Iterative Independent Component Analysis Research Articles

Independent Component Extraction from the Incomplete Coordinate Time Series of Regional GNSS Networks

Independent component analysis (ICA) is one of the most effective approaches in extracting independent signals from a global navigation satellite system (GNSS) regional station network. However, ICA requires the involved time series to be complete, thereby the missing data of incomplete time series should be interpolated beforehand. In this contribution, a modified ICA is proposed, by which the missing data are first recovered based on the reversible property between the original time series and decomposed principal components, then the complete time series are further processed with FastICA. To evaluate the performance of the modified ICA for extracting independent components, 24 regional GNSS network stations located in North China from 2011 to 2019 were selected. After the trend, annual and semiannual terms were removed from the GNSS time series, the first two independent components captured 17.42, 18.44 and 17.38% of the total energy for the North, East and Up coordinate components, more than those derived by the iterative ICA that accounted for 16.21%, 17.72% and 16.93%, respectively. Therefore, modified ICA can extract more independent signals than iterative ICA. Subsequently, selecting the 7 stations with less missing data from the network, we repeatedly process the time series after randomly deleting parts of the data and compute the root mean square error (RMSE) from the differences of reconstructed signals before and after deleting data. All RMSEs of modified ICA are smaller than those of iterative ICA, indicating that modified ICA can extract more exact signals than iterative ICA.

An Iterative ICA-Based Reconstruction Method to Produce Consistent Time-Variable Total Water Storage Fields Using GRACE and Swarm Satellite Data

Observing global terrestrial water storage changes (TWSCs) from (inter-)seasonal to (multi-)decade time-scales is very important to understand the Earth as a system under natural and anthropogenic climate change. The primary goal of the Gravity Recovery And Climate Experiment (GRACE) satellite mission (2002–2017) and its follow-on mission (GRACE-FO, 2018–onward) is to provide time-variable gravity fields, which can be converted to TWSCs with ∼ 300 km spatial resolution; however, the one year data gap between GRACE and GRACE-FO represents a critical discontinuity, which cannot be replaced by alternative data or model with the same quality. To fill this gap, we applied time-variable gravity fields (2013–onward) from the Swarm Earth explorer mission with low spatial resolution of ∼ 1500 km. A novel iterative reconstruction approach was formulated based on the independent component analysis (ICA) that combines the GRACE and Swarm fields. The reconstructed TWSC fields of 2003–2018 were compared with a commonly applied reconstruction technique and GRACE-FO TWSC fields, whose results indicate a considerable noise reduction and long-term consistency improvement of the iterative ICA reconstruction technique. They were applied to evaluate trends and seasonal mass changes (of 2003–2018) within the world’s 33 largest river basins.

A simple iterative independent component analysis algorithm for vibration source signal identification of complex structures

Independent Component Analysis (ICA), one of the blind source separation methods, can be applied for extracting unknown source signals only from received signals. This is accomplished by finding statistical independence of signal mixtures and has been successfully applied to myriad fields such as medical science, image processing, and numerous others. Nevertheless, there are inherent problems that have been reported when using this technique: instability and invalid ordering of separated signals, particularly when using a conventional ICA technique in vibratory source signal identification of complex structures. In this study, a simple iterative algorithm of the conventional ICA has been proposed to mitigate these problems. The proposed method to extract more stable source signals having valid order includes an iterative and reordering process of extracted mixing matrix to reconstruct finally converged source signals, referring to the magnitudes of correlation coefficients between the intermediately separated signals and the signals measured on or nearby sources. In order to review the problems of the conventional ICA technique and to validate the proposed method, numerical analyses have been carried out for a virtual response model and a 30 m class submarine model. Moreover, in order to investigate applicability of the proposed method to real problem of complex structure, an experiment has been carried out for a scaled submarine mockup. The results show that the proposed method could resolve the inherent problems of a conventional ICA technique.

Neural spike sorting using iterative ICA and a deflation-based approach

We propose a spike sorting method for multi-channel recordings. When applied in neural recordings, the performance of the independent component analysis (ICA) algorithm is known to be limited, since the number of recording sites is much lower than the number of neurons. The proposed method uses an iterative application of ICA and a deflation technique in two nested loops. In each iteration of the external loop, the spiking activity of one neuron is singled out and then deflated from the recordings. The internal loop implements a sequence of ICA and sorting for removing the noise and all the spikes that are not fired by the targeted neuron. Then a final step is appended to the two nested loops in order to separate simultaneously fired spikes. We solve this problem by taking all possible pairs of the sorted neurons and apply ICA only on the segments of the signal during which at least one of the neurons in a given pair was active. We validate the performance of the proposed method on simulated recordings, but also on a specific type of real recordings: simultaneous extracellular–intracellular. We quantify the sorting results on the extracellular recordings for the spikes that come from the neurons recorded intracellularly. The results suggest that the proposed solution significantly improves the performance of ICA in spike sorting.

Consistency of the auditory evoked response: The presence of aberrant responses and their effect on N100 localization

The structure and distribution of the sources underlying the generation of evoked potentials (EPs) is often very complex. In an effort to improve localization accuracy of the auditory N100 (negative response occurring around 100ms poststimulus) component, we analyzed 13 datasets of single-trial EPs obtained from normal subjects using an iterative independent component analysis procedure which allowed us to detect a clear N100 component in each single trial and to study gross changes in component morphology across trials. We found that single-trial N100 amplitude was most often negative in polarity, as expected, but occasionally exhibited a marked reversal to become positive. The average N100, however, showed the typical negative polarity, in all subjects.Based on this observation, we separated the processed single trials in two groups of typical and aberrant responses, and from each group, we computed a partial EP that was used to localize the underlying intracranial sources. Additionally, we localized the classical ensemble average EP. Before processing, the N100 sources were identified correctly in the primary auditory cortex in only four datasets, while after processing, all 13 datasets yielded correct localizations, and the confidence volume of the sources improved by about 80%. Further analysis demonstrated that in nine datasets the improvement was mostly due to the typical responses, while the aberrant responses had an antagonistic effect.Our results suggest that aberrant responses should not be included in source localizations, especially when EEG-based brain mapping is intended as a clinical tool.

Single-trial analysis of auditory evoked potentials improves separation of normal and schizophrenia subjects

ObjectiveIn this study, we employed iterative independent component analysis of single-trial auditory evoked responses to identify features of the P50 and N100 components that provide maximum separation between normal controls and schizophrenia subjects and compared the results against classical ensemble averaging. MethodWe analyzed data from 13 schizophrenia and 20 normal control subjects. Responses were obtained in a paired-stimulus paradigm, in which an auditory stimulus S1 is followed by an identical S2. The amplitude and latency of the P50 and N100 components in response to the S1 and S2 stimuli were measured in each single trial and used as features to classify the responses into two groups. Several methods were used for classification, while their performance was quantified in a 10-fold stratified cross-validation approach. ResultsWe found that normal controls tended to respond earlier and their individual responses had significantly higher amplitude (p<0.01) and significantly less latency variability (p<0.01) compared to schizophrenia patients. The S1 latency was the most significant discriminatory feature (p<0.01) followed by S2 latency (p<0.01). The S2 amplitude, though relatively larger in normal subjects (p<0.05), was the least discriminatory feature. Classification based on single-trial analysis yielded 100% accuracy, while the classical ensemble averaging yielded only a maximum of 76% accuracy. ConclusionsOur results demonstrate that single-trial analysis can accurately separate schizophrenia patients from normal controls and suggest that inter-trial variability plays a significant role in information processing in the human brain. SignificanceThe proposed technique may have a significant impact as a clinical tool in the quest for identifying physiological markers of schizophrenia.

ICA Based Automatic Segmentation of Dynamic &lt;formula formulatype="inline"&gt;&lt;tex Notation="TeX"&gt;${\bf H}_{\bf 2}^{\bf 15}{\bf O}$&lt;/tex&gt;&lt;/formula&gt; Cardiac PET Images

In this study, we applied an iterative independent component analysis (ICA) method for the separation of cardiac tissue components (myocardium, right, and left ventricle) from dynamic positron emission tomography (PET) images. Previous phantom and animal studies have shown that ICA separation extracts the cardiac structures accurately. Our goal in this study was to investigate the methodology with human studies. The ICA separated cardiac structures were used to calculate the myocardial perfusion in two different cases: 1) the regions of interest were drawn manually on the ICA separated component images and 2) the volumes of interest (VOI) were automatically segmented from the component images. For the whole myocardium, the perfusion values of 25 rest and six drug-induced stress studies obtained with these methods were compared to the values from the manually drawn regions of interest on differential images. The separation of the rest and stress studies using ICA-based methods was successful in all cases. The visualization of the cardiac structures from H (2) (15) O PET studies was improved with the ICA separation. Also, the automatic segmentation of the VOI seemed to be feasible.

Estimation of individual evoked potential components using iterative independent component analysis

Independent component analysis (ICA) has been successfully employed in the study of single-trial evoked potentials (EPs). In this paper, we present an iterative temporal ICA methodology that processes multielectrode single-trial EPs, one channel at a time, in contrast to most existing methodologies which are spatial and analyze EPs from all recording channels simultaneously. The proposed algorithm aims at enhancing individual components in an EP waveform in each single trial, and relies on a dynamic template to guide EP estimation. To quantify the performance of this method, we carried out extensive analyses with artificial EPs, using different models for EP generation, including the phase-resetting and the classical additive-signal models, and several signal-to-noise ratios and EP component latency jitters. Furthermore, to validate the technique, we employed actual recordings of the auditory N100 component obtained from normal subjects. Our results with artificial data show that the proposed procedure can provide significantly better estimates of the embedded EP signals compared to plain averaging, while with actual EP recordings, the procedure can consistently enhance individual components in single trials, in all subjects, which in turn results in enhanced average EPs. This procedure is well suited for fast analysis of very large multielectrode recordings in parallel architectures, as individual channels can be processed simultaneously on different processors. We conclude that this method can be used to study the spatiotemporal evolution of specific EP components and may have a significant impact as a clinical tool in the analysis of single-trial EPs.

A feature-selective independent component analysis method for functional MRI.

In this work, we propose a simple and effective scheme to incorporate prior knowledge about the sources of interest (SOIs) in independent component analysis (ICA) and apply the method to estimate brain activations from functional magnetic resonance imaging (fMRI) data. We name the proposed method as feature-selective ICA since it incorporates the features in the sample space of the independent components during ICA estimation. The feature-selective scheme is achieved through a filtering operation in the source sample space followed by a projection onto the demixing vector space by a least squares projection in an iterative ICA process. We perform ICA estimation of artificial activations superimposed into a resting state fMRI dataset to show that the feature-selective scheme improves the detection of injected activation from the independent component estimated by ICA. We also compare the task-related sources estimated from true fMRI data by a feature-selective ICA algorithm versus an ICA algorithm and show evidence that the feature-selective scheme helps improve the estimation of the sources in both spatial activation patterns and the time courses.

Single-trial evoked potential estimation: Comparison between independent component analysis and wavelet denoising

Objective Brain responses to repeated sensory stimuli are typically buried in the more prominent background activity, and thus analysis of these responses on a single-trial basis would require advanced procedures to estimate the brain activity related only to the experimental task. Recently, we have proposed a new iterative independent component analysis ( iICA) approach to estimate single-trial responses. In this paper, we compare the performance of iICA at estimating single-trial responses with ensemble averaging and wavelet transform (WT) denoising. Methods We analyzed simulated evoked potentials (EPs) and actual recordings of the auditory N100 component from 33 normal subjects, and the performance of each method was quantified in terms of the average root-mean-square error and average correlation before and after processing. Results We found that WT gave a smoother overall average EP, while iICA could isolate the N100 component out of the entire EP waveform. With simulated data, iICA provided significantly better estimates of the true EP compared to plain averaging ( p < 0.01) and WT ( p < 0.01). With actual data, iICA showed clear responses in single trials, in all subjects. Additionally, the corresponding average EPs had a sharper N100–P200 complex, with flatter preceding and following regions, resulting in an enhanced N100 component. Conclusions The iICA procedure can provide clear responses in each single trial, and the resulting average N100 component is significantly improved compared to plain averaging and wavelet denoising. Significance The proposed technique may have a significant impact as a clinical tool in the analysis of single-trial responses.

Performance of blind source separation algorithms for fMRI analysis using a group ICA method

Independent component analysis (ICA) is a popular blind source separation technique that has proven to be promising for the analysis of functional magnetic resonance imaging (fMRI) data. A number of ICA approaches have been used for fMRI data analysis, and even more ICA algorithms exist; however, the impact of using different algorithms on the results is largely unexplored. In this paper, we study the performance of four major classes of algorithms for spatial ICA, namely, information maximization, maximization of non-Gaussianity, joint diagonalization of cross-cumulant matrices and second-order correlation-based methods, when they are applied to fMRI data from subjects performing a visuo-motor task. We use a group ICA method to study variability among different ICA algorithms, and we propose several analysis techniques to evaluate their performance. We compare how different ICA algorithms estimate activations in expected neuronal areas. The results demonstrate that the ICA algorithms using higher-order statistical information prove to be quite consistent for fMRI data analysis. Infomax, FastICA and joint approximate diagonalization of eigenmatrices (JADE) all yield reliable results, with each having its strengths in specific areas. Eigenvalue decomposition (EVD), an algorithm using second-order statistics, does not perform reliably for fMRI data. Additionally, for iterative ICA algorithms, it is important to investigate the variability of estimates from different runs. We test the consistency of the iterative algorithms Infomax and FastICA by running the algorithm a number of times with different initializations, and we note that they yield consistent results over these multiple runs. Our results greatly improve our confidence in the consistency of ICA for fMRI data analysis.

Spectral tuning of visual pigments: the rhodopsins of cottoid fish from Lake Baikal

Brain responses to repeated sensory stimuli are typically buried in the more prominent background activity, and thus analysis of these responses on a single-trial basis would require advanced procedures to estimate the brain activity related only to the experimental task. Recently, we have proposed a new iterative independent component analysis (iICA) approach to estimate single-trial responses. In this paper, we compare the performance of iICA at estimating single-trial responses with ensemble averaging and wavelet transform (WT) denoising.We analyzed simulated evoked potentials (EPs) and actual recordings of the auditory N100 component from 33 normal subjects, and the performance of each method was quantified in terms of the average root-mean-square error and average correlation before and after processing.We found that WT gave a smoother overall average EP, while iICA could isolate the N100 component out of the entire EP waveform. With simulated data, iICA provided significantly better estimates of the true EP compared to plain averaging (p < 0.01) and WT (p < 0.01). With actual data, iICA showed clear responses in single trials, in all subjects. Additionally, the corresponding average EPs had a sharper N100–P200 complex, with flatter preceding and following regions, resulting in an enhanced N100 component.The iICA procedure can provide clear responses in each single trial, and the resulting average N100 component is significantly improved compared to plain averaging and wavelet denoising.The proposed technique may have a significant impact as a clinical tool in the analysis of single-trial responses.