Neuroinformatics最新文献

The recent advance in technology allows for the photorealistic digitalization of anatomical specimens that can now be presented through various dynamic visualization modalities, enabling a more interactive learning experience. This study explores a comprehensive workflow for reproducibly integrating photorealistic three-dimensional (3D) anatomical scans of the supra- and infratentorial venous system with stereoscopic visualization and virtual reality (VR) for anatomical learning. A formaldehyde-fixed head and neck specimen was injected with radiopaque dye into its vessels, and a post-mortem computed tomography (CT) venography was performed. A layered anatomical dissection of the intracranial venous system was performed. Photogrammetry surface scanning was employed to create 3D anatomical models, which were then post-processed to produce stereoscopic 3D images and videos using open-source software. In addition, the 3D models were formatted for immersive VR environment integration. Six photorealistic surface models and one CT venography-based reconstruction were generated. These were incorporated into several platforms: multiplayer VR environment using stand-alone headsets, and stereoscopic materials suitable for phone-based VR viewers, 3D multimedia projectors, or 3D monitors with passive or active glasses. These formats supported multiple learning scenarios (VR in single or multiplayer sessions), 3D stereoscopic lectures using 3D multimedia, real-time 3D stereoscopic visualization, or prerecorded videos for phone-based VR visualization. Building on these formats, the proposed workflow enables a realistic and spatially accurate representation of the anatomical data with photorealistic 3D models and facilitates the creation of accessible educational content for 3D stereoscopic presentations and immersive dedicated VR sessions, all through a user-friendly technical approach.

Global interpretability in machine learning holds great potential for extracting meaningful insights from neuroimaging data to improve our understanding of brain function. Although various approaches exist to identify key contributing features at both local and global levels, the high dimensionality and correlations in neuroimaging data require careful selection of interpretability methods to achieve reliable global insights into brain function using machine learning. In this study, we evaluate multiple interpretability techniques such as SHAP, which relies on feature independence, as well as recent advances that account for feature dependence in the context of global interpretability, and inherently global methods such as SAGE. To demonstrate the practical application, we trained XGBoost models to predict age and fluid intelligence using neuroimaging measures from the UK Biobank dataset. By applying these interpretability methods, we found that mean intensities in subcortical regions are consistently and significantly associated with brain aging, while the prediction of fluid intelligence is driven by contributions of the hippocampus and the cerebellum, alongside established regions such as the frontal and temporal lobes. These results underscore the value of interpretable machine learning methods in understanding brain function through a data-driven approach.

Accurate detection of brain midline shift is critical for the diagnosis and monitoring of neurological conditions such as traumatic brain injuries, strokes, and tumors. This study aims to address the lack of dedicated datasets and tools for this task by introducing a novel dataset and a 3D Slicer extension, evaluating the effectiveness of multiple deep learning models for automatic detection of brain midline shift. We introduce the brain-midline-detection dataset, specifically designed for identifying three brain landmarks-Anterior Falx (AF), Posterior Falx (PF), and Septum Pellucidum (SP)-in MRI scans. A comprehensive performance evaluation was conducted using deep learning models including YOLOv5 (n, s, m, l), YOLOv8, and YOLOv9 (GELAN-C model). The best-performing model was integrated into the 3D Slicer platform as a custom extension, incorporating steps such as MRI preprocessing, filtering, skull stripping, registration, and midline shift computation. Among the evaluated models, YOLOv5l achieved the highest precision (0.9601) and recall (0.9489), while YOLOv5m delivered the best mAP@0.5:0.95 score (0.6087). YOLOv5n and YOLOv5s exhibited the lowest loss values, indicating high efficiency. Although YOLOv8s achieved a higher mAP@0.5:0.95 score (0.6382), its high loss values reduced its practical effectiveness. YOLOv9-GELAN-C performed the worst, with the highest losses and lowest overall accuracy. YOLOv5m was selected as the optimal model due to its balanced performance and was successfully integrated into 3D Slicer as an extension for automated midline shift detection. By offering a new annotated dataset, a validated detection pipeline, and open-source tools, this study contributes to more accurate, efficient, and accessible AI-assisted medical imaging for brain midline assessment.

Neuroimaging studies have and continue to advance our understanding of the neurobiology of dyslexia. Integration of data from these studies has the potential to replicate findings, deepen understanding through theoretically focused research, and provide for unexpected discovery. This data integration can be important for questions where a sufficiently large and well-defined group of participants is necessary for sufficient experimental power, particularly for a complex disorder where age, language background, and cognitive profiles can impact imaging results. We have developed a data-sharing platform to provide a data repository, image processing resources, and data analysis tools, with an emphasis on data harmonization across retrospective datasets ( https://dyslexiadata.org ). Here, we summarize data sharing, download, imaging metrics, and quality and privacy considerations in the design of and resources available through this repository. By providing access to a relatively large multisite dataset, researchers can test hypotheses about reading development and disability, test novel data analysis methods, even within the platform, and advance understanding of dyslexia.

With recent developments in recording hardware, the processing of neuronal data must keep up with the increasing volumes and complexity by capturing the intrinsic relationships between instances of neuronal activity while remaining invariant to noise. Here, we explore a suite of non-linear manifold feature extraction methods - including PHATE, t-SNE, UMAP, TriMap - in an attempt to identify the most adequate method for automated spike sorting. Spike sorting is the process of clustering instances of neuronal activity, called spikes, based on similarity. By embedding high-dimensional spike shapes into low-dimensional manifolds that preserve local and global structure, we demonstrate more separable and robust clusters than those obtained via traditional feature extraction methods, such as PCA. We evaluated all feature extraction methods analyzed on 95 single-channel synthetic datasets and 2 single-channel real datasets spanning a range of cluster counts. Quantitative evaluation using clustering performance metrics (such as Adjusted Rand Index, Silhouette Score, etc.) indicates that several manifold feature extractions outperform other feature extraction methods. Our results suggest that the embeddings obtained by non-linear manifold approaches can offer a powerful, high-precision option in the spike sorting of the next-generation of electrophysiological recordings. While this study focuses on single-channel data and a subset of manifold learning techniques, a baseline has been established, and future avenues of research have been opened through this work. Future work may extend these insights to multi-channel settings, such as high-density probes and incorporate emerging manifold methods, such as hierarchical and multi-view extensions, which could further improve the robustness and accuracy of spike sorting.

Brain functional connectivity alterations, that is, pathological changes in the signal exchange between areas of the brain, are occurring in several neurological diseases, including neurodegenerative and neuropsychiatric ones. They consist in changes in how brain functional networks work. By conceptualising a brain space as a space whose points are connectome configurations representing brain functional states, changes in brain network functionality can be represented by paths between these points. Paths from a healthy state to a diseased one, or between diseased states as instances of disease progression, are modelled as the action of the Krankheit-Operator, that produces changes from a brain functional state to another one. This study proposes a formal representation of the space of brain states and presents its computational definition. Moreover, references to patients affected by Parkinson's disease, schizophrenia, and Alzheimer-Perusini's disease are included for discussing the proposed approach and possible developments of the research toward a generalisation.

Harmonizing multisite functional magnetic resonance imaging (fMRI) data is crucial for eliminating site-specific variability that hinders the generalizability of machine learning models. Traditional harmonization techniques, such as ComBat, depend on additive and multiplicative factors, and may struggle to capture the non-linear interactions between scanner hardware, acquisition protocols, and signal variations between different imaging sites. In addition, these statistical techniques require data from all the sites during their model training which may have the unintended consequence of data leakage for ML models trained using this harmonized data. The ML models trained using this harmonized data may result in low reliability and reproducibility when tested on unseen data sets, limiting their applicability for general clinical usage. In this study, we propose Autoencoders (AEs) as an alternative for harmonizing multisite fMRI data. Our designed and developed framework leverages the non-linear representation learning capabilities of AEs to reduce site-specific effects while preserving biologically meaningful features. Our evaluation using Autism Brain Imaging Data Exchange I (ABIDE-I) dataset, containing 1,035 subjects collected from 17 centers demonstrates statistically significant improvements in leave-one-site-out (LOSO) cross-validation evaluations. All AE variants (AE, SAE, TAE, and DAE) significantly outperformed the baseline mode (p < 0.01), with mean accuracy improvements ranging from 3.41% to 5.04%. Our findings demonstrate the potential of AEs to harmonize multisite neuroimaging data effectively enabling robust downstream analyses across various neuroscience applications while reducing data-leakage, and preservation of neurobiological features. Our open-source code is made available at https://github.com/pcdslab/Autoencoder-fMRI-Harmonization .

Early diagnosis of mild hepatic encephalopathy is important for the reversion of hepatic encephalopathy. Brain hyper-connectivity networks with hyperedges have showed good performance for diagnosis of neurological disorders. However, the previous hyper-connectivity networks is essentially low-level since the temporal synchronization of regional signal fluctuation is merely considered. Here, we propose a novel high-level hyper-connectivity network based on the resting state functional magnetic resonance imaging to capture the complex interactions among brain regions for better diagnosis of neurological disorders. Resting-state functional magnetic resonance imaging data from 36 mild hepatic encephalopathy patients and 36 cirrhotic patients with no mild hepatic encephalopathy are included in the study. Multi-level high-level hyper-connectivity networks are constructed firstly. Then, we define and extract node hyperdegree, hyperedge global importance and hyperedge dispersion from both low-level and high-level hyper-connectivity networks and combine them. Finally, gradient boosting decision tree is used for feature selection and classification. The leave-one-out cross-validation is used to evaluate the performance. The public ASD resting state functional magnetic resonance imaging datasets from 3 sites are also used as testing set to evaluate the generalization power of our method. Our method showed considerable performance in both experiments which confirms the effectiveness and generalization ability of the model. Besides, important regions and hyperedge features are identified for the interpretability.