China
Africa
Explore chapters and articles related to this topic
Review of the Human Brain and EEG Signals
Published in Teodiano Freire Bastos-Filho, Introduction to Non-Invasive EEG-Based Brain–Computer Interfaces for Assistive Technologies, 2020
Alessandro Botti Benevides, Alan Silva da Paz Floriano, Mario Sarcinelli-Filho, Teodiano Freire Bastos-Filho
The two mentioned ICA algorithms are based on the two broadest definitions of independence for ICA that are the maximization of the non-Gaussianity and the minimization of the mutual information (MMI) [13,24,25,32]. The non-Gaussianity family of ICA algorithms, which includes “fastICA” algorithm, is based on measures of non-Gaussianity such as kurtosis30 and negentropy [34]. As the Gaussian distribution has above second-order cumulants equal to zero, the functions of its third- and fourth-order cumulants, such as the skewness31 and kurtosis, are also zero, and then, they are used as a measurement for non-Gaussianity. The negentropy, or differential entropy, can also be approximated by a function of the first- and fourth-order cumulants, and then, it is also used as a measurement of non-Gaussianity. The “fastICA” is a fixed point algorithm, which, maximizing the absolute value of the kurtosis, leads to the identification of non-Gaussian sources [33].
Skin image analysis granulation tissue for healing assessment of chronic ulcers
Published in Ahmad Fadzil Mohamad Hani, Dileep Kumar, Optical Imaging for Biomedical and Clinical Applications, 2017
Ahmad Fadzil Mohamad Hani, Leena Arshad
ICA is applied on the observation dataset to excerpt the source grey-level images that demonstrate the spread of the pigment haemoglobin representing the regions of granulation tissue in the colour images of chronic ulcers. ICA is a multivariate data analysis approach used to recover source signals from their observed linear combinations. ICA assumes that the observed signals, in this case the RGB data vectors of each picture of ulcer, are generated by linearly mixing original source signals with an unknown mixing matrix [29,30]. ICA estimates the source images assuming that they are statistically independent, which is achieved when their probability density function can be derived as the outcome of their marginal independent distributions [30]. In this study, ICA is implemented using the FastICA that is referencing on a fixed-point iteration, which utilises maximisation of non-Gaussianity as a measurement of independence in order to project the independent components [29,31–33]. In this method, an approximation of negentropy, which is Newton's iterative approximation, is used to measure non-Gaussianity and estimate the independent source images.
Exploratory Data Analysis with Unsupervised Machine Learning
Published in Altuna Akalin, Computational Genomics with R, 2020
ICA requires that the columns of the S matrix, the “metagenes” in our example above, are statistically independent. This is a stronger constraint than uncorrelatedness. In this case, there should be no relationship between non-linear transformation of the data either. There are different ways of ensuring this statistical indepedence and this is the main constraint when finding the optimal A and S matrices. The various ICA algorithms use different proxies for statistical independence, and the definition of that proxy is the main difference between many ICA algorithms. The algorithm we are going to use requires that metagenes or sources in the S matrix are non-Gaussian (non-normal) as possible. Non-Gaussianity is shown to be related to statistical independence (Hyvärinen, 2013). Below, we are using the fastICA::fastICA() function to extract 2 components and plot the rows of matrix A which represents metagenes shown in Figure 4.16. This way, we can visualize samples in a 2D plot. If we wanted to plot the relationship between genes we would use the columns of matrix S.
The Influence of Experience on Neuromuscular Control of the Body When Cutting at Different Angles
Published in Journal of Motor Behavior, 2023
Zhengye Pan, Lushuai Liu, Xingman Li, Yunchao Ma
To describe the spinal motor output pattern, the 13 sEMG signals collected were mapped to the rostrocaudal location of the pool of alpha-motor neurons (MNs) in the ninth thoracic (T9) to the third sacral (S3) vertebral segments, with the thoracic segment (T9-12) predominantly innervating the trunk muscle groups and the lumbar (L1-5) and sacral (S1-3) segments predominantly innervating the lower limb muscle groups. The original sEMG signal is high-pass filtered (50 Hz), full-wave rectified and low-pass filtered (20 Hz) using a 4th-order IIR Butterworth zero-phase filter (Santuz et al., 2017a) based on Python (v3.9.13, Delawwere, US), and a linear envelope is constructed with amplitude normalisation based on maximum activation (Santuz et al., 2017b). In addition, artefacts in sEMG were removed using fastICA (Hu et al., 2007). The contribution of each muscle to the activity of the spinal cord segments is calculated using the neuromuscular map (Kendall et al., 2014). The motor output of each spinal cord segment Sj is estimated using the following equation, assuming a common spinal topography among the investigated participants: mj is the muscle innervated by each spinal cord segment, ni is the number of spinal cord segments innervating ith muscle, and kij is the weight factor of each muscle relative to the innervated spinal cord segment (la Scaleia et al., 2014).
Attentional dysfunction and recovery in concussion: effects on the P300m and contingent magnetic variation
Published in Brain Injury, 2018
Lauren Petley, Tim Bardouille, Darrell Chiasson, Patrick Froese, Steve Patterson, Aaron Newman, Antonina Omisade, Steven Beyea
Sources of magnetic noise external to the participant were first removed via spatiotemporal signal space separation (45) using default parameters. The data was then low-pass filtered (40 Hz cutoff) and downsampled to 250 Hz using the MaxFilter software (Elekta Oy). A 0.1 Hz highpass filter was subsequently applied using MNE Suite (version 2.7.0 (46);). Head movement compensation was not applied. MEG artifact correction was performed in Matlab (R2015b; Mathworks Inc.) using independent component analysis (FastICA v2.5, based on (47)). All components with a Pearson correlation ≥ 0.4 with one of the artefact reference channels (horizontal EOG, vertical EOG, and ECG), or with amplitudes that exceeded a manually-optimized threshold (1e−10 T for magnetometers, 1e−9 T/cm for gradiometers) were removed from the MEG signal. The corrected MEG signal was then segmented and averaged in 3500 ms epochs that were time-locked to cue onset, with a 500 ms pre-cue baseline separately for each combination of cue and flanker condition. Epoching and averaging were performed via command-line calls to the MNE Suite software package (version 2.7.0 (46);).
Human resting-state EEG and radiofrequency GSM mobile phone exposure: the impact of the individual alpha frequency
Published in International Journal of Radiation Biology, 2022
Jasmina Wallace, Lydia Yahia-Cherif, Christophe Gitton, Laurent Hugueville, Jean-Didier Lemaréchal, Brahim Selmaoui
Resting EEG data of baseline phases (Run 3 and Run 4) and exposure phases (from Run 5 to Run 8) were analyzed using MNE-Python, Version 0.19.2 (Gramfort et al. 2013). Electrodes near to ear-slits of the EEG cap (i.e., T9, TP9, T10, TP10) were removed from all data analysis for all subjects. The signal was band-pass filtered between 1–20 Hz using Finite Impulse Response (FIR) filter design with Hamming window. Independent component analysis (ICA) with FastICA algorithm was performed to decompose EEG signal and reject ocular artifacts. Pearson correlations were used to find vertical and horizontal EOG related components and bad epochs were created 500 ms before and after each blink, with band pass filtering between 1 and 10 Hz.