Analytic estimation of statistical significance maps for support vector machine based multi-variate image analysis and classification

By Bilwaj Gaonkar, Christos Davatzikos. 2013

123 citations recorded by the source corpus.

Multivariate pattern analysis (MVPA) methods such as support vector machines (SVMs) have been increasingly applied to fMRI and sMRI analyses, enabling the detection of distinctive imaging patterns. However, identifying brain regions that significantly contribute to the classification/group separation requires computationally expensive permutation testing. In this paper we show that the results of SVM-permutation testing can be analytically approximated. This approximation leads to more than a thousand fold speed up of the permutation testing procedure, thereby rendering it feasible to perform such tests on standard computers. The speed up achieved makes SVM based group difference analysis competitive with standard univariate group difference analysis methods.

Subfields: Computer Vision and Pattern Recognition, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Molecular Biology

Citations in the corpus

Visual Feature Attribution Using Wasserstein GANs (2017, methodology): “The most common approach is to use regression analysis or machine learning tools to generate population average maps, which highlights features that are salient across the population [3, 32, 61, 16, 39, 45, 14].”
Interpreting support vector machine models for multivariate group wise analysis in neuroimaging (2015, positive): “They are also quite sensitive to regularization Gaonkar and Davatzikos (2013).”
Visualizing Deep Neural Network Decisions: Prediction Difference Analysis (2017, background): “These are independent of the input image, and, as argued by Gaonkar & Davatzikos (2013) and Haufe et al. (2014), interpreting these weights can be misleading in general.”
Speeding up Permutation Testing in Neuroimaging (2013, background): “This step alone can run from a few days up to many weeks and even longer [14, 15].”
Constructing Subject- and Disease-Specific Effect Maps: Application to Neurodegenerative Diseases (2016, methodology): “Predictive modeling approaches use machine-learning tools to identify measurements that are highly informative for diagnosing the condition [9,10,11,12].”
Metabolomics and Multi-Omics Integration: A Survey of Computational Methods and Resources (2020, background): “Another approach to decipher the analytes most relevant to the model is to consider both the weight and margin between the hyperplane [175].”
Single Subject Prediction of Brain Disorders in Neuroimaging: Promises and Pitfalls (2016, methodology): “There are number of recent studies looking at combined biomarkers of sMRI, FDG-PET, and CSF (mostly for ADNI dataset) to discriminate between AD,MCI and HC (Gray et al., 2013; Xu et al., 2015; Yu et al., 2015; Zhang et al., 2011, 2014).”
Reconstructing subject‐specific effect maps (2017, unknown): “Both of these were then converted to p-value maps using permutation testing (Gaonkar and Davatzikos, 2013; Good, 2005).”
SubCMap: Subject and Condition Specific Effect Maps (2017, methodology): “The RF importance values and SVM weights were then converted to p-values with permutation testing similar to other works in the literature [18, 38].”