https://www.na-mic.org/w/index.php?action=history&feed=atom&title=FBIRN%3ANeuroInfStatsTuesAM2006FBIRN:NeuroInfStatsTuesAM2006 - Revision history2026-04-17T12:12:05ZRevision history for this page on the wikiMediaWiki 1.33.0https://www.na-mic.org/w/index.php?title=FBIRN:NeuroInfStatsTuesAM2006&diff=3626&oldid=prevAndy: Update from Wiki2006-12-18T13:26:28Z<p>Update from Wiki</p>
<p><b>New page</b></p><div>= Data Mining =<br />
<br />
* History of Data Mining<br />
** 1960's Tukey<br />
** Predictive Modeling (CART)<br />
** Machine Learning (CS)<br />
** Relational Databases<br />
** Web<br />
** Hardware<br />
* Convergence of above technologies has lead to data mining<br />
** There is a lot of hype regarding data mining<br />
* FBIRN has a small N in the data mining sense<br />
** Clustering, look for subgroups in the schizophrenia patients<br />
** fBIRN data has a rich structure<br />
** Focused on data analysis right now, do algorithm comparisons<br />
** Make sure that the raw spatial data, timecourse data is available<br />
* Standardized data for data analyses<br />
** MRI Machine --> "raw data" --> intermediate data --> Beta-maps, etc<br />
*** inter-subject issues in preprocessing<br />
* data analysis recommendations<br />
** give someting standard and consistent across sites<br />
** should data vetting be performed, runs or subjects tossed based on results of first level analysis<br />
** time series of variance, auto-correlation<br />
** having standard datasets available so other disciplines (e.g., computer scientists, statisticians) can look at the data<br />
** proposal, FIPS 1.0 produces standard results (beta-maps) on phaseII, then spend 6 months improving the pipeline<br />
** make data available after each major preprocessing step<br />
*** after pre-whitening/smoothing<br />
*** after motion correction (would need to modify FIPS/FEAT)<br />
*** after slice timing correction (would need to modify FIPS/FEAT)<br />
*** currently FIPS final products create the output of FEAT<br />
** phaseII<br />
*** uncompressed raw data is just over 1Gb/subject visit<br />
*** intermediate time series 20Gb/subject visit<br />
*** may be problematic to store all of the intermediate steps, due to disk space and downloading times<br />
* what can we do for every subject?<br />
* what can we do on a subset of subjects?<br />
* FIAC<br />
** standard data products<br />
** 1st level<br />
*** raw<br />
*** motion corrected<br />
**** motion parameters<br />
*** slice-time corrected<br />
*** meta-data, details of the paradigm<br />
**** best practices, what we think the design matrix should be<br />
**** what happened at what time, how long the blocks were, etc.<br />
** 2nd level<br />
*** constrast copes in standard space<br />
*** varcopes in standard space<br />
*** meta-data,<br />
**** what effect<br />
**** degrees of freedom<br />
**** smoothness<br />
**** T image<br />
**** threshold<br />
* Standard Data Products Plan<br />
** Establish a working group<br />
** Establish the requirements of what is released besides raw data and when<br />
** QA reports on raw and on intermediate data products<br />
* Data QA<br />
** Variance of image over time series<br />
** Variacne oc (Xt-Xt+1) over time series<br />
*** scale median to 100<br />
** Good for residuals<br />
*** (Outlier count per image / Expected outlier count ) * 100 for each image yields a time series<br />
*** Normality test at each voxel, look at how many p values smaller than 0.05<br />
**** (Number of significant voxels / Expected significant voxels) * 100</div>Andy