3. Goals of quality control
Deciding which data to include in your study and which to
reject.
Deciding on using a public dataset (is it appropriate for my
design/study?)
Diagnosing fixable problems with data acquisition process:
Types of sequences
Scanner malfunctions
Head padding
Participant instructions
4. When to perform quality control?
Early! As soon as you get data:
Helps fix problems with the scanner before the next subject
Allows to recruit extra subjects if you know some data needs
to be discarded
QC (when done with the right tools) takes very
little effort - but can save a lot of money and time
in the long run!
5. Basics: data consistency
Check if:
Scans for a new subjects have the same (prescribed) parameters:
Resolution
Field of view
Number of timepoints (fMRI)
Each subject has all of the scans
21. Motion and spin history effects
http://www.jonathanpower.net/2016-ni-the-plot.html
22. Motion and spin history effects
http://www.pnas.org/content/111/16/6058.full.pdf
23. QC metrics
Noise measurement
Signal-to-noise ratio (SNR) - higher is better
Contrast-to-noise ration (CNR) - higher is better
Sharpness (full-width half maximum estimations) - smaller FWHM is better
Goodness of fit of a noise model into the noise in the background (QI2) - lower is better
Coefficient of Joint Variation (CJV) - lower is better
Information theory
Foreground-Background Energy Ratio (FBER) - higher is better
Entropy Focus Criterion (EFC) - lower is better
Artifacts
Segmentation using mathematical morphology (QI1) - lower is better
24. QC metrics
Noise measurement: SNR, tSNR, temporal standard deviation
Information theory: EFC, FBER
Confounds and artifacts:
Framewise Displacement (FD) - lower is better
(Standardized) DVARS (D referring to temporal derivative of timecourses, VARS referring to RMS
variance over voxels) - lower is better
Ghost-to-Signal ratio (GSR) - lower is better
Global correlation (GCOR) - lower is better
Energy of spectrum (ES) - lower is better
AFNI’s outlier detection and quality indexes
25. More thoughts about QC
There are not strict rules which data to discard
Some artefacts and distortions can be recovered by smart
algorithms
QC can help you decide results from which subjects you should
interrogate more closely