A Data-Driven Multi-scale Technique for fMRI Mapping of the Human Somatosensory Cortex.

A previously introduced Bayesian non-parametric multi-scale technique, called iterated Multigrid Priors (iMGP) method, is used to map the topographic organization of human primary somatosensory cortex (S1). We analyze high spatial resolution fMRI data acquired at ultra-high field (UHF, 7T) in individual subjects during vibrotactile stimulation applied to each distal phalange of the left hand digits using both a travelling-wave (TW) and event-related (ER) paradigm design. We compare the somatotopic digit representations generated in S1 using the iMGP method with those obtained using established fMRI paradigms and analysis techniques: Fourier-based analysis of travelling-wave data and General Linear Model (GLM) analysis of event-related data. Maps derived with the iMGP method are similar to those derived with the standard analysis, but in contrast to the Fourier-based analysis, the iMGP method reveals overlap of activity from adjacent digit representations in S1. These findings validate the use of the iMGP method as an alternative to study digit representations in S1, particularly with the TW design as an attractive means to study cortical reorganization in patient populations such dystonia and carpal tunnel syndrome, where the degree of spatial overlap of cortical finger representations is of interest.

Individual trial analysis for 7T fMRI data by a data-driven multi scale approach.

An important interest in event-related single trial fMRI is the possibility of studying cognitive processes that vary in time (e.g. learning or adaptation). Region-specific modelling and the inter-trial variability of the evoked response play an important role. We showed how the use of the iterated multigrid priors (iMGP) method, a previously introduced data-driven multi scale Bayesian iterative approach, may be extended for a trial-by-trial analysis on ultra-high magnetic field data. We used both artificial (present real physiological noise) and real (unilateral finger tapping experiment) data at 7T and compared to other methods. Since the iMGP does not need to spatially smooth the data, avoiding a loss of sensitivity, we take advantage of the high SNR available at 7T. For artificial data, we showed receiver operating characteristic curves parametrized by the activity threshold and by the addition of extra thermal noise and compared with correlation technique results.The method showed be very robust in terms of specificity for very noisy data and capable of capturing the temporal variability imposed artificially across regions. For real data, we examined the inter-trial spatial relationships for four subjects and the time-to-peak of the evoked response estimated by the iMGP across trials, regions and subjects. To stress the reliability of the iMGP in single trial studies, an illustrative comparison with the variational Bayes approach (implemented in the very popular Statistical Parametric Mapping software) was done for a single subject. Despite the extravascular signals are still present at 7T and the confounds of physiological noise and hemodynamic variability affecting single trial approaches, we showed that with the iMGP method it is possible to detect individual HR robustly.

BOLD response analysis by iterated local multigrid priors.

We present a non parametric Bayesian multiscale method to characterize the Hemodynamic Response HR as function of time. This is done by extending and adapting the Multigrid Priors (MGP) method proposed in (S.D.R. Amaral, S.R. Rabbani, N. Caticha, Multigrid prior for a Bayesian approach to fMRI, NeuroImage 23 (2004) 654-662; N. Caticha, S.D.R. Amaral, S.R. Rabbani, Multigrid Priors for fMRI time series analysis, AIP Conf. Proc. 735 (2004) 27-34). We choose an initial HR model and apply the MGP method to assign a posterior probability of activity for every pixel. This can be used to construct the map of activity. But it can also be used to construct the posterior averaged time series activity for different regions. This permits defining a new model which is only data-dependent. Now in turn it can be used as the model behind a new application of the MGP method to obtain another posterior probability of activity. The method converges in just a few iterations and is quite independent of the original HR model, as long as it contains some information of the activity/rest state of the patient. We apply this method of HR inference both to simulated and real data of blocks and event-related experiments. Receiver operating characteristic (ROC) curves are used to measure the number of errors with respect to a few hyperparameters. We also study the deterioration of the results for real data, under information loss. This is done by decreasing the signal to noise ratio and also by decreasing the number of images available for analysis and compare the robustness to other methods.

Multigrid priors for a Bayesian approach to fMRI.

We introduce multigrid priors to construct a Bayesian-inspired method to asses brain activity in functional magnetic resonance imaging (fMRI). A sequence of different scale grids is constructed over the image. Starting from the finest scale, coarse grain data variables are sequentially defined for each scale. Then we move back to finer scales, determining for each coarse scale a set of posterior probabilities. The posterior on a coarse scale is used as the prior for activity at the next finer scale. To test the method, we use a linear model with a given hemodynamic response function to construct the likelihood. We apply the method both to real and simulated data of a boxcar experiment. To measure the number of errors, we impose a decision to determine activity by setting a threshold on the posterior. Receiver operating characteristic (ROC) curves are used to study the dependence on threshold and on a few hyperparameters in the relation between specificity and sensitivity. We also study the deterioration of the results for real data, under information loss. This is done by decreasing the number of images in each period and also by decreasing the signal to noise ratio and compare the robustness to other methods.