Projects:GiniContrast
Back to NA-MIC Collaborations, MIT Algorithms,
Gini contrast for multivariate activation detection
In this work, we study the benefits of Random Forest classifiers and the associated Gini importance measure for selecting voxel subsets that form a multivariate neural response in a fMRI experiment. The Gini importance measure quantifies the predictive power of a particular feature when considered as part of the entire pattern. The measure is based on a random sampling of fMRI time points and voxels. As a consequence the resulting voxel score, or Gini contrast, is highly reproducible and reliably includes all informative features. The method does not rely on {\em a priori} assumptions about the signal distribution, a specific statistical or functional model or regularization. Instead it uses the predictive power of features to characterize their relevance for encoding task information. The Gini contrast offers an additional advantage of directly quantifying the task-relevant information in a multi-class setting, rather than reducing the problem to several binary classification sub-problems. In a multi-category visual fMRI study, the proposed method identified informative regions not detected by the univariate criteria, such as the t-test or the F-test. Including these additional regions in the feature set improves the accuracy of multi-category classification. Moreover, we demonstrate higher classification accuracy and stability of the detected spatial patterns across runs than the traditional methods such as the recursive feature elimination used in conjunction with Support Vector Machines.
Fig.1 Scheme of the algorithm
Experiments
In our experiments, we observed that Gini contrast as a voxel selection score identifies regions detected by univariate criteria and additional informative regions consistently missed by univariate criteria. Regions selected by the Gini contrast measure exhibit substantial overlap for different fMRI data trials for the same subject and across subjects.
Fig.2 Face category: Top row: top ranked voxels by univeriate t-test, Gini contrast, and exclusively Gini contrast. Second row: comparison between connected regions with more than 9 voxels detected by a univariate criterion ($t$-test, blue) and regions selected exclusively by Gini contrast (red). The additional regions detected by Gini contrast primarily contribute multivariate relationships to the category. For one pair of regions the de-trended BOLD values are illustrated. Together they hold significantly more information about the category than random regions. Two example regions carrying joint information are indicated by red and green curves. They exhibit characteristic joint behavior for faces: single mutual information vs. face images: I(face;red)=0.11, I(face;green)=0.054, pair-wise mutual information vs. face images: I(face;red,green)= 0.213. U - selected by univariate t-test; Gini only - selected only by Gini contrast.
Conclusion
We learn an atlas of the functional connectivity structure that emerges during a cognitive process from a group of individuals. The atlas is a group-wise generative model that describes the fMRI responses of all subjects in the embedding space. The embedding space is a low dimensional representation of fMRI time courses that encodes the functional connectivity patterns within each subject. Future work will focus on the application of the framework to the study of reorganization processes.
Key Investigators
- MIT: Georg Langs, Bjoern Menze, Danial Lashkari, Polina Golland
Publications
[1] G. Langs, B. Menze, D. Lashkari, and P. Golland. Detecting stable distributed patterns of brain activation using gini contrast. NeuroImage, 2010.
NA-MIC Publications Database on Atlas of Functional Connectivity