FIX Hand-Training Datasets

With major data and FIX hand-classification contributions from the Human Connectome Project and the Whitehall II imaging study.

Here we supply the data that was fed into FIX to train it - to create the trained-weights files that are supplied with FIX. These data are primarily the MELODIC ICA outputs (ICA component spatial maps, timecourses and temporal power spectra) that are fed into FIX, which it then classifies into "good vs bad" ICA components. We also supply the hand-classifications. The intention is primarily that researchers may want to look at these files in order to see in more detail how we did the hand classifications.

The new FSL viewer FSLeyes now contains a convenient mode for viewing ICA components and classifications, and hand-editing classifications.

Note that the final *.RData trained-weights files, which is all that you need to run FIX to classify new data, are not here - they are supplied with FIX.

Here is the list of downloads (for descriptions of the different sets of training datasets, see the main FIX web page)
HCP_hp2000 - subjects: 1-5, 6-10, 11-15, 16-20, 21-25 (4GB per file)
HCP_hp2000 (just one run)
WhII_MB6
WhII_Standard
Standard

The following individuals contributed to developing FIX and carrying out hand classifications: Gholamreza Salimi-Khorshidi, Ludovica Griffanti, Gwenaelle Douaud, David Flitney, Claire Sexton, Eniko Zsoldos, Klaus Ebmeier, Nicola Filippini, Clare Mackay, Stephen Smith (Oxford); Matthew Glasser, Donna Dierker, Erin Reid, David Van Essen (WashU); Edward Auerbach, Steen Moeller, Junqian Xu, Essa Yacoub, Kamil Ugurbil (Minnesota); Giuseppe Baselli (Milan); Christian Beckmann (Donders).

For general queries regarding FIX/FSL, please see the FSL Wiki, or email the FSL support list. Please send other feedback to steve@fmrib.ox.ac.uk.