The Statistical Analysis of Multi-Voxel Patterns in Functional Imaging (original) (raw)
Related papers
A critique of multi-voxel pattern analysis
2010
Abstract Multi-voxel pattern analysis (MVPA) is a popular analytical technique in neuroscience that involves identifying patterns in fMRI BOLD signal data that are predictive of task conditions. But the technique is also frequently used to make inferences about the regions of the brain that are most important to the tasks in question, and our analysis shows that this is a mistake. MVPA does not provide a reliable guide to what information is being used by the brain during cognitive tasks, nor where that information is.
NeuroImage, 2013
An ever-increasing number of functional magnetic resonance imaging (fMRI) studies are now using informationbased multi-voxel pattern analysis (MVPA) techniques to decode mental states. In doing so, they achieve a significantly greater sensitivity compared to when they use univariate frameworks. However, the new brain-decoding methods have also posed new challenges for analysis and statistical inference on the group level. We discuss why the usual procedure of performing t-tests on accuracy maps across subjects in order to produce a group statistic is inappropriate. We propose a solution to this problem for local MVPA approaches, which achieves higher sensitivity than other procedures. Our method uses random permutation tests on the single-subject level, and then combines the results on the group level with a bootstrap method. To preserve the spatial dependency induced by local MVPA methods, we generate a random permutation set and keep it fixed across all locations. This enables us to later apply a cluster size control for the multiple testing problem. More specifically, we explicitly compute the distribution of cluster sizes and use this to determine the p-values for each cluster. Using a volumetric searchlight decoding procedure, we demonstrate the validity and sensitivity of our approach using both simulated and real fMRI data sets. In comparison to the standard t-test procedure implemented in SPM8, our results showed a higher sensitivity. We discuss the theoretical applicability and the practical advantages of our approach, and outline its generalization to other local MVPA methods, such as surface decoding techniques.
Magnetic Resonance Imaging, 1998
The general aims of functional brain magnetic resonance imaging (fMRI) studies are to ascertain which areas of the brain are activated during a specific task, the extent of this activation, whether different groups of subjects demonstrate different patterns of activation, and how these groups behave in different tasks. Many steps are involved in answering such questions and if each step is not carefully controlled the results may be influenced. This work has three objectives. Firstly, to present a technique for quantitatively evaluating methods used in functional imaging data analysis. While receiver-operator-characteristic (ROC) analysis has been used effectively to evaluate the ability of post-processing algorithms to detect true activations while rejecting false activations, it is difficult to adapt such a technique for comparisons of methods for quantitating activations. We present a technique based on the ANOVA, between two or more regions of interest (ROIs), subject groups, or activation tasks, over a range of statistical thresholds, which reveals the sensitivity of different activation quantification metrics to noise and other variables. Secondly, we use this technique to compare two methods of quantifying localized brain activation. There are numerous ways of quantifying the amount of activation present in a specific region of the brain in an individual subject. We compare the pixel count approach, which simply counts the number of pixels above an arbitrary statistical threshold, with an approach based on the sum of t-values above the same arbitrary t-value threshold. Finally, we examine the sensitivity of the results from an analysis of variance, to user defined parameters such as threshold and region of interest size. Both simulated and real functional magnetic resonance data are used to demonstrate these techniques.
NeuroImage, 2014
Multi-voxel pattern analysis (MVPA) has led to major changes in how fMRI data are analyzed and interpreted. Many studies now report both MVPA results and results from standard univariate voxel-wise analysis, often with the goal of drawing different conclusions from each. Because MVPA results can be sensitive to latent multidimensional representations and processes whereas univariate voxel-wise analysis cannot, one conclusion that is often drawn when MVPA and univariate results differ is that the activation patterns underlying MVPA results contain a multidimensional code. In the current study, we conducted simulations to formally test this assumption. Our findings reveal that MVPA tests are sensitive to the magnitude of voxel-level variability in the effect of a condition within subjects, even when the same linear relationship is coded in all voxels. We also find that MVPA is insensitive to subject-level variability in mean activation across an ROI, which is the primary variance component of interest in many standard univariate tests. Together, these results illustrate that differences between MVPA and univariate tests do not afford conclusions about the nature or dimensionality of the neural code. Instead, targeted tests of the informational content and/or dimensionality of activation patterns are critical for drawing strong conclusions about the representational codes that are indicated by significant MVPA results.
Journal of Magnetic Resonance Imaging, 1999
In 23 fMRI studies on six subjects, we examined activation in visual and motor tasks. We modeled the expected activation time course by convolving a temporal description of the behavioral task with an empirically determined impulse response function. We evaluated the signal activation intensity as both the number of activated voxels over arbitrary correlation thresholds and as the slope of the regression line between our modeled time course and the actual data. Whereas the voxel counting was strikingly unstable (standard deviation 74% in visual trials at a correlation of 0.5), the slope was relatively constant across trials and subjects (standard deviation F14%). Using Monte Carlo methods, we determined that the measured slope was largely independent of the contrast-to-noise ratio. Voxel counting is a poor proxy for activation intensity, with greatly increased scatter, much reduced statistical power, and increased type II error. The data support an alternative approach to functional magnetic resonance imaging (fMRI) that allows for quantitative comparisons of fMRI response magnitudes across trials and laboratories.
Comparison of different statistical analyses in visual stimulus fMRI
Journal of neuroradiology. Journal de neuroradiologie, 2006
The present study evaluated four different clinically relevant statistical approaches with respect to a response to a visual stimulus paradigm. Healthy volunteers were subjected to a visual stimulus consisting of a checkerboard black-and-white box car pattern with on-off blocks of 10s. Simultaneously, sensitivity encoding (SENSE) dynamic MR imaging was acquired using a 1.5 T MR system. Statistical analyses were conducted with z-cluster analysis, Student's t-test, Spearman's correlation, and time-series normalized cross-correlation. A figure-of-merit for neural activity was measured from calculated maps using pixel counting. The results demonstrated that the index of activity estimated from the number of "activated" pixels did not differ markedly among the four different statistical methods, except when comparing the cross-correlation statistics with z-clustering in the whole brain, implying that all methods lead to similar statistical information when using fMRI to...
Multivariate fMRI Analysis Using Optimally-discriminative Voxel-based Analysis
2012 Second International Workshop on Pattern Recognition in NeuroImaging, 2012
This significantly extends Multi-Voxel Pattern Analysis (MVPA) methods, such as the Searchlight and related methods, by building on an approach that was recently proposed for structural brain images, and was named Optimally-Discriminative Voxel-Based Analysis (ODVBA), which uses machine learning models to determine the optimal anisotropic filtering of images that enhances group differences. Precise spatial maps of activation are computed by tallying the weights of each voxel to all of the neighborhood in which it belongs, and significance maps are obtained via permutation testing. We adapt this idea to both single and multi-subject fMRI analysis. Both simulated data and real data from 12 adolescent subjects who completed a standard working memory task demonstrated the use of ODVBA in fMRI improves accuracy and spatial specificity of activation detection over Searchlight.
All that glitters is not BOLD: inconsistencies in functional MRI
Scientific Reports, 2014
The blood oxygenation level dependent (BOLD) functional magnetic resonance imaging (fMRI) signal is a widely-accepted marker of brain activity. The acquisition parameters (APs) of fMRI aim at maximizing the signals related to neuronal activity while minimizing unrelated signal fluctuations. Currently, a diverse set of APs is used to acquire BOLD fMRI data. Here we demonstrate that some fMRI responses are alarmingly inconsistent across APs, ranging from positive to negative, or disappearing entirely, under identical stimulus conditions. These discrepancies, resulting from non-BOLD effects masquerading as BOLD signals, have remained largely unnoticed because studies rarely employ more than one set of APs. We identified and characterized non-BOLD responses in several brain areas, including posterior cingulate cortex and precuneus, as well as AP-dependence of both the signal time courses and of seed-based functional networks, noticing that AP manipulation can inform about the origin of the measured signals.
Enhanced fMRI Response Detection and Reduced Latency through Spatial Analysis of BOLD Signals
2008
In conventional functional magnetic resonance imaging (fMRI) analysis, activation is often inferred by examining only the intensity modulation of blood-oxygen-level dependent (BOLD) signal of each voxel in isolation or in small, local clusters. However, as has been recently demonstrated, activation can in fact be detected by examining the spatial modulation of the BOLD distribution within a region of interest (ROI). In this paper, we propose and demonstrate with real fMRI data that analyzing such spatial changes can enhance the effect size of fMRI response detection over using intensity information alone. Furthermore, we show that such spatial changes consistently and significantly antecede mean intensity changes in multiple ROIs. We hence foresee spatial analysis of BOLD distribution to be a promising direction to explore in complementing pure intensity-based approaches.