Brain Informatics最新文献

Brain-Computer Interfaces provide promising alternatives for detecting stress and enhancing emotional resilience. This study introduces a lightweight, subject-independent method for detecting stress during arithmetic tasks, designed for low computational cost and real-time use. Stress detection is performed through ElectroEncephaloGraphy (EEG) signal analysis using a simplified processing pipeline. The method begins with preprocessing the EEG recordings to eliminate artifacts and focus on relevant frequency bands (α, β, and γ). Features are extracted by calculating band power and its deviation from a baseline. A statistical thresholding mechanism classifies stress and no-stress epochs without the need for subject-specific calibration. The approach was validated on a publicly available dataset of 36 subjects and achieved an average accuracy of 88.89%. The method effectively identifies stress-related brainwave patterns while maintaining efficiency, making it suitable for embedded and wearable devices. Unlike many existing systems, it does not require subject-specific training, enhancing its applicability in real-world environments.

Objective: The reconstruction of visual stimuli and captions from brain activity offers a distinctive viewpoint on how perception reconstructs the external world within neural dynamics. Despite considerable advancements in deep generative models in recent years, simultaneously generating images and captions with both detailed accuracy and semantic consistency remains a significant challenge.

Methods: We introduce panoptic segmentation and generative semantics for the first time, offering enhanced, multi-level data support and a novel perspective in the domain of brain decoding. Using multi-scale fusion techniques, we integrate pixel features from natural images with structural features from panoptic segmentation, creating a state-of-the-art "initial guess." Building upon the neural paradigm that we discovered, we propose an innovative semantic connection strategy to guide image reconstruction. Additionally, by fine-tuning visual semantics within the encoded space compressed by a language model and further combining our advanced retrieval module with the comprehension capabilities of large language models (LLMs), we generate high-quality brain captions.

Results: Experimental results demonstrate that we surpass current methods in visual decoding and brain captioning tasks. We offer a webpage to showcase the results: www.neuai4science.cn:5001/brain_visual_decode .

Conclusion: Our proposed Brain-Imager framework, which incorporates multi-level data and semantic guidance, sets a new standard in the domain.

Significance: This work provides a novel perspective on the relationship between text and image semantics and the visual pathways of the human brain, with potential applications in downstream tasks such as brain-computer interfaces. Additionally, our code is publicly available at https://github.com/songqianyi01/Brain-Imager .

Alzheimer's disease (AD) diagnosis at an early yet accurate stage is critical to support effective treatment or intervention. Still it is not very feasible due to the presence of image data class imbalance, low interpretability of models, and a high computational cost. This research proposes a novel, end-to-end diagnostic framework that considers a Medical Genetic Algorithm (MedGA)-optimized Convolutional Neural Network (CNN) with a Deep Convolutional Generative Adversarial Network (DCGAN) to generate synthetic MRIs and SHapley Additive Explanations (SHAP) to analyse and interpret the model. The given methodology is trained and tested on the Open Access Series of Imaging Studies (OASIS) dataset. The DCGAN component introduces 700 structurally coherent synthetic images (SSIM = 0.92) into the underrepresented Moderate Dementia class, improving the overall recall by 10% and balancing the dataset. MedGA succeeds in optimizing CNN hyperparameters and resulting in complexity reduction (20%) in networks without loss of testing accuracy (97%) at the four demonstrated stages of AD: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. SHAP analysis emphasises the role of key brain areas, the hippocampus and the amygdala in the results of classification accuracy, leading to 25% greater interpretability and clinician confidence. Comparative evaluation shows that the current framework is exceptionally better in terms of predictive performance and explainability than current state-of-the-art approaches. This combined method provides a powerful and adaptable device to categorize AD at an early age, with promising outcomes in precise diagnosis in health facilities.

Background: The Biotuner Toolbox is an open-source Python toolbox for biosignals that integrates concepts from neuroscience, music theory, and signal processing. It introduces a harmonic perspective on physiological oscillations by applying musical constructs such as consonance, rhythm, and scale construction.

Methods: The core biotuner_object processes neural, cardiac, and auditory time series, providing a unified interface for extracting spectral peaks, computing harmonicity metrics, and supporting downstream analyses. Companion modules extend harmonic analyses across temporal (time-resolved harmonicity), spatial (harmonic connectivity), and spectral (harmonic spectrum) dimensions.

Results: Biotuner identifies harmonic structure across different biosignals, revealing significant variations in harmonicity between physiological states. Specifically, the toolbox extracts spectral peaks from complex signals using multiple algorithms, ensuring robust peak detection under varying signal-to-noise ratios. Moreover, we show how harmonicity metrics change across distinct sleep stages and capture variations in the slopes of the aperiodic (1/f) component of the power spectrum.

Conclusion: Biotuner provides an extensible framework that unifies music-theoretic constructs with biosignal processing, enabling hypothesis-driven analyses for researchers and, in parallel, creative exploration of complex natural patterns for artists.

Repetitive transcranial magnetic stimulation (rTMS) is an effective, non-invasive neuromodulation therapy for major depressive disorder (MDD) and treatment-resistant depression (TRD). However, current clinical practice typically relies on standardized protocols that may not adequately account for individual patient variability. To address this gap, we propose a novel, interpretable framework for personalized rTMS treatment recommendations that combines a pretrained sentence embedding model with large language model (LLM)-based reasoning in a retrieval-augmented generation (RAG) setting. Specifically, our approach leverages a pretrained sentence embedding model to encode structured patient profiles into a dense semantic representation, enabling the retrieval of clinically similar cases. These retrieved examples serve as few-shot prompts for in-context learning (ICL), enabling the LLM to reason over these examples and synthesize customized rTMS treatment parameters. Unlike previous approaches that focus solely on individual aspects of personalization, our framework integrates all key parameters (frequency, intensity, and stimulation mode) into a comprehensive recommendation. We systematically evaluated various sentence embedding models and LLMs. Among them, Bge-large-en-v1.5 for few-shot retrieval and GPT-4o-mini for reasoning achieved the highest rTMS protocol matching accuracy of 78.18% using 15 few-shot examples. Our approach is fine-tuning-free, interpretable, and adaptable to real-world, resource-poor clinical settings, providing a promising step forward in data-driven personalized neurostimulation therapy.

Neural signal spikes are recorded by microelectrode technology for specific applications. Nevertheless, spikes frequently encounter significant contamination from several noise sources, making efficient denoising quite challenging. Hence, this study presents a spike-denoising model with a bidirectional long short-term memory and attention mechanism combined with a shallow autoencoder to enhance the signal quality. To assess the effectiveness of the proposed method, various types of synthetic data, such as simulated white noise, correlated noise, colored noise and integrated noise, are used to show the performance. At very high noise levels, the proposed method maintains a high signal-to-noise ratio above 27 dB and the average Pearson 0.91. And the performance metrics of spike detection outperforms the traditional signal processing methods and the partial deep learning methods. Ultimately, the proposed method is used to process the real-world C57 fetal rat neural signals, which can recover a substantial amount of spikes from the obscured noise, showing the proposed method can effectively enhance the quality of neural signals damaged by noise.

Chronic Traumatic Encephalopathy (CTE) and other neurodegenerative disorders (NDs) pose diagnostic challenges due to their diffuse and subtle pathological changes. Traditional diagnostic methods relying on manual histopathological slide inspection are labor-intensive and prone to variability, often missing subtle structural alterations. This study introduces an age-informed, attention-based multiple instance learning pipeline to predict AT8 density, a key marker of p-tau aggregation in CTE. Using Luxol Fast Blue and Hematoxylin & Eosin stained images, our model identifies critical pathological regions and generates interpretable attention maps highlighting structural changes linked to tau pathology. Incorporating patient age enhances predictive accuracy and contextual understanding, addressing aging's confounding effects. We also develop quantitative evaluation procedures for foundation models (FMs), assessing attention map smoothness, faithfulness, and robustness to perturbations like stain variability and noise. These benchmarks facilitate informed FM selection and optimization for neuropathological tasks. By enabling scalable, automated whole-slide image analysis, our approach advances digital neuropathology, supporting earlier and more precise ND diagnoses and uncovering subtle markers with potential applications in clinical imaging.

Understanding how demographic factors influence visual attention is crucial for the development of adaptive and user-centered web interfaces. This paper presents a gender-aware saliency prediction system based on fine-tuned deep learning models and demographic-specific gaze behavior. We introduce the WIC640 dataset, which includes 640 web page screenshots categorized by content type and country of origin, along with eye-tracking data from 85 participants across four age groups and both genders. To investigate gender-related differences in visual saliency, we fine-tuned TranSalNet, a Transformer-based saliency prediction model, on the WIC640 dataset. Our experiments reveal distinct gaze behavior patterns between male and female users. The female-trained model achieved a correlation coefficient (CC) of 0.7786, normalized scanpath saliency (NSS) of 2.4224, and Kullback-Leibler divergence (KLD) of 0.5447; the male-trained model showed slightly lower performance (CC = 0.7582, NSS = 2.3508, KLD = 0.5986). Interestingly, the general model trained on the complete dataset outperformed both gender-specific models, highlighting the importance of inclusive training data. Statistical analysis revealed significant gender-related differences in 9 out of 12 saliency features and a trend of reduced fixation dispersion with increasing age. While this study does not yet incorporate temporal gaze modeling, the results suggest practical benefits for intelligent systems aiming to personalize user experiences based on demographic features. The WIC640 dataset is publicly available and offers a valuable resource for future research on adaptive AI systems, visual attention modeling, and demographic-aware interface design.

Background: Alzheimer's disease (AD) and type 2 diabetes mellitus (T2DM) are distinct yet interconnected disorders that frequently co-occur. While insulin resistance and impaired glucose metabolism have been implicated in their shared pathogenesis, the molecular mechanisms underlying this comorbidity remain incompletely understood. Emerging evidence suggests that circular RNAs (circRNAs), particularly those enriched in neural and metabolic tissues, may play regulatory roles in both diseases.

Methods: We conducted integrated transcriptomic analyses using Gene Expression Omnibus (GEO) datasets to identify differentially expressed genes in AD and T2DM. Protein-protein interaction (PPI) network construction and enrichment analyses identified common hub genes and dysregulated pathways. Functional studies were performed in SH-SY5Y and HEK293 cell models to explore the biological impact of Circular RNA Cwc27 (circCwc27), a circRNA derived from the Cwc27 gene.

Results: Among 86 commonly upregulated genes, Cwc27 emerged as a central hub with significant connectivity in the AD-T2DM interaction network. Functional enrichment analysis revealed circCwc27's association with RNA splicing, mRNA surveillance, and PI3K-Akt signaling. Overexpression of circCwc27 increased total and phosphorylated Tau protein levels, enhanced Tau seeding activity, and reduced intracellular glycogen storage-hallmarks of AD neuropathology and metabolic dysregulation in T2DM. Notably, these effects occurred independently of Akt-GSK3β activation or APP expression, suggesting a unique regulatory axis involving Tau protein.

Conclusion: Our findings identify circCwc27 as a novel molecular bridge linking AD and T2DM via Tau upregulation and metabolic impairment. This dual role highlights its potential as both a biomarker and therapeutic target for addressing the shared pathophysiological mechanisms of neurodegeneration and metabolic disease.