Myelin plays a critical role in the central nervous system, and its maturation is essential for understanding brain development. However, assessing myelin progression remains challenging due to variability across age groups. Radiologists typically rely on developmental atlases and age-based milestones, but manual evaluation is time-consuming and prone to inter-observer variability. This paper presents a novel dual-input deep learning framework that leverages both [Formula: see text] and [Formula: see text]-weighted MRI modalities for automated myelin maturation assessment. Each modality is processed through an in-domain trained DenseNet121 feature extractor, followed by Channel and Multi-Head Attention Blocks to enhance feature prioritization and spatial contextualization. Cross-Attention enables effective inter-modality information exchange, while early fusion via concatenation integrates structural insights from both contrasts. The fused features are refined using Global Average Pooling and passed to a regression-optimized dense layer. Trained on 710 samples and tested on 123 from a publicly available dataset (833 total), the model achieved a Mean Absolute Error (MAE) of 1.18 months, a Pearson Correlation Coefficient (PCC) of 0.98, a Coefficient of Determination ([Formula: see text]) of 0.96, and a Concordance Correlation Coefficient (CCC) of 0.98. Visual interpretability through Grad-CAM revealed the model's focus on clinically meaningful brain regions, with abnormal cases showing heightened activation in peripheral and ventral areas. These findings confirm the model's ability to deliver accurate and interpretable predictions, supporting its potential for real-world diagnostic integration in pediatric neuroimaging.

A significant challenge for neurofeedback training research and related clinical applications, is participants' difficulty in learning to induce specific brain patterns during training. Here, we address this issue in the context of fMRI-based decoded neurofeedback (DecNef). Arguably, discrepancies between the data used to construct the decoder and the data used for neurofeedback training, such as differences in data distributions and experimental contexts, neural and non-neural noise, are likely the cause of the difficulties of the aforementioned participants. Here, we developed a co-adaptation procedure using standard machine learning algorithms. The procedure involves an adaptive decoder algorithm that is updated in real time based on its predictions across neurofeedback trials. First, we tested the procedure via simulations using a previous DecNef dataset and showed that decoder co-adaptation can improve performance during neurofeedback training. Importantly, a drift analysis demonstrated the stability of the co-adapted decoder throughout the neurofeedback training sessions. We then collected real time fMRI data in a DecNef training procedure to provide proof of concept evidence that co-adaptation enhances participant's ability to induce the target state during training. Thus, personalized decoders through co-adaptation can improve the precision and reliability of DecNef training protocols to target specific brain representations, with ramifications in translational research. The tools are made openly available to the scientific community.

Muscle weakness after immobilization often exceeds that explained by loss of muscle mass alone, suggesting a role for neuromuscular synaptic changes. To quantify these adaptations, we developed a composite transcriptomic Neuromuscular Junction (NMJ) Remodeling Score and evaluated its behavior relative to classical atrophy pathways during short-term unloading. We analyzed vastus lateralis RNA sequencing data from adults undergoing 10 days of unilateral lower-limb suspension followed by a 21-day recovery, generating NMJ and atrophy scores for 15 and 10 genes, respectively. Transcriptome-wide testing across more than twenty thousand genes identified a broad pattern of metabolic suppression. The NMJ score showed a large effect increase during unloading and partial normalization with recovery, while the atrophy score rose more strongly and reversed during recovery. The two scores demonstrated weak correlation, consistent with distinct biological processes. Individual NMJ-related genes displayed coordinated regulation, including marked upregulation of several acetylcholine receptor subunits and modest downregulation of muscle signaling kinase (MuSK), reflecting a denervation-like transcriptional pattern. Directional replication in a 60-day bed rest cohort supported generalizability across disuse conditions. Together, these findings indicate that limb unloading elicits measurable transcriptomic remodeling at the NMJ that is only partially aligned with atrophy signaling, providing a framework for investigating neural contributions to immobilization-induced weakness.

Tauopathies are characterised by a progressive accumulation of hyperphosphorylated tau. However, early and intermediate stages remain challenging to quantify due to subtle and heterogeneous morphological characteristics. This study evaluates a deep learning framework for classifying multiple temporal stages of tauopathy progression using AT8 (anti-phospho-tau antibody)-stained cortical micrographs in a controlled traumatic brain injury mouse model - an underexplored application. Three convolutional neural network (CNN) architectures were examined: a custom CNN and two transfer-learning models (InceptionV3 and DenseNet). Images were grouped into four post-injury stages: 1day, 1week, 1month and 3months. Preprocessing included normalisation, augmentation and oversampling to address imbalance. Performance was assessed using stratified k-fold cross-validation with accuracy, macro-F1, per-class F1, and one-vs-rest area under the receiver operating characteristic curve (AUC). DenseNet achieved the best overall performance (accuracy = 70.9%, macro-F1 = 0.68) with strong discrimination for the 1-week stage (F1 = 0.95). All models showed limited separability in the earliest post-injury stage (1day), while intermediate to late stages (1-3months) exhibited partial overlap, consistent with the progressive nature of tau accumulation. These results indicate that deep learning, particularly transfer learning, offers a scalable approach for automated temporal staging of tauopathy in preclinical histology. Although the results are based on internal cross-validation without independent animal-level identifiers or external cohorts, the proposed framework provides a reliable foundation for incorporating CNN-based analysis into digital neuropathology workflows. Larger multi-centre datasets and slide-level modelling will be required to assess generalisation and support applications in early detection, longitudinal tracking, and treatment evaluation of tau-related neurodegeneration.

The study of structural brain networks (SBNs) offers critical insights into brain-cognition relationships. However, a comprehensive comparison of these methods in terms of their topological properties, cognitive relevance, and sensitivity to connection density remains lacking. This study compares two types of individual-level SBNs-morphometric similarity networks (MSNs) and morphometric inverse divergence (MIND) networks-by analyzing their associations with cognitive performance using sMRI data from 29 male children. Group- and individual-level analyses were conducted to evaluate differences in hemispheric connectivity, topological features, and their correlations with cognitive performance across different connection densities. In our analyses, a connection density of [Formula: see text] appeared optimal for stabilizing network properties and maximizing cognitive correlations in both MSN and MIND. Moreover, advanced network segregation and integration metrics (such as local efficiency and node versatility, along with their global summaries) demonstrated greater sensitivity to cognitive performance. However, MSNs appeared to provide a more reliable framework, demonstrating more stable associations across connection densities in topological and hemispheric dimensions. Specifically, higher cognitive performance may be linked to stronger left intra-hemispheric connectivity, weaker inter-hemispheric connectivity, and more modular network organization-consistent with established theories of hemispheric specialization and efficient modularity. In contrast, MIND networks exhibit reduced effectiveness and stability across metrics and densities in our data. These preliminary insights enhance our understanding of brain-cognition relationships and provide practical guidelines for parameter selection and metric identification in network-based cognitive analyses.

This study addresses the important question of how neuron model choice and learning rules shape the classification performance of Spiking Neural Networks (SNNs) in bio-signal processing. By systematically contrasting Leaky Integrate-and-Fire, metaneurons, and probabilistic Levy-Baxter (LB) neurons across spike-timing dependent plasticity, tempotron, and reward-modulated learning, we identify model-rule combinations best suited for capturing the temporal richness of neural data. A novel contribution is the integration of a complexity-driven evaluation into the SNN pipeline. Using Lempel-Ziv Complexity (LZC), an entropy-related measure of spike-train regularity, we provide a consistent and interpretable benchmark of classification outcomes across architectures. To probe neural dynamics under controlled conditions, we employed synthetic datasets with varying temporal dependencies and stochasticity, including Markov and Poisson processes established models of neuronal spike-trains. Moreover, we validated the observed trends on real data by testing the same architectures on an MNIST dataset. Performance trends reveal strong dependence on the interaction between neuron model, learning rule, and network size. The LZC based evaluation highlights configurations resilient to weak or noisy signals. The LB-tempotron combination proved most effective for tasks with complex temporal patterns, leveraging adaptive neuronal dynamics and precise spike-timing exploitation. LIF-based architectures with Bio-inspired Active Learning delivered solid accuracy at lower computational cost, while hybrid models offered a versatile middle ground when paired with appropriate learning algorithms. This work delivers the first systematic mapping of neuron model learning rule synergies in SNNs and introduces complexity-based evaluation framework that sets a robust benchmark for biosignal classification. Beyond benchmarking, our results provide actionable guidelines for building next-generation SNNs capable of handling the variability and complexity of real neural data.