Real Activation Study

As we have described above, it is difficult to make accurate measurements pertaining to the accuracy of a motion correction scheme when presented with real data. With data exhibiting activation, examination of the time series after correction using an animation tool reveals no visible extreme affine movement although some motion artefacts remain. In particular, we are able to show good localisation of activations which would not be possible without motion correction first being carried out. The thresholded statistical maps shown in Figure 17 correspond to a 180 volume audio-visual experiment. Analysis was carried out using FEAT, FMRIBs Easy Analysis Tool using an Improved Linear Model [20]. In order to test the effectiveness of MCFLIRT on real data, the subject was asked to move his head during the experiment.

Figure 17: The resulting thresholded z-statistics from a FEAT analysis of real audiovisual data (Auditory 45s OFF, 45s ON and visual 30s OFF, 30s ON) which has been motion corrected. Prior to motion correction, the series exhibited a large degree of motion as the subject had been requested to move as much as possible during the experiment. Maps are: (from top) Visual paradigm z-stats (uncorrected), auditory z-stats (uncorrected), visual z-stats (MCFLIRT), auditory z-stats (MCFLIRT)

It can be seen, by comparing the activations from the uncorrected and corrected data sets, that in the uncorrected set a large number of false positives exist in both visual and auditory results. While it can be argued that both data sets are highly corrupted by this large motion (indeed even the corrected data set still exhibits some visible movement, albeit at a significantly smaller scale than the uncorrected data), the MCFLIRT-corrected visual stimulus is well localised and allows an otherwise corrupted set of experimental data to yield potentially useful results.