Identifying Predictive Regions from fMRI with TV-L1 Prior

Gramfort, Alexandre; Thirion, Bertrand; Varoquaux, Gaël

doi:10.1109/prni.2013.14

Cited by 61 publications

(67 citation statements)

References 9 publications

(24 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compare our regularizer against several common regularizers ( 1 and 2 ) and popular structured regularizers for problems with similar structure. In recent work TV + 1 , which adds the TV and 1 constraints, has been heavily utilized for data with similar spatial assumptions (Dohmatob et al 2014;Gramfort et al 2013) and is thus one of our main benchmarks. Source code for learning with the k-support/TV regularizer is available at https://github.com/eugenium/StructuredSparsityRegularization.…”

Section: Resultsmentioning

confidence: 99%

Convex relaxations of penalties for sparse correlated variables with bounded total variation

2015

View full text Add to dashboard Cite

We study the problem of statistical estimation with a signal known to be sparse, spatially contiguous, and containing many highly correlated variables. We take inspiration from the recently introduced k-support norm, which has been successfully applied to sparse prediction problems with correlated features, but lacks any explicit structural constraints commonly found in machine learning and image processing. We address this problem by incorporating a total variation penalty in the k-support framework. We introduce the (k, s) support total variation norm as the tightest convex relaxation of the intersection of a set of sparsity and total variation constraints. We show that this norm leads to an intractable combinatorial graph optimization problem, which we prove to be NP-hard. We then introduce a tractable relaxation with approximation guarantees that scale well for grid structured graphs. We devise several first-order optimization strategies for statistical parameter estimation with the described penalty. We demonstrate the effectiveness of this penalty on classification in the low-sample regime, classification with M/EEG neuroimaging data, and image recovery with synthetic and real data background subtracted image recovery tasks. We extensively analyse the application of our penalty on the complex task of identifying predictive regions from low-sample high-dimensional fMRI brain data, we show that our method is particularly useful compared to existing methods in terms of accuracy, interpretability, and stability.

show abstract

Section: Resultsmentioning

confidence: 99%

Convex relaxations of penalties for sparse correlated variables with bounded total variation

2015

View full text Add to dashboard Cite

show abstract

“…These methods differ in various aspects, including the selection process for voxels used to predict group membership. While L2-regularized (dense) methods use all available voxels to make a prediction, L1-regularized (sparse) methods choose only a subset of voxels to classify subjects (Grosenick et al, 2013;Gramfort et al, 2013). During training, i.e.…”

Section: Multivariate Group Comparisonsmentioning

confidence: 99%

Sparse regularization techniques provide novel insights into outcome integration processes

et al. 2015

View full text Add to dashboard Cite

By exploiting information that is contained in the spatial arrangement of neural activations, multivariate pattern analysis (MVPA) can detect distributed brain activations which are not accessible by standard univariate analysis. Recent methodological advances in MVPA regularization techniques have made it feasible to produce sparse discriminative whole-brain maps with highly specific patterns. Furthermore, the most recent refinement, the Graph Net, explicitly takes the 3D-structure of fMRI data into account. Here, these advanced classification methods were applied to a large fMRI sample (N=70) in order to gain novel insights into the functional localization of outcome integration processes. While the beneficial effect of differential outcomes is well-studied in trial-and-error learning, outcome integration in the context of instruction-based learning has remained largely unexplored. In order to examine neural processes associated with outcome integration in the context of instruction-based learning, two groups of subjects underwent functional imaging while being presented with either differential or ambiguous outcomes following the execution of varying stimulus-response instructions. While no significant univariate group differences were found in the resulting fMRI dataset, L1-regularized (sparse) classifiers performed significantly above chance and also clearly outperformed the standard L2-regularized (dense) Support Vector Machine on this whole-brain between-subject classification task. Moreover, additional L2-regularization via the Elastic Net and spatial regularization by the Graph Net improved interpretability of discriminative weight maps but were accompanied by reduced classification accuracies. Most importantly, classification based on sparse regularization facilitated the identification of highly specific regions differentially engaged under ambiguous and differential outcome conditions, comprising several prefrontal regions previously associated with probabilistic learning, rule integration and reward processing. Additionally, a detailed post-hoc analysis of these regions revealed that distinct activation dynamics underlay the processing of ambiguous relative to differential outcomes. Together, these results show that L1-regularization can improve classification performance while simultaneously providing highly specific and interpretable discriminative activation patterns.

show abstract

“…Such an approach, also known as sparse Total Variation (sparse TV) regularization (Baldassarre et al, 2012), TV-L 1 regularization (Gramfort et al, 2013) or fused LASSO (Tibshirani and Saunders, 2005), has previously been used in image processing (Ma et al, 2008) and fMRI prediction (Baldassarre et al, 2012;Gramfort et al, 2013), where it has been shown to lead to robust solutions, but is new in the field of brain source imaging. Note though that the combination of sparsity in the original source domain and in a transformed domain that is different from the total variation has been explored in (Chang et al, 2010) for MEG source imaging.…”

Section: Source Imaging Based On Structured Sparsity (Sissy) This Almentioning

confidence: 99%

SISSY: An efficient and automatic algorithm for the analysis of EEG sources based on structured sparsity

et al. 2017

View full text Add to dashboard Cite

Over the past decades, a multitude of different brain source imaging algorithms have been developed to identify the neural generators underlying the surface electroencephalography measurements. While most of these techniques focus on determining the source positions, only a small number of recently developed algorithms provides an indication of the spatial extent of the distributed sources. In a recent comparison of brain source imaging approaches, the VB-SCCD algorithm has been shown to be one of the most promising algorithms among these methods. However, this technique suffers from several problems: it leads to amplitude-biased source estimates, it has difficulties in separating close sources, and it has a high computational complexity due to its implementation using second order cone programming.To overcome these problems, we propose to include an additional regulariza-

show abstract

Identifying Predictive Regions from fMRI with TV-L1 Prior

Cited by 61 publications

References 9 publications

Convex relaxations of penalties for sparse correlated variables with bounded total variation

Convex relaxations of penalties for sparse correlated variables with bounded total variation

Sparse regularization techniques provide novel insights into outcome integration processes

SISSY: An efficient and automatic algorithm for the analysis of EEG sources based on structured sparsity

Contact Info

Product

Resources

About