Frontiers in Neuroinformatics最新文献

Anomaly detection is a typical binary classification problem under the condition of unbalanced samples, which has been widely used in various fields of data mining. For example, it can help detect heart murmurs when the heart is structurally abnormal, to tell if a newborn has congenital heart disease. Due to the low time and high efficiency, most work focuses on the semi- supervised anomaly detection method. However, the anomaly detection effect of this method is not high because of massive data with uneven samples and different noise. To improve the accuracy of anomaly detection under unbalanced sample conditions, we propose a new semi-supervised anomaly detection method (WCOS) based on semi-supervised clustering, which combines wavelet reconstruction, convolutional autoencoder, and one classification support vector machine. In this way, we can not only distinguish a small proportion of abnormal heart sounds in the huge data scale but also filter the noise through the noise reduction network, thus significantly improving the detection accuracy. In addition, we evaluated our method using real datasets. When the noise of sigma = 0.5, the AUC standard deviation of the WR-CAE-OCSVM is 19.2, 54.1, and 29.8% lower than that of WR-OCSVM, CAE-OCSVM and OCSVM, respectively. The results confirmed the higher accuracy of anomaly detection in WCOS compared to other state-of-the-art methods.

This article develops a fundamental insight into the behavior of neuronal membranes, focusing on their responses to stimuli measured with power spectra in the frequency domain. It explores the use of linear and nonlinear (quadratic sinusoidal analysis) approaches to characterize neuronal function. It further delves into the random theory of internal noise of biological neurons and the use of stochastic Markov models to investigate these fluctuations. The text also discusses the origin of conductance noise and compares different power spectra for interpreting this noise. Importantly, it introduces a novel sequential chemical state model, named p ₂, which is more general than the Hodgkin-Huxley formulation, so that the probability for an ion channel to be open does not imply exponentiation. In particular, it is demonstrated that the p ₂ (without exponentiation) and n ⁴ (with exponentiation) models can produce similar neuronal responses. A striking relationship is also shown between fluctuation and quadratic power spectra, suggesting that voltage-dependent random mechanisms can have a significant impact on deterministic nonlinear responses, themselves known to have a crucial role in the generation of action potentials in biological neural networks.

Purpose: The Multicentre Acute ischemic stroke imaGIng and Clinical data (MAGIC) repository is a collaboration established in 2024 by seven stroke centres in Europe. MAGIC consolidates clinical and radiological data from acute ischemic stroke (AIS) patients who underwent endovascular therapy, intravenous thrombolysis, a combination of both, or conservative management.

Participants: All centres ensure accuracy and completeness of the data. Only patients who did not refuse use of their routine data collected during or after their hospital stay are included in the repository. Approvals or waivers are obtained from the responsible ethics committees before data exchange. A formal data transfer agreement (DTA) is signed by all contributing centres. The centres then share their data, and files are stored centrally on a safe server at the University Hospital Zurich. There, patient identifiers are removed and images are algorithmically de-faced. De-identified structured clinical data are connected to the imaging data by a new identifier. Data are made available to participating centres which have entered into a DTA for stroke research projects.

Repository setup: Initially, MAGIC is set to comprise initial and first follow-up imaging of 2,500 AIS patients. Clinical data consist of a comprehensive set of patient characteristics and routine prehospital metrics, treatment and laboratory variables.

Outlook: Our repository will support research by leveraging the entire range of routinely collected imaging and clinical data. This dataset reflects the current state of practice in stroke patient evaluation and management and will enable researchers to retrospectively study clinically relevant questions outside the scope of randomized controlled clinical trials. New centres are invited to join MAGIC if they meet the requirements outlined here. We aim to reach approximately 10,000 cases by 2026.

Introduction: The prevalence of age-related brain issues has risen in developed countries because of changes in lifestyle. Alzheimer's disease leads to a rapid and irreversible decline in cognitive abilities by damaging memory cells.

Methods: A ResNet-18-based system is proposed, integrating Depth Convolution with a Squeeze and Excitation (SE) block to minimize tuning parameters. This design is based on analyses of existing deep learning architectures and feature extraction techniques. Additionally, pre-trained ResNet-18 models were created with and without the SE block to compare ROC and accuracy values across different hyperparameters.

Results: The proposed model achieved ROC values of 95% for Alzheimer's Disease (AD), 95% for Cognitively Normal (CN), and 93% for Mild Cognitive Impairment (MCI), with a maximum test accuracy of 88.51%. However, the pre-trained model with SE had 93.26% accuracy and ROC values of 98%, 99%, and 98%, while the model without SE had 94%, 97%, and 94% ROC values and 92.41% accuracy.

Discussion: Collecting medical data can be expensive and raises ethical concerns. Small data sets are also prone to local minima issues in the cost function. A scratch model that experiences extensive hyperparameter tuning may end up being either overfitted or underfitted. Class imbalance also reduces performance. Transfer learning is most effective with small, imbalanced datasets, and pre-trained models with SE blocks perform better than others. The proposed model introduced a method to reduce training parameters and prevent overfitting from imbalanced medical data. Overall performance findings show that the suggested approach performs better than the state-of-the-art techniques.

Introduction: Mental health monitoring utilizing EEG analysis has garnered notable interest due to the non-invasive characteristics and rich temporal information encoded in EEG signals, which are indicative of cognitive and emotional conditions. Conventional methods for EEG-based mental health evaluation often depend on manually crafted features or basic machine learning approaches, like support vector classifiers or superficial neural networks. Despite the potential of these approaches, they often fall short in capturing the intricate spatiotemporal relationships within EEG data, leading to lower classification accuracy and poor adaptability across various populations and mental health scenarios.

Methods: To overcome these limitations, we introduce the EEG Mind-Transformer, an innovative deep learning architecture composed of a Dynamic Temporal Graph Attention Mechanism (DT-GAM), a Hierarchical Graph Representation and Analysis (HGRA) module, and a Spatial-Temporal Fusion Module (STFM). The DT-GAM is designed to dynamically extract temporal dependencies within EEG data, while the HGRA models the brain's hierarchical structure to capture both localized and global interactions among different brain regions. The STFM synthesizes spatial and temporal elements, generating a comprehensive representation of EEG signals.

Results and discussion: Our empirical results confirm that the EEG Mind-Transformer significantly surpasses conventional approaches, achieving an accuracy of 92.5%, a recall of 91.3%, an F1-score of 90.8%, and an AUC of 94.2% across several datasets. These findings underline the model's robustness and its generalizability to diverse mental health conditions. Moreover, the EEG Mind-Transformer not only pushes the boundaries of state-of-the-art EEG-based mental health monitoring but also offers meaningful insights into the underlying brain functions associated with mental disorders, solidifying its value for both research and clinical settings.

Understanding brain function relies on identifying spatiotemporal patterns in brain activity. In recent years, machine learning methods have been widely used to detect connections between regions of interest (ROIs) involved in cognitive functions, as measured by the fMRI technique. However, it's essential to match the type of learning method to the problem type, and extracting the information about the most important ROI connections might be challenging. In this contribution, we used machine learning techniques to classify tasks in a working memory experiment and identify the brain areas involved in processing information. We employed classical discriminators and neural networks (convolutional and residual) to differentiate between brain responses to distinct types of visual stimuli (visuospatial and verbal) and different phases of the experiment (information encoding and retrieval). The best performance was achieved by the LGBM classifier with 1-time point input data during memory retrieval and a convolutional neural network during the encoding phase. Additionally, we developed an algorithm that took into account feature correlations to estimate the most important brain regions for the model's accuracy. Our findings suggest that from the perspective of considered models, brain signals related to the resting state have a similar degree of complexity to those related to the encoding phase, which does not improve the model's accuracy. However, during the retrieval phase, the signals were easily distinguished from the resting state, indicating their different structure. The study identified brain regions that are crucial for processing information in working memory, as well as the differences in the dynamics of encoding and retrieval processes. Furthermore, our findings indicate spatiotemporal distinctions related to these processes. The analysis confirmed the importance of the basal ganglia in processing information during the retrieval phase. The presented results reveal the benefits of applying machine learning algorithms to investigate working memory dynamics.

Introduction: Modeling multi-channel electroencephalographic (EEG) time-series is a challenging tasks, even for the most recent deep learning approaches. Particularly, in this work, we targeted our efforts to the high-fidelity reconstruction of this type of data, as this is of key relevance for several applications such as classification, anomaly detection, automatic labeling, and brain-computer interfaces.

Methods: We analyzed the most recent works finding that high-fidelity reconstruction is seriously challenged by the complex dynamics of the EEG signals and the large inter-subject variability. So far, previous works provided good results in either high-fidelity reconstruction of single-channel signals, or poor-quality reconstruction of multi-channel datasets. Therefore, in this paper, we present a novel deep learning model, called hvEEGNet, designed as a hierarchical variational autoencoder and trained with a new loss function. We tested it on the benchmark Dataset 2a (including 22-channel EEG data from 9 subjects).

Results: We show that it is able to reconstruct all EEG channels with high-fidelity, fastly (in a few tens of epochs), and with high consistency across different subjects. We also investigated the relationship between reconstruction fidelity and the training duration and, using hvEEGNet as an anomaly detector, we spotted some data in the benchmark dataset that are corrupted and never highlighted before.

Discussion: Thus, hvEEGNet could be very useful in several applications where automatic labeling of large EEG dataset is needed and time-consuming. At the same time, this work opens new fundamental research questions about (1) the effectiveness of deep learning models training (for EEG data) and (2) the need for a systematic characterization of the input EEG data to ensure robust modeling.