What are the most effective solutions to clean EEG recorded during fMRI?4 answersThe most effective solutions for cleaning EEG recorded during fMRI include the use of ballistocardiographic (BCG) artifact correction methods such as Optimal Basis Set (OBS), Independent Component Analysis (ICA), and a combination of OBS and ICA. Another approach is the use of a new open-source BCG removal software called EEG-LLAMAS, which adapts and advances existing artifact removal techniques for low-latency experiments. Surrogate source models, created using principal components analysis (PCA-S) or manual selection of artifact components utilizing independent components analysis (ICA-S), have also been shown to effectively separate artifact-related signals from brain signals with minimal distortion. Additionally, a comprehensive open-source toolbox called APPEAR integrates average template subtraction and independent component analysis to suppress both MRI-related and physiological artifacts. These methods have been validated through simulations and real-time EEG-fMRI recordings, demonstrating their ability to recover EEG waveforms, power spectra, and brain activity of interest.
Why is Z-Score Normalisation necessary in pre-processing EEG signal?3 answersZ-Score Normalization is necessary in pre-processing EEG signals because it allows for the comparison of EEG data across different recording sessions and subjects. EEG signals are non-stationary, meaning they can vary in each recording session, making it difficult to compare data directly. Z-Score Normalization addresses this issue by transforming the EEG data into a standardized distribution with a mean of 0 and a standard deviation of 1. This normalization technique ensures that the EEG data from different sessions and subjects are on the same scale, allowing for meaningful comparisons and analysis. Additionally, Z-Score Normalization simplifies the EEG biofeedback process by providing a reference point for determining threshold settings.
What are the different methods for reducing noise in EEG recordings?5 answersDifferent methods for reducing noise in EEG recordings include adaptive noise cancellation (ANC) with linear regression and modified independent component analysis (ICA). ANC is effective in removing line noise artifact without removing physiological activity, but it is sensitive to the quality of the reference signal. Another approach is the use of variational autoencoders (VAE) algorithm, which reduces noise in EEG signals by training neural networks with stochastic gradient descent. VAE has been compared with the fast fixed-point algorithm for independent component analysis (FastICA) to measure performance using machine learning algorithms. Preprocessing steps such as band-pass filtering, re-referencing, segmenting, removal of bad channels and trials, and independent component analysis can also be used to reduce artifacts in EEG recordings.
What is the relationship between reduced power specturm and connectivity of scalp?3 answersReduced power spectrum in the scalp is related to changes in connectivity. Studies have shown that chronic recurrent seizures lead to differences in brain activity, including power spectral density (PSD) and functional connectivity. In individuals with consecutive epileptic bursts, postseizure energy accumulation is observed, indicating impaired brain function. Additionally, total sleep deprivation (TSD) has been found to cause a decrease in alpha-band power and an increase in delta-band power, along with impaired functional connectivity in specific brain regions. These findings suggest that changes in power spectrum are associated with alterations in functional connectivity, indicating a disruption in the normal functioning of the brain.
How can SPD domain-specific batch normalization be used to improve the performance of unsupervised domain adaptation in EEG?5 answersSPD domain-specific batch normalization (SPDDSMBN) can be used to improve the performance of unsupervised domain adaptation in EEG. SPDDSMBN is a building block for geometric deep learning that transforms domain-specific SPD inputs into domain-invariant SPD outputs. It can be applied to multi-source/-target and online UDA scenarios. By using SPDDSMBN, a theory-based machine learning framework enables learning domain-invariant tangent space mapping (TSM) models in an end-to-end fashion. This approach achieves state-of-the-art performance in inter-session and -subject transfer learning (TL) with a simple, interpretable network architecture called TSMNet. The proposed method leverages the knowledge in a label-rich source domain to facilitate learning in an unlabeled target domain with a different distribution. It captures domain-specific information and aligns domains and labels to learn more robust and discriminative invariant features for domain adaptation.
How to normalize non parametric data from different psychometric scales?4 answersNon-parametric methods can be used to normalize non-parametric data from different psychometric scales. These methods do not require the data to follow a specific distribution and can handle different types of response variables measured on different scales. Non-parametric tests, such as permutation tests, can be used to compare multivariate data samples and identify significant subsets of response variables and factor levels. These tests can be applied to low- or high-dimensional data with small or large sample sizes. Additionally, non-parametric rank statistics can be used for testing spectral power and coherence in neural signals, providing robustness against artefactual components. These non-parametric methods offer new possibilities for testing the complex coherency function, including both magnitude and phase. Therefore, non-parametric methods are recommended for normalizing non-parametric data from different psychometric scales.