Homepage of Matthijs Vink

Functional MRI typically aims to detect changes in neuronal processing using BOLD. However, any process that affects the oxygen-level in the blood, or for example the level of blood-flow will also affect the fMRI-BOLD signal. Experiments testing for differences between patients and controls, investigating the effects of medication (pharmacological MRI), or even resting-state fMRI studies may therefore be confounded by cardiorespiratory (CR) influences. It is estimated that physiological noise makes up about 35 percent of the signal in BOLD-fMRI. Controlling for this noise therefore has the potential to dramatically increase the reliability of fMRI and hence reduce the number of subjects needed for both test-retest and genetic association studies. We have developed a matlab-tool, AZTEC, that removes both cyclic variance and non-cyclic variance due to CR processes [ download ].  

We have already used AZTEC for cleaning up the fMRI time series data for resting-state Default-Mode Network analyses [ download paper ]

Introduction

High reliability is crucial for the validity of any method. Unfortunately, fMRI does not seem to be very reliable on an individual level. This is problematic as fMRI results are now starting to be used as biomarkers in genetic studies or to test effects of treatment (TMS, medication).

In our first fMRI study on reliability, the test-retest reliability of activation maps during the antisaccade paradigm was assessed for individual and group results. Functional MRI images were acquired during two sessions of prosaccades and antisaccades in twelve healthy subjects using an event-related fMRI design. Reliability was assessed for both individual and group-wise results. In addition, the reliability of differences between subjects was established in predefined regions of interest. The reliability of group activation maps was high for prosaccades and antisaccades, but only moderate for antisaccades vs. prosaccades, probably as a result of low statistical power of individual results. Reproducibility of individual subject maps was highly variable, indicating that reliable results can be obtained in some but not all subjects. Reliability of individual activity maps was largely explained by individual differences in the global temporal signal to noise ratio (SNR). As the global SNR was stable over sessions, it explained a large portion of the differences between subjects in regional brain activation. A low SNR in some subjects may be dealt with either by improving the statistical sensitivity of the fMRI procedure or by subject exclusion. Differences in the global SNR between subjects should be addressed before using regional brain activation as phenotype in genetic studies.

Test-retest reliability of fMRI activation during prosaccades and antisaccades. Raemaekers M, Vink M, Zandbelt BB, van Wezel RJ, Kahn RS, & Ramsey NF.  Neuroimage 2007, Jul 1;36(3):532-42 [ download paper ]

Reliability metrics: ICC and sigma(w) 

ICC or intraclass correlation coefficient refers to the degree of similarity between sessions given the formula:

            MSbetween - MSwithin

ICC =  ------------------------------

            MSbetween + MSwithin

where MSbetween and MSwithin represent the mean squared error for between and within-subjects variance. As such, a high ICC is achieved when the within subject variance is low. However, a high ICC can also be achieved when the between subject variance is high. In other words; when using ICC, the estimate of reliability depends upon group characteristics. This is not desirable, as we want a measure of replicability for a single subject, independent of group. For this reason, we use a different metric for reliability, namely sigma(w), or within subject standard deviation, which does not depend on the between-subjects variance in the sampled group. Sigma(w) for an individual is given by:

                                           (activation session1 - overall mean)^2 + (activation session2 - overall mean)^2

sigma(w) indiv = sqrt    -----------------------------------------------------------------------------------------------------------

                                                                                     number of sessions

where activation can be either the value of a single voxel or, for example, the activation within a Region of Interest. The individual sigma(w)s can be averaged over subjects to give an impression of reliability for a specific group. In the paper, we describe results from ten healthy subjects who performed a motor task on three occasions, separated by one week, while being scanned. In order to exclude training effects on fMRI stability, all subjects were trained extensively on the task. Task performance, spatial activation pattern, and group-wise BOLD signal changes were highly stable over sessions. In contrast, we found substantial fluctuations (up to half the size of the group mean activation level) in individual activation levels, both in ROIs and in voxels. Given this large degree of instability over sessions, and the fact that the amount of within-subject variation plays a crucial role in determining the success of an fMRI study with repeated measurements, improving stability is essential. In order to guide future studies, sample sizes are provided for a range of experimental effects and levels of stability. Obtaining estimates of these latter two variables is essential for selecting an appropriate number of subjects.

Within-subject variation in BOLD-fMRI signal changes across repeated measurements: Quatification and implications for sample size. BB Zandbelt, TE Gladwin, M Raemaekers, M van Buuren, SF Neggers, RS Kahn, NF Ramsey & M Vink. Neuroimage 2008, August 1;42(1):196-206  [ download paper ]