NI Biofizikos ir bioinformatikos laboratorija (11.03)

Collagenous colitis (CC) is diagnosed histologically and is characterised by a thickened subepithelial collagen band together with inflammatory and epithelial changes. Although routine haematoxylin and eosin (H&E) staining is sufficient for diagnosis in most cases, visual assessment of the collagen band can be challenging in borderline or heterogeneous specimens. Additional stains may be required in diagnostically difficult situations.

Background There is no universally accepted definition of perioperative blood loss in cardiac surgery. Existing methods are based on chest tube output and are not normalised for patient weight. Objective To validate machine learning-derived blood loss severity clusters based on a haemoglobin mass loss per kilogram (Hb/kg index). Methods This single-center prospective study included 195 patients undergoing cardiac surgery between October 2023 and November 2024. Three clusters derived using K-Medoids were mapped to the Hb/kg index to define cut-offs. Cluster discrimination was assessed by receiver operating characteristics (ROC) analysis (area under the curve (AUC)). Group comparisons were performed using analysis of covariance adjusted for age and gender. Associations between the Hb/kg index and clinical outcomes, including transfusion requirements and complications were analysed using Chi-square tests and adjusted two-way Analysis of Covariance (ANCOVA). Results Clustering identified three groups (Mild, Moderate, Severe) defined by optimal Hb/kg thresholds of 1.72 and 2.10. The Severe cluster demonstrated strong discriminative performance (AUC = 0.790, 95% confidence interval 0.721-0.859). Chest tube output did not differ significantly between clusters (= 0.097), while haemoglobin mass loss through chest tubes demonstrated a significant effect (p = 0.011). Conclusions The Hb/kg Index is a validated, data-driven, objective metric for perioperative blood loss, offering greater precision than traditional chest tube drainage volume. It effectively stratifies bleeding severity and identifies high-risk patients with lower BMI.

Image-guided radiotherapy (IGRT) has enhanced the precision of cancer treatment by integrating imaging modalities such as computed tomography (CT), magnetic resonance imaging (MRI) and cone-beam computed tomography (CBCT) into daily radiotherapy workflows. In head and neck cancer, where anatomical changes are common, accurate image registration between planning and treatment scans is essential to ensure dose accuracy. However, geometric distortions in CBCT (such as translation, rotation, and scaling resulting from patient positioning variations observed in daily CBCT images) can affect tumour targeting and dose delivery. This pilot study assesses a MATLAB-based image correction algorithm that uses rigid bony landmarks and point cloud registration together with spatial transformation to align CBCT with planning CT. Two head and neck cancer patients were retrospectively analysed, selected for their contrasting anatomical responses: one with substantial tumour regression and one with minimal change. Imaging was performed on the Halcyon V3.1 linear accelerator (Varian Medical Systems), with 25 daily CBCT scans per patient (85–96 slices per scan), resulting in 50 datasets for analysis. Spatial deviations were measured along the X, Y, and Z axes, and dose recalculations were performed for each treatment fraction. The correction method significantly improved spatial congruence and reduced geometric discrepancies caused by voxel spacing and acquisition parameters. Uncorrected scans showed dose deviations of up to ± 12% in organs at risk, notably the spinal cord and parotid glands. These findings demonstrate the feasibility and dosimetric relevance of automated CBCT correction in daily head and neck radiotherapy. Although limited in sample size, the study provides a detailed technical and dosimetric analysis of spatial distortions and supports future validation in larger patient cohorts.

Postoperative bleeding following cardiopulmonary bypass (CPB) remains a significant challenge.Although viscoelastic testing is increasingly used, the relative contributions of fibrinogen, platelet count and clot firm-ness to blood loss remain debated. We evaluated the diagnostic accuracy of thromboelastometry (ROTEM) comparedwith platelet aggregometry (PA) and standard tests, using the Hb/kg index to quantify blood loss.

Introduction Parkinson’s disease (PD) is a neurodegenerative disorder affecting motor and cognitive function. While dopaminergic medications provide symptom relief, deep brain stimulation (DBS) is often used in advanced cases to manage motor dysfunction. However, the response to DBS varies between individuals, and optimal stimulation settings are typically determined through clinical trial and error. This highlights the need for biomarkers to monitor DBS-induced changes in brain function. Methods In this pilot study, we explored EEG-based connectivity features as potential biomarkers for DBS effects in two PD patients. EEG was recorded before and after DBS implantation. Preoperative EEGs were collected while patients were off medication; postoperative EEGs were obtained six months later with DBS turned on. Preprocessing followed Makoto’s pipeline in EEGLAB, and source reconstruction was conducted in Brainstorm using MRI-based Boundary Element Method head models generated via Open-MEEG. EEG signals were projected onto the Desikan-Killiany atlas and segmented into 3-second epochs. We extracted three connectivity metrics from source- space data: Amplitude Envelope Correlation (AEC), Phase Locking Value (PLV), and Avalanche Transition Matrix (ATM). A Wilcoxon permutation test (1000 iterations) with max-statistic correction was used to assess changes between conditions. Results ATM showed no significant differences. In contrast, AEC and PLV both revealed reductions in beta- band connectivity after DBS activation (Fig 1). Subject specific patterns were observed: subject 1 showed stronger AEC changes, while subject 2 exhibited more pronounced PLV effects. These results suggest that beta-band connectivity is sensitive to neuromodulation and that EEG-based metrics can capture individualized brain dynamics. Conclusions This pilot study supports the use of source-space EEG connectivity analysis for tracking DBS- elated changes and lays the groundwork for future integration into modeling frameworks such as The Virtual Brain (TVB).

Sensory processing sensitivity (SPS) is a personality trait characterized by heightened responsiveness to environmental stimuli, often linked to introversion and emotional reactivity. This study evaluated the factorial structure of the Lithuanian version of the Highly Sensitive Person Scale (HSPS) and its associations with anxiety, depression, and personality traits. A cross-sectional survey of 1130 Lithuanian employees was conducted. Exploratory and confirmatory factor analyses supported a three-factor model: Low Sensory Threshold and Withdrawal, Ease of Excitation, and Aesthetic Sensitivity. HSPS scores were negatively correlated with Extraversion and Emotional Stability and positively associated with anxiety and depression symptoms. The Lithuanian HSPS is a reliable and valid instrument for assessing SPS in adults.

Introduction Parkinson's disease, the second most common neurodegenerative disorder, is marked by progressive loss of dopaminergic neurons. Dopaminergic disruption causes severe motor symptoms (resting tremor, rigidity, bradykinesia, postural instability), cognitive deficits, and reduced quality of life in aging. While incurable, treatments such as deep brain stimulation can alleviate symptoms. We aim to develop personalized virtual brain models for Parkinson's patients that explicitly incorporate dopaminergic modulation. Building on our earlier work, we now refine these models by integrating individual structural connectivity, advanced neural mass formulations, and EEG-derived biomarkers, enabling more precise characterization of patient-specific dynamics. Methods We introduced a modular framework to capture dopaminergic regulation at the neural mass scale (Depannemaecker, 2024; Casagrande, 2025). Targeting D1-type receptor dynamics, it enables investigation of how fluctuations in dopamine availability shape population responses. The framework adopts a mean-field formulation (Chen & Campbell, 2022), providing a tractable yet biologically plausible description of macroscopic activity, well-suited to study dopamine-mediated modulation of basal ganglia circuits in health and disease.To identify biomarkers of brain dynamics in Parkinson’s disease (PD), we analyzed resting-state EEG acquired before and after DBS electrode implantation. Preoperatively, patients were tested ON and OFF L-DOPA; postoperatively, six months later, they were assessed with DBS ON and OFF, under both medication conditions. EEG features—Amplitude Envelope Correlation (AEC) and Phase Locking Value (PLV)—were extracted, and significant effects identified with permutation ANOVA. Results Our framework reveals that dopaminergic modulation critically reshapes the dynamical repertoire of neural mass models. By systematically varying dopamine input levels, we identified transitions between qualitatively distinct regimes, including fixed-point convergence, sustained oscillatory activity (limit cycles), and bursting dynamics. Progressive increases in dopaminergic tone induced a marked expansion of quiescent states, accompanied by reduced oscillatory frequencies across both fast spiking and inter-burst domains. Importantly, these model-derived regimes correspond to electrophysiological features observed in EEG and DBS recordings from Parkinsonian patients, thereby providing a mechanistic account of dopamine-dependent neural dynamics. Analysis of EEG data showed that the most significant changes were found in the reduction in the Beta band for both PLV and AEC features, as reported in previous studies. Discussion Our results show that dopaminergic tone critically shapes neural mass dynamics, inducing transitions between quiescent, oscillatory, and bursting regimes. The observed reduction in oscillatory frequency and expansion of quiescent states parallel electrophysiological signatures in Parkinsonian EEG and DBS data, particularly in the Beta band. Next, we will integrate model-derived features with empirical connectivity into a unified framework, extending earlier approaches (Angiolelli, 2025) by incorporating DBS effects. This strategy advances toward personalized dynamical models to guide understanding and optimization of neuromodulation therapies.

INTRODUCTION The EBRAINS Research Infrastructure [1,2] offers an ecosystem of data, tools, and resources for neuroscience research. EBRAINS accommodates heterogeneous, multilevel data that are carefully curated and made findable and interoperable through structured metadata and connected to digital brain atlases and computational resources for exploration and analysis. EBRAINS was delivered by the Human Brain Project (2013-2023) [3], with several follow-up projects contributing to operating, maintaining and developing the core infrastructure and distributed services. Providing novel solutions for sharing, integration and analysis of data [4], EBRAINS has opened new avenues for the field of neuroscience. In 2021 EBRAINS was added to the ESFRI (European Strategy Forum on Research Infrastructures) roadmap, initiating efforts to configure EBRAINS for the future. To establish a sustainable future for EBRAINS, a scientific, technical, financial, legal and governance framework is now being developed. METHODS The hub-node structure of EBRAINS was established in 2019, with seven founding member institutions forming the EBRAINS AISBL as a legal entity. Services are provided through a distributed network of National Nodes offering scientific expertise, software, services, and support. The EBRAINS PREP project (2022-2025) defined an organizational framework with statutes, business plan, technical design, and formal agreements, connecting EBRAINS nodes into a consortium dedicated to providing services and contributing to the development of EBRAINS. The EBRAINS 2.0 project is improving, automating, and monitoring services, as well as onboarding new services. RESULTS AND DISCUSSION Through the EBRAINS PREP project, the National Nodes developed governance and financial models formalizing the nodes internally with mandates for signing national node agreements and contributing membership fees. The National Node Agreements, signed in 2025, regulate the collaborative work, to facilitate the operation and development of EBRAINS and expand the panEuropean EBRAINS user base to increase scientific exchange and advancement. The EBRAINS PREP project delivered updated formal policy documents regulating activities, as well as updated catalogues of services, tools, and training resources. To ensure seamless operation services, new service onboarding and monitoring routines were developed, defining core services and distributed operational services, together with agreements regulating service levels and commitment to support. The National Node Board serves as a liaison between the EBRAINS and distributed user communities, coordinating the National Nodes to facilitate service onboarding and provisioning, collaborative training efforts and outreach to recruit new users and expand the national node network to new member states. Finally, following the inclusion of EBRAINS in the ESFRI 2021 roadmap, the choice was made to develop EBRAINS towards becoming a European Digital Infrastructure Consortium (EDIC), as a multi-country collaboration delivering an extensive portfolio of digital research tools and services for neuroscience. The EBRAINS hub-node network, with a defined governance structure underpinned by formal agreements, policies, and procedures, provides a foundation for the further development of a financial, legal and governance framework suitable for implementing EBRAINS as an EDIC. To achieve this ambition and secure a longer-term sustainable future for EBRAINS, it is necessary to describe how EBRAINS can operate as an EDIC, with a suitable governance structure and a suite of service offerings needed by the brain research community. The National Nodes are central to this endeavour, connecting national user communities with EBRAINS services, maintaining interoperability with other research infrastructures, and contributing to community engagement and matching services to needs and trends across the European neuroscience landscape.

Transformer networks (Vaswani et al., 2017) have revolutionized artificial intelligence with their powerful capabilities, yet it remains unknown if they process information in a way that is similar to the human brain. Recent theories suggest that the transformer’s attention mechanism may mimic the brain’s cortico-thalamic circuits (Granier & Senn, 2025). However, a working computational model is needed to test this idea against realworld brain data. This study aimed to build and train a cortical transformer model, which is a transformer based on the proposed cortico-thalamic architecture, which maps the mathematical components of the self-attention mechanism—queries, keys, and values—directly onto the biological dynamics of specific cortical layers and thalamic loops (Granier & Senn, 2025). The goal was to create a tool that can generate internal activity patterns for direct comparison with human brain recordings during a language task. A cortical transformer was built in PyTorch, using a 4-layer, 6-head design. The model, containing 11.71 million parameters, was trained for 34 epochs on the Multi30k dataset (Elliott et al., 2016) to complete an English-to-German translation task. A data pipeline was developed to extract the model’s internal activations (Query, Key, and Value vectors) in response to linguistic input. The cortical transformer was successfully trained, achieving a BLEU score of 25.29. While not state-of-the-art, this result serves as a critical proof of concept, demonstrating that the cortico-thalamic architecture is computationally viable and capable of learning a complex language task. The system is now prepared for future neuroscientific studies that will correlate its activations with human brain recordings, providing a framework to investigate the biological basis of attention and cognition.

Ageing changes are heterogenous, with substantial variation in health impacts of ageing across populations, individuals and tissues [Grimbly MJ, et al. BMJ Open Ophth 2024;9:e001794. doi:10.1136/bmjophth-2024-001794]. Biological ageing markers have emerged to better represent the ageing process and predict the risk of diseases. Retinal age is an imaging-based biomarker for biological age assessment from retinal fundus photographs. Accelerated ageing, or Retinal age gap (RAG) – the difference between calculated retinal age and chronological age, provides a valuable metric for quantifying the risk of vascular, neurodegenerative, metabolic diseases and even assessing the risk of mortality. However, an undisclosed controversy remains whether generated retinal age models are accurate predictors of biological age. Moreover, further clinical trials exploring their applicability across diverse healthy populations are warrant. We aimed to determine the accuracy of retinal age prediction models and evaluate their ability to reflect age-related parameters from retinal images exploring their estimation to chronological age of healthy Lithuanian population subjects. Color fundus images from 92 patients were analyzed for age prediction, allocating 80% of patients (n = 74) for training and 20% (n = 18) for validation. Images were split into training and validation sets at the patient level to avoid data leakage. Each patient contributed 8 images, except for one patient who had 6 images, resulting in a total of 734 fundus images included in the analysis. Preliminary experiments using grayscale OCTA images resulted in insufficient prediction performance, therefore the analysis focused exclusively on color fundus data. A convolutional neural network based on EfficientNet-B2 was adapted for regression by replacing the classification head with a single linear output unit predicting normalized age values. EfficientNet-B2 was selected based on findings from our previous study, in which several convolutional and transformer-based architectures were compared for classifying fundus images of healthy versus pathological images. The evaluated models in-