Benchmark Data

This page will contain some 'benchmark' data to be used by everyone assessing the effectiveness of new methods for dealing with the motion problem. There are several datasets that are good candidates to be used as benchmarks. We will need to converge on one dataset, but for now they are listed below:

Jonathan Power's data

The data on development used in the Power et al 2012 paper can be found at [[1]]

Karen and Simon's data: drug users

The advantage of this dataset is that Simon has shown it to display a large motion-related artifact (by reproducing Power et al's cloud plots - see Simon's presentation in 'meeting 1').

Please email pv226@cam.ac.uk for access to these data (password protected access will be up soon).

The related scanner settings are available at: [[2]]

Notes by Simon:

• The naming convention is subject_id_wbic_id_image_session. Subject ids beginning 1 are controls, 2 are drug users, 3 are siblings, 4 are recreational users. e.g. 2001_16758_mprage.nii is the mprage of drug user number 1 and 3001_16759_rest.nii is the resting state epi session for 2001's sibling.

• There are 170 subjects in all. Not all drug users have a matching sibling and vice versa.

• Some subjects should be removed for coverage reasons as well as movement.

• Each directory also contains the DICOM header for each image.

• Some EPIs are unwrapped to put the back of the head that appears in front of the nose to the back of the image.

• The mprages are cropped to remove some neck. I find it makes life easier and saves disk space to remove space outside the brain.

• Note, the EPI slice code is 5 but the header says 0 (not known) because they were downloaded from the WBIC before Guy updated dcmconv to add that feature for Cinly.

• slicesdir contains index.html of with pictures of all mprage images next to the corresponding mean epi (inverted contrast).

• I believe the coregistrations could be improved. I used a target volume of 78 (79th volume) as it was the first common volume that was undamaged in all subjects. I think it would be better to coregister the mean EPI run image and reslice (if really necessary) all EPIs to the mean run EPI rather than a single volume as is done at present. The mean image should then be used to coregister to the MPRAGE (with 6 dof). (And the MPRAGE coregistered or warped to a T1 weighted MNI target). Using a mean image for coregistration takes out some of the randomness from a single image and better represents the session. If you have a lot of subjects it is difficult to find a single target volume number that is good for all. However, I did not do this as I stuck to the script but used target 78. But all that is just fiddling compared to the movement issue.

Prantik's datasets

The advantage of Prantik's data is that it corresponds to a single echo extracted from multi-echo data. Any new denoising techniques tested on this will thus be comparable to Prantik's multi-echo ICA denoising. Prantik has provided data for:

• 3 subjects at rest
• 5 subjects watching a movie
• 3 tremor patients

The decription can be found at [[3]]

The data can be downloaded from: http://wbic.cam.ac.uk/~pk355/me_sample_datasets.tar

Back To FMRI Page or Main Page

FMRI

Main Page