Group information guided ICA for fMRI data analysis

Du, Yuhui; Fan, Yong

doi:10.1016/j.neuroimage.2012.11.008

Cited by 339 publications

(371 citation statements)

References 44 publications

Supporting

Mentioning

370

Contrasting

Order By: Relevance

“…Data were back-reconstructed using the moo-icar option, i.e. the multivariate objective optimization ICA with reference (Du and Fan, 2013). Since the reconstruction adds some noise to component activity, a 35 Hz low pass filter was reapplied to the time courses with the use of the EEGLab iirfilt() function.…”

Section: Eeg Data Acquisition and Analysismentioning

confidence: 99%

When three is greater than five: EEG and fMRI signatures of errors in numerical and physical comparisons

et al. 2017

View full text Add to dashboard Cite

Unravelling the neural mechanisms, which determine performance accuracy, is one of the key concepts in cognitive neuroscience. When compared to correct responses, shorter reaction times are commonly observed behavioural feature of errors committed in typical conflict tasks. Yet, little is known about the origins of this phenomenon. In this study, EEG and fMRI experiments were conducted using the numerical version of the Stroop paradigm, which yielded unique behavioural outcomes. Particularly, errors in numerical comparison had shorter reaction times than correct trials, whereas physical comparison resulted in the opposite pattern. This criss-crossing interaction effect was used as a marker when exploring time-courses of brain activity. Group independent component analysis was applied to neurophysiological data and event-related analysis was conducted on the components' time-courses. Results revealed one centro-parietal EEG component and one temporo-parietal fMRI neural network, which exhibited significant task and accuracy interactions. Showing linear increase that peaked right after the response onset, the activity of centro-parietal EEG component was linked to the decision variable signal, which reflects a process of accumulating evidence until reaching an actiontriggering threshold. Both amplitude measurements and linear fits to the signal provided evidence for distinctive characteristics between numerical and physical comparisons, thereby explaining the behavioural outcomes: errors are committed due to accumulation of evidence in favour of the other (wrong) task instruction. The architecture of the temporo-parietal network, which comprises bilateral inferior temporal and intraparietal regions, is highly consistent with the recently established core ''number network''. These findings link perceptual decisions with the generalized magnitude system and impart novel insights into the neural determinants of errors in humans.

show abstract

Section: Eeg Data Acquisition and Analysismentioning

confidence: 99%

When three is greater than five: EEG and fMRI signatures of errors in numerical and physical comparisons

et al. 2017

View full text Add to dashboard Cite

show abstract

“…The final sample included six patients (Table 2). GIG-ICA 20 was performed across D+ and DÀ datasets using the seven group spatial maps of interest (i.e., the ICs selected from the first group decomposition across all participants) as spatial priors. Paired t-tests were used to explore differences in FNC between D+ and DÀ datasets.…”

Section: Network Interactions During Discharge-affected and Dischargementioning

confidence: 99%

Abnormal cognitive network interactions in Lennox‐Gastaut syndrome: A potential mechanism of epileptic encephalopathy

et al. 2016

View full text Add to dashboard Cite

SUMMARYObjective: In patients with Lennox-Gastaut syndrome (LGS), recurrent epileptic activity is thought to contribute to impaired cognition (epileptic encephalopathy). Using concurrent electroencephalography-functional magnetic resonance imaging (EEG-fMRI), we recently showed that epileptiform discharges in LGS recruit largescale networks that normally support key cognitive processes. In LGS, given that epileptic activity engages cognitive networks, and cognition is pervasively impaired, we hypothesized that cognitive network interactions in LGS are persistently abnormal. Methods: We studied 15 LGS patients (mean age AE 1 standard deviation [SD] = 28.7 AE 10.6 years) and 17 healthy controls (mean age AE 1 SD = 27.6 AE 6.6 years) using task-free EEG-fMRI. Four networks of interest (default-mode, dorsal attention, executive control, and anterior salience) were defined using group-level independent components analysis (ICA). Functional connectivity within and between networks was determined for each subject, and then LGS network interactions were compared to network behavior in the control group. To test whether group differences were present in periods without scalp-detectable epileptiform discharges (i.e., persistent), we separately assessed discharge-affected and discharge-unaffected epochs in six patients with sufficient data for this analysis. Results: In LGS, cognitive networks showed (1) reduced within-network integration, including weaker connectivity within the default-mode network, and (2) impaired between-network segregation, including stronger connectivity between the defaultmode and dorsal attention networks. Abnormal interactions were present during fMRI periods with and without discharges, indicating that impaired network behavior may endure during periods without scalp-detectable epileptic activity. Significance: In LGS, cognitive network interactions are persistently abnormal. Given that cognition typically worsens with the onset of LGS, and may improve after seizure control, our findings are consistent with the hypothesis that the epileptic process in LGS may initiate and perhaps sustain abnormal network behavior. We propose that epileptic encephalopathy may be a consequence of persistently disrupted cognitive network interactions.

show abstract

“…GIG-ICA involves steps: (1) application of group level ICA to all subject datasets, (2) identification of artifact group ICs, (3) computation of individual ICs via one-unit ICA with reference algorithm on individual datasets using the non-artifact group ICs as spatial references [7] , and (4) computation of individual TCs using regression. Relevant free parameters in GIG-ICA include the number of PCs in the subject-level PCAs of step 1 (G1), and the number of PCs/ICs in the group-level PCA/ICA of step 1 (G2).…”

Section: Methodsmentioning

confidence: 99%

“…This is particularly the case if many subjects are involved, although approaches for training classifiers can help mitigate this to a degree [6] . The second approach skips the expensive single-subject ICA step and directly applies a variant of group ICA based on a new one-unit ICA with reference algorithm [7] . This approach, called Group Information Guided ICA (GIG-ICA) [7] , implements ICA on all data, and then uses the non-artifact group ICs as references to compute individual networks.…”

Section: Introductionmentioning

confidence: 99%

“…The second approach skips the expensive single-subject ICA step and directly applies a variant of group ICA based on a new one-unit ICA with reference algorithm [7] . This approach, called Group Information Guided ICA (GIG-ICA) [7] , implements ICA on all data, and then uses the non-artifact group ICs as references to compute individual networks. Compared to IRPG, GIG-ICA does not require individual identification of artifact ICs for each subject.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Brain functional networks extraction based on fMRI artifact removal: Single subject and group approaches

Allen

et al. 2014

2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society

View full text Add to dashboard Cite

Abstract-Independent component analysis (ICA) has been widely applied to identify brain functional networks from multiple-subject fMRI. However, the best approach to handle artifacts is not yet clear. In this work, we study and compare two ICA approaches for artifact removal using simulations and real fMRI data. The first approach, recommended by the human connectome project, performs ICA on individual data to remove artifacts, and then applies group ICA on the cleaned data from all subjects. We refer to this approach as Individual ICA artifact Removal Plus Group ICA (IRPG). A second approach, Group Information Guided ICA (GIG-ICA), performs ICA on group data, and then removes the artifact group independent components (ICs), followed by individual subject ICA using the remaining group ICs as spatial references. Experiments demonstrate that GIG-ICA is more accurate in estimation of sources and time courses, more robust to data quality and quantity, and more reliable for identifying networks than IRPG.

show abstract

Group information guided ICA for fMRI data analysis

Cited by 339 publications

References 44 publications

When three is greater than five: EEG and fMRI signatures of errors in numerical and physical comparisons

When three is greater than five: EEG and fMRI signatures of errors in numerical and physical comparisons

Abnormal cognitive network interactions in Lennox‐Gastaut syndrome: A potential mechanism of epileptic encephalopathy

Brain functional networks extraction based on fMRI artifact removal: Single subject and group approaches

Contact Info

Product

Resources

About