We now have available a silent (more accurately quiet) EPI sequence, version r50
Although some improvements have been made, several image artifacts were encountered when testing the silent EPI sequence (but see updates below!). The choice of TE may bear some relevance to the presence of these artifacts - the quiet sequence achieves reduced acoustic noise by increasing the echo time in a manner that is suboptimal for fMRI acquisition.
Gary Chandler and John Deeks have measured noise levels for three different variants, varying in TR and TE:
1) TR=4s (TE=44; 32 slices; inter-slice gap of 52ms) - 24db quieter than the standard CBU_EPI); 2) TR=2.656s (TE=44; 32 slices; 21db quieter); 3) TR=2.2s (TE=38ms; 32 slices; 8db quieter).
As is apparent above, noise-reduction and TR is optimal for sequence 2.
UPDATE: After some phantom work, the following parameters of the sequence were changed and the images are now of much higher quality: shim mode = standard; Matrix Coil Mode = Triple; Series = descending; flip angle = 78 deg. (for 24 slices; 83 deg. for 32 slices).
Artifacts
* Dropout
Dropout and distortion scale with TE (best for the standard sequence with TE=30ms, worst for TE=44ms), so if acquisition noise is not at issue it would be best to use the standard sequence and avoid these artifacts. However, if noise reduction is required and ROIs only include peri-auditory areas (e.g. STG), then the quiet sequence (TE=44, TR=2.656) may be preferable (see Peelle et al., 2010, NeuroImage).
Given the susceptibility artefact, this does not seem to be a good sequence for studying regions that are prone to susceptibility artifact, e.g. inferior temporal and inferior frontal lobes. In addition, Matt pointed out that there are artifacts (e.g. ghosting) in other ventral structures, such as cerebellum and Rhodri noted image degradation in the Putamen and other subcortical structures.
UPDATE: with the updated sequence drop out seems to have decreased; to be compared to standard EPI
* Distortion, including squashing of the brain. Similar to old Bruker data. Field-maps could be used for undistortion. However, these need modification to ensure that they work correctly with this non-standard sequence.
Action point: Rhodri would liaise with Gayaneh, Matt, and Jonathan.
UPDATE: fieldmap undistortion did not make a big difference but with the updated sequence distortion has decreased considerably.
* Wrap around. In some cases the nose ended up at the back of the brain.
The wrap around artifact was the in prefrontal cortex. After these images were acquired Marta received a new sequence (version r50) from the developers, which seems to avoid the wrap around effect. However, problems with distortion/drop-out remain.
UPDATE: new version of the sequence (v50) does not seem to show any wrap around artifacts.
* Ghosting, in the form of streaks of image at the back of the head. Also, there is what looks like tiger stripes on the brain surface, around the edges.)
It could be helped by adopting a small field of view. Or, one could use a normal field of view and then remove the data after acquisition. Careful orientation of the slices might avoid the problem of ghosting. However, whether this is feasible depends on which region is of interest for the data acquisition.
UPDATE: with the updated sequence ghosting seems to have disappeared.
* High inter-subject variability in the data quality. In two different datasets, the subject acquired earlier in time had worse data than the later subject.
UPDATE: this has not been confirmed in later scans.
The inter-subject variability may be caused by the shimming of the scanner.
To address these questions about the data quality there will be some piloting of the silent sequence. A number of acquisitions are going to be acquired from subjects that are being run at CBU from week 04/10/2010. There will be two sessions of 20s each. Field-maps would be acquired for each participant.
The data would be pre-processed with AA to explore the data quality with regards to the artifacts. Undistortion could be applied to see whether this improves the data quality. The mean EPI image and tsdiffana can be used to explore data quality. Lower slices are expected to have higher inter-slice variability. The piloting could also help to see whether the inter-subject variability is affected by the shimming of the scanner, by looking at whether noise/artifacts change over the acquisition time across subjects.
UPDATE: piloting has been completed, issues with artifacts have been resolved and functional scans are now being acquired.
How about BOLD sensitivity? It is not clear whether these problems affect the sequence’s BOLD sensitivity. But, see paper by Peelle et al 2010 http://dx.doi.org/10.1016/j.neuroimage.2010.05.015
UPDATE: the quiet sequence seems to produce reliable and strong BOLD signal with signal being as high as or higher than for the standard sequence in auditory regions (sound vs. silence contrast).
Please improve these notes and add any other issues that you have encountered.