retrieve the musical stimuli most similar to those that originated the fMRI data. Our results demonstrate state-of-the-art identification accuracy, outperforming existing approaches.

Brain decoding aims to reconstruct external stimuli from brain activity, providing insights into the neural representation of cognitive experiences. Music decoding from functional magnetic resonance imaging (fMRI) is particularly challenging due to the complexity of auditory processing and the temporal limitations of fMRI signals. In this study, we introduce a novel decoding framework that improves the alignment between fMRI activity and latent musical representations extracted using a pre-trained multimodal model (CLAP). We propose a dual-loss approach combining Optimal Transport and Contrastive Learning to enhance feature mapping and retrieval accuracy. The first loss ensures structural consistency between brain-predicted and true musical embeddings, while the contrastive loss refines the embedding space by maximizing similarities between corresponding pairs and minimizing non-correspondences. Using fMRI data from five subjects listening to music tracks from the GTZAN dataset, our method achieves improved decoding performance, surpassing traditional regression-based approaches from 22.1% top-1 accuracy to 29.3%. These results highlight the potential of integrating Optimal Transport and Contrastive Learning to improve brain decoding performance, paving the way for extending the approach to different sensory domains and applications in Brain-Computer Interfaces (BCI).Clinical relevance- This study could have clinical implications for understanding auditory processing disorders and developing neurorehabilitation strategies. By elucidating how the brain encodes complex auditory stimuli, this approach may contribute to BCI applications for speech and music perception restoration in individuals with hearing impairments or neurological conditions affecting auditory cognition.

Reconstructing music directly from brain activity provides insight into the neural representations underlying auditory processing and paves the way for future brain-computer interfaces. We introduce a fully data-driven pipeline that combines cross-subject functional alignment with bayesian decoding in the latent space of a diffusion-based audio generator. Functional alignment projects individual fMRI responses onto a shared representational manifold, increasing the performance of cross-participant accuracy with respect to anatomically normalized baselines. A bayesian search over latent trajectories then selects the most plausible waveform candidate, stabilizing reconstructions against neural noise. Crucially, we bridge CLAP's multi-modal embeddings to music-domain latents through a dedicated aligner, eliminating the need for hand-crafted captions and preserving the intrinsic structure of musical features. Evaluated on ten diverse genres, the model achieves a cross-subject-averaged identification accuracy of [Formula: see text], and produces audio that human listeners recognize above chance in 85.7% of trials. Voxel-wise analyses locate the predictive signal within a bilateral circuit spanning early auditory, inferior-frontal, and premotor cortices, consistent with hierarchical and sensorimotor theories of music perception. The framework establishes a principled bridge between generative audio models and cognitive neuroscience.

Decoding visual representations from human brain activity has emerged as a thriving research domain, particularly in the context of brain-computer interfaces. Our study presents an innovative method that employs knowledge distillation to train an EEG classifier and reconstruct images from the ImageNet and THINGS-EEG 2 datasets using only electroencephalography (EEG) data from participants who have viewed the images themselves (i.e. "brain decoding"). We analyzed EEG recordings from 6 participants for the ImageNet dataset and 10 for the THINGS-EEG 2 dataset, exposed to images spanning unique semantic categories. These EEG readings were converted into spectrograms, which were then used to train a convolutional neural network (CNN), integrated with a knowledge distillation procedure based on a pre-trained Contrastive Language-Image Pre-Training (CLIP)-based image classification teacher network. This strategy allowed our model to attain a top-5 accuracy of 87%, significantly outperforming a standard CNN and various RNN-based benchmarks. Additionally, we incorporated an image reconstruction mechanism based on pre-trained latent diffusion models, which allowed us to generate an estimate of the images that had elicited EEG activity. Therefore, our architecture not only decodes images from neural activity but also offers a credible image reconstruction from EEG only, paving the way for, e.g., swift, individualized feedback experiments.

Many complex networks exhibit hierarchical, tree-like structures, making hyperbolic space a natural candidate wherein to learn representations of them. Based on this observation, Hyperbolic Graph Neural Networks (HGNNs) have been widely adopted as a principled choice for representation learning on tree-like graphs. In this work, we question this paradigm by proposing the additional condition of geometry--task alignment, i.e., whether the metric structure of the target follows that of the input graph. We theoretically and empirically demonstrate the capability of HGNNs to recover low-distortion representations on regression problems, and show that their geometric inductive bias becomes helpful when the problem requires preserving metric structure. By jointly analyzing predictive performance and embedding distortion, we further show that HGNNs gain an advantage on link prediction, a naturally geometry-aligned task, whereas this advantage largely disappears on standard node classification benchmarks, which are typically not geometry--aligned. Overall, our findings shift the focus from only asking "Is the graph hyperbolic?" to also questioning "Is the task aligned with hyperbolic geometry?", showing that HGNNs consistently outperform Euclidean models under such alignment, while their advantage vanishes otherwise.

Hyperbolic geometry has gained attention for its ability to naturally embed hierarchical and tree-like structures with low distortion, outperforming Euclidean spaces in various graph representation tasks. While previous work has demonstrated the advantage of graph neural networks embedding in hyperbolic space for link prediction and node classification, the benefits for graph classification remain less understood. Moreover, these studies typically attribute the benefits of hyperbolic models to the hierarchical or tree-like nature of graph structures, often neglecting the important role that node features play in leveraging these geometric advantages. With this in mind, we designed an experiment specifically aimed at evaluating the interplay between geometry and node features in graph classification, creating a dataset composed exclusively of tree-structured graphs. Each graph is generated by sampling the number of children per node at each level from a predefined range of branching factors, which varies across levels. The dataset defines two distinct classes based on these branching factor patterns. Node features are either random , structural embeddings obtained via node2vec , or layout-based embeddings derived from kamada-kawai algorithm, which provide a strong hierarchical prior. We evaluated a fully hyperbolic graph neural network against its Euclidean counterpart and a standard Graph Convolutional Network (GCN), using node embeddings learned in low-dimensional latent spaces. Across all feature types, the Euclidean counterpart consistently outperformed the hyperbolic one. This indicates that the benefits of hyperbolic geometry do not arise solely from its alignment with global graph structure. These findings call for a critical reassessment of hyperbolic models in graph classification tasks where preserving graph distances is not essential.

Graph neural networks (GNNs) have demonstrated success in learning representations of brain graphs derived from functional magnetic resonance imaging (fMRI) data. However, existing GNN methods assume brain graphs are static over time and the graph adjacency matrix is known prior to model training. These assumptions contradict evidence that brain graphs are time-varying with a connectivity structure that depends on the choice of functional connectivity measure. Incorrectly representing fMRI data with noisy brain graphs can adversely affect GNN performance. To address this, we propose DynDepNet, a novel method for learning the optimal time-varying dependency structure of fMRI data induced by downstream prediction tasks. Experiments on real-world fMRI datasets, for the task of sex classification, demonstrate that DynDepNet achieves state-of-the-art results, outperforming the best baseline in terms of accuracy by approximately 8 and 6 percentage points, respectively. Furthermore, analysis of the learned dynamic graphs reveals prediction-related brain regions consistent with existing neuroscience literature.

Understanding how speech foundation models relate to human cortical activity is a key challenge for computational neuroscience. Here, we investigate how internal representations from Whisper predict intracranial ECoG responses during naturalistic speech perception. We introduce a time-resolved neural encoder that combines speech embeddings with a recurrent temporal model and soft attention, allowing us to examine layer-wise brain alignment. Intermediate Whisper layers provide the strongest correspondence with neural activity, supporting a hierarchical match between model representations and cortical speech processing. Comparisons with baselines show that high-resolution ECoG responses benefit from temporally structured modelling beyond linear mappings from the same speech representations. In addition, attention maps reveal temporally local alignment between speech embeddings and neural responses, while a phonemic interpretability analysis identifies anatomically coherent phoneme-category organization among encoding-informative electrodes. Together, these results suggest that speech foundation models offer a useful framework for studying time-resolved cortical speech representations.

Objective.Brain decoding is a field of computational neuroscience that aims to infer mental states or internal representations of perceptual inputs from measurable brain activity. This study proposes a novel approach to brain decoding that relies on semantic and contextual similarity.Approach.We use several functional magnetic resonance imaging (fMRI) datasets of natural images as stimuli and create a deep learning decoding pipeline inspired by the bottom-up and top-down processes in human vision. Our pipeline includes a linear brain-to-feature model that maps fMRI activity to semantic visual stimuli features. We assume that the brain projects visual information onto a space that is homeomorphic to the latent space of last layer of a pretrained neural network, which summarizes and highlights similarities and differences between concepts. These features are categorized in the latent space using a nearest-neighbor strategy, and the results are used to retrieve images or condition a generative latent diffusion model to create novel images.Main results.We demonstrate semantic classification and image retrieval on three different fMRI datasets: Generic Object Decoding (vision perception and imagination), BOLD5000, and NSD. In all cases, a simple mapping between fMRI and a deep semantic representation of the visual stimulus resulted in meaningful classification and retrieved or generated images. We assessed quality using quantitative metrics and a human evaluation experiment that reproduces the multiplicity of conscious and unconscious criteria that humans use to evaluate image similarity. Our method achieved correct evaluation in over 80% of the test set.Significance.Our study proposes a novel approach to brain decoding that relies on semantic and contextual similarity. The results demonstrate that measurable neural correlates can be linearly mapped onto the latent space of a neural network to synthesize images that match the original content. These findings have implications for both cognitive neuroscience and artificial intelligence.

Brain decoding is a field of computational neuroscience that uses measurable brain activity to infer mental states or internal representations of perceptual inputs. Therefore, we propose a novel approach to brain decoding that also relies on semantic and contextual similarity. We employ an fMRI dataset of natural image vision and create a deep learning decoding pipeline inspired by the existence of both bottom-up and top-down processes in human vision. We train a linear brain-to-feature model to map fMRI activity features to visual stimuli features, assuming that the brain projects visual information onto a space that is homeomorphic to the latent space represented by the last convolutional layer of a pretrained convolutional neural network, which typically collects a variety of semantic features that summarize and highlight similarities and differences between concepts. These features are then categorized in the latent space using a nearest-neighbor strategy, and the results are used to condition a generative latent diffusion model to create novel images. From fMRI data only, we produce reconstructions of visual stimuli that match the original content very well on a semantic level, surpassing the state of the art in previous literature. We evaluate our work and obtain good results using a quantitative semantic metric (the Wu-Palmer similarity metric over the WordNet lexicon, which had an average value of 0.57) and perform a human evaluation experiment that resulted in correct evaluation, according to the multiplicity of human criteria in evaluating image similarity, in over 80% of the test set.

Bio-inspired networks offer rich dynamic capabilities with minimal energy demands, especially when implemented on neuromorphic hardware. In particular, the recurrence in brain circuits enables Recurrent Spiking Neural Networks (RSNNs) to generate complex spatio-temporal spike patterns, forming internal representations of time-varying signals. Despite this biological sophistication, such architectures are often applied to machine learning tasks with limited biological relevance. In this work, we present a spiking reservoir computing architecture, implemented as a fully spiking Liquid State Machine (LSM) with Leaky Integrate-and-Fire (LIF) neurons, designed to recognize and decode internal brain states. We simulate epileptic activity using a spiking model and develop a complete pipeline where a source generates Local Field Potentials (LFPs). These signals are encoded through population coding and processed by the LSM, which performs regression on the biophysical parameters controlling epileptic dynamics, thereby inferring the source state. To enhance the LSM's performance, we apply a biologically inspired synaptic plasticity mechanism to the RSNN. Our results demonstrate that a simple, unsupervised plasticity mechanism can optimize the internal parameters of the reservoir, particularly in smaller networks. This approach offers a hardware-efficient strategy for task-specific adaptation of general-purpose circuits, highlighting its suitability for edge-device implementations. Our findings emphasize the potential of spiking reservoir computing for real-time decoding of complex brain dynamics, such as epileptic activity, and underscore the advantages of biologically inspired, energy-efficient methods for neuromorphic systems in real-world applications.Clinical relevance- The ability to decode internal brain states from Local Field Potentials (LFPs) using a biologically inspired spiking reservoir computing architecture has significant implications for clinical neuroscience. By providing a real-time, energy-efficient method for tracking epileptic dynamics, this approach could aid in the development of advanced braincomputer interfaces (BCIs) and personalized neurostimulation therapies. Such a system may help clinicians monitor seizure progression, optimize treatment strategies, and improve patient outcomes in clinical epilepsy management.