Large Sample Group Independent Component Analysis of Functional Magnetic Resonance Imaging Using Anatomical Atlas-Based Reduction and Bootstrapped Clustering

Temporal embedding and spatiotemporal feature selection boost multi-voxel pattern analysis decoding accuracy

BACKGROUND

In fMRI decoding, temporal embedding of spatial features of the brain allows the incorporation of brain activity dynamics into the multivariate pattern classification process, and provides enriched information about stimulus-specific response patterns and potentially improved prediction accuracy.

NEW METHOD

This study investigates the possibility of enhancing the classification performance by exploring temporal embedding, to identify the optimum combination of spatiotemporal features based on their classification performance. We investigated the importance of spatiotemporal feature selection using a slow event-related design adapted from the classic Haxby study (Haxby et al., 2001). Data were collected using a multiband fMRI sequence with temporal resolution of 0.568 s.

COMPARISON WITH EXISTING METHODS

A wide range of spatiotemporal observations were created as various combinations of spatiotemporal features. Using both random forest, and support vector machine, classifiers prediction accuracies for these combinations were then compared with the single spatial multivariate pattern approach that uses only a single temporal observation.

RESULTS

Our findings showed that, on average, spatiotemporal feature selection improved prediction accuracy. Moreover, the random forest algorithm outperformed the support vector machine and benefitted from temporal information to a greater extent.

CONCLUSIONS

As expected, the most influential temporal durations were found to be around the peak of the hemodynamic response function, a few seconds after the stimuli onset until -4 s after the peak of the hemodynamic response function. The superiority of spatiotemporal feature selection over single time-point spatial approaches invites future work to design optimal approaches that incorporate spatiotemporal dependencies into feature selection for decoding.

Decreased small-world functional network connectivity and clustering across resting state networks in schizophrenia: an fMRI classification tutorial

Functional network connectivity (FNC) is a method of analyzing the temporal relationship of anatomical brain components, comparing the synchronicity between patient groups or conditions. We use functional-connectivity measures between independent components to classify between Schizophrenia patients and healthy controls during resting-state. Connectivity is measured using a variety of graph-theoretic connectivity measures such as graph density, average path length, and small-worldness. The Schizophrenia patients showed significantly less clustering (transitivity) among components than healthy controls (p < 0.05, corrected) with networks less likely to be connected, and also showed lower small-world connectivity than healthy controls. Using only these connectivity measures, an SVM classifier (without parameter tuning) could discriminate between Schizophrenia patients and healthy controls with 65% accuracy, compared to 51% chance. This implies that the global functional connectivity between resting-state networks is altered in Schizophrenia, with networks more likely to be disconnected and behave dissimilarly for diseased patients. We present this research finding as a tutorial using the publicly available COBRE dataset of 146 Schizophrenia patients and healthy controls, provided as part of the 1000 Functional Connectomes Project. We demonstrate preprocessing, using independent component analysis (ICA) to nominate networks, computing graph-theoretic connectivity measures, and finally using these connectivity measures to either classify between patient groups or assess between-group differences using formal hypothesis testing. All necessary code is provided for both running command-line FSL preprocessing, and for computing all statistical measures and SVM classification within R. Collectively, this work presents not just findings of diminished FNC among resting-state networks in Schizophrenia, but also a practical connectivity tutorial.

Single trial decoding of belief decision making from EEG and fMRI data using independent components features

The complex task of assessing the veracity of a statement is thought to activate uniquely distributed brain regions based on whether a subject believes or disbelieves a given assertion. In the current work, we present parallel machine learning methods for predicting a subject's decision response to a given propositional statement based on independent component (IC) features derived from EEG and fMRI data. Our results demonstrate that IC features outperformed features derived from event related spectral perturbations derived from any single spectral band, yet were similar to accuracy across all spectral bands combined. We compared our diagnostic IC spatial maps with our conventional general linear model (GLM) results, and found that informative ICs had significant spatial overlap with our GLM results, yet also revealed unique regions like amygdala that were not statistically significant in GLM analyses. Overall, these results suggest that ICs may yield a parsimonious feature set that can be used along with a decision tree structure for interpretation of features used in classifying complex cognitive processes such as belief and disbelief across both fMRI and EEG neuroimaging modalities.

The utility of data-driven feature selection: re: Chu et al. 2012

The recent Chu et al. (2012) manuscript discusses two key findings regarding feature selection (FS): (1) data driven FS was no better than using whole brain voxel data and (2) a priori biological knowledge was effective to guide FS. Use of FS is highly relevant in neuroimaging-based machine learning, as the number of attributes can greatly exceed the number of exemplars. We strongly endorse their demonstration of both of these findings, and we provide additional important practical and theoretical arguments as to why, in their case, the data-driven FS methods they implemented did not result in improved accuracy. Further, we emphasize that the data-driven FS methods they tested performed approximately as well as the all-voxel case. We discuss why a sparse model may be favored over a complex one with similar performance. We caution readers that the findings in the Chu et al. report should not be generalized to all data-driven FS methods.

Time course based artifact identification for independent components of resting-state FMRI

In functional magnetic resonance imaging (fMRI) coherent oscillations of the blood oxygen level-dependent (BOLD) signal can be detected. These arise when brain regions respond to external stimuli or are activated by tasks. The same networks have been characterized during wakeful rest when functional connectivity of the human brain is organized in generic resting-state networks (RSN). Alterations of RSN emerge as neurobiological markers of pathological conditions such as altered mental state. In single-subject fMRI data the coherent components can be identified by blind source separation of the pre-processed BOLD data using spatial independent component analysis (ICA) and related approaches. The resulting maps may represent physiological RSNs or may be due to various artifacts. In this methodological study, we propose a conceptually simple and fully automatic time course based filtering procedure to detect obvious artifacts in the ICA output for resting-state fMRI. The filter is trained on six and tested on 29 healthy subjects, yielding mean filter accuracy, sensitivity and specificity of 0.80, 0.82, and 0.75 in out-of-sample tests. To estimate the impact of clearly artifactual single-subject components on group resting-state studies we analyze unfiltered and filtered output with a second level ICA procedure. Although the automated filter does not reach performance values of visual analysis by human raters, we propose that resting-state compatible analysis of ICA time courses could be very useful to complement the existing map or task/event oriented artifact classification algorithms.

Likelihood-based population independent component analysis

Independent component analysis (ICA) is a widely used technique for blind source separation, used heavily in several scientific research areas including acoustics, electrophysiology, and functional neuroimaging. We propose a scalable two-stage iterative true group ICA methodology for analyzing population level functional magnetic resonance imaging (fMRI) data where the number of subjects is very large. The method is based on likelihood estimators of the underlying source densities and the mixing matrix. As opposed to many commonly used group ICA algorithms, the proposed method does not require significant data reduction by a 2-fold singular value decomposition. In addition, the method can be applied to a large group of subjects since the memory requirements are not restrictive. The performance of our approach is compared with a commonly used group ICA algorithm via simulation studies. Furthermore, the proposed method is applied to a large collection of resting state fMRI datasets. The results show that established brain networks are well recovered by the proposed algorithm.