Randomized parcellation based inference

Abstract

Neuroimaging group analyses are used to relate inter-subject signal differences observed in brain imaging with behavioral or genetic variables and to assess risks factors of brain diseases. The lack of stability and of sensitivity of current voxel-based analysis schemes may however lead to non-reproducible results. We introduce a new approach to overcome the limitations of standard methods, in which active voxels are detected according to a consensus on several random parcellations of the brain images, while a permutation test controls the false positive risk. Both on synthetic and real data, this approach shows higher sensitivity, better accuracy and higher reproducibility than state-of-the-art methods. In a neuroimaging-genetic application, we find that it succeeds in detecting a significant association between a genetic variant next to the COMT gene and the BOLD signal in the left thalamus for a functional Magnetic Resonance Imaging contrast associated with incorrect responses of the subjects from a Stop Signal Task protocol.

Robust regression for large-scale neuroimaging studies

“…Many data processing and statistical analysis methods have been proposed in the literature to perform neuroimaging group analyses. These deal with the three main issues mentioned above: local averages within regions of interest (Flandin et al, 2002;Nieto-Castanon et al, 2003;Thirion et al, 2006) and feature selection (Hansen et al, 1999;Thirion and Faugeras, 2003;Spetsieris et al, 2009) are used to reduce the data dimension and the dependence between descriptors; prior smoothing of the images reduces registration mismatches (Worsley et al, 1996) and can be accounted for in standard multiple comparison corrections (Worsley et al, 1992); introducing noise regressors into the model aims at improving the sensitivity of the analyses (Lund et al, 2006); cluster-size analysis (Roland et al, 1993;Friston et al, 1993;Poline and Mazoyer, 1993), Threshold-Free Cluster Enhancement (TFCE) (Smith and Nichols, 2009;Salimi-Khorshidi et al, 2011) and Randomized Parcellation Based Inference (RPBI) (Da Mota et al, 2013) are state-of-the-art methods that combine several of the above-mentioned concepts to improve the statistical sensitivity of the analyses. For a more complete review, see Da Mota et al, 2013;Moorhead et al (2005); and Petersson et al (1999).…”

“…RPBI is a neuroimaging group analysis method recently proposed by Da Mota et al (Da Mota et al, 2013). It notably introduces spatial context in the fitting of the regression model.…”

Towards a faster randomized parcellation based inference

“…RPBI was proposed in [4] as an alternative to the ThresholdFree Cluster Enhancement (TFCE) [10] Formally, let C be the set of parcellations, and V be the set of voxels under consideration. Given a voxel v and a parcellation C, the parcel-based thresholding function θ t is defined as:…”

“…In this section, we investigate the computation time and recovery performance of various inference methods. We compare the execution time and recovery of i) voxel-level group analysis via ordinary least squares (OLS), which is the standard method in neuroimaging; ii) threshold-free cluster enhancement (TFCE) [10]; iii) RPBI with Ward [4]; iv) RPBI with ReNA, and v) ReNA aggregation. When clustering is applied, we set the number k of clusters to 5% of the number p of voxels 2 .…”

“…One proposed approach to overcome the lack of correspondence between individual images at the voxel level is Randomized Parcellation-Based Inference (RPBI) [4]. This algorithm has shown a good empirical performance, which may be linked to its data-driven parcellation step, and semi-parametric nature.…”