Contextual clustering for analysis of functional MRI data.

We present a contextual clustering procedure for statistical parametric maps (SPM) calculated from time varying three-dimensional images. The algorithm can be used for the detection of neural activations from functional magnetic resonance images (fMRI). An important characteristic of SPM is that the intensity distribution of background (nonactive area) is known whereas the distributions of activation areas are not. The developed contextual clustering algorithm divides an SPM into background and activation areas so that the probability of detecting false activations by chance is controlled, i.e., hypothesis testing is performed. Unlike the much used voxel-by-voxel testing, neighborhood information is utilized, an important difference. This is achieved by using a Markov random field prior and iterated conditional modes (ICM) algorithm. However, unlike in the conventional use of ICM algorithm, the classification is based only on the distribution of background. The results from our simulations and human fMRI experiments using visual stimulation demonstrate that a better sensitivity is achieved with a given specificity in comparison to the voxel-by-voxel thresholding technique. The algorithm is computationally efficient and can be used to detect and delineate objects from a noisy background in other applications.

Accuracy of a registration procedure for brain SPET and MRI: phantom and simulation studies.

Phantom experiments and simulations were performed to evaluate the significance of different error sources in a clinical registration procedure for brain SPET and MRI based on external markers. The results from the phantom experiments were used to adjust the error model for simulations. In the phantom experiments, 13-14 external markers were attached to the surface of a three-dimensional brain phantom for computing registration. Three internal test markers were used to estimate the accuracy of registration. The phantom was imaged with two different SPET and MRI devices. The mean root-mean-squared (RMS) residual of the locations of the test markers after registration using different combinations of four external markers varied from 3.5 +/- 1.0 to 5.2 +/- 1.3 mm depending on the imaging equipment and parameters used. The accuracy improved with an increasing number of external markers, from 3.2 +/- 0.5 to 4.9 +/- 0.5 mm for 6 markers and from 3.1 +/- 0.1 to 4.7 +/- 0.1 mm for 13 markers. In simulations, the external markers had an error comparable to the corresponding error in the phantom experiments. The error in the test markers was varied independently of that of the external markers. When the locating error of the test markers was removed, about 2 mm of the residuals of the test markers were found to come from this source. When an error comparable to the resolution of the original images (7-10 mm for SPET, 2 mm for MRI) was included in the test markers, the largest mean RMS residual after registration was smaller than the resolution error (8.8 +/- 1.1 mm). This was due to the accuracy of localization of the external markers and the fact that the direction of the error was random for each marker. The size of the registration error of an image volume was site-dependent, being minimal near the centre of mass of the external markers. When comparing the error with the spatial resolution of SPET, it was concluded that the accuracy of registration is not the limiting factor in region-of-interest analysis of registered images, provided that the design and attachment of the marker system are appropriate.

Reproducibility of fMRI: effect of the use of contextual information.

We studied the effect of use of contextual information on the reproducibility of the results in analysis of fMRI data. We used data from a repeated simple motor fMRI experiment. In the first approach, statistical parametric maps were computed from a spatially unsmoothed data and thresholded using a Bonferroni corrected threshold. In the second approach, the maps were computed from a spatially unsmoothed data but were segmented into nonactive and active regions using a spatial contextual clustering method. In the third approach, the statistical parametric maps were computed from spatially smoothed data and thresholded, using, optionally, a spatial extent threshold. The variation in the classification was largest in the Bonferroni thresholded statistical parametric maps. There were no significant differences in variation between statistical parametric maps generated with all the other methods. In addition to reproducibility, the detection rates of weak simulated activations in the presence of measured scanner and physiological noise were investigated. Contextual clustering method was the most sensitive method, while the least sensitive method was the Bonferroni corrected thresholding. Using simulated data, we demonstrated that the contextual clustering method preserves the shapes of activation regions better than the method using spatial smoothing of the data.