Outliers too far from the "trend"

Hello,

I am working through the AFNI tutorials for pre-processing and I have a question regarding the “auto block: outcount” procedure. It indicates that “at each voxel, the data is detrended using the default polynominal degree based on the duration of the first run. For that voxel time series, any value that is too far from the trend is considered an outlier.”

Does this mean that (1) the data takes into account scanner drift by using the default polynominal degree and then (2) outliers are removed based on whether they are too far from the trend? If so, what does “trend” refer to? Does it refer to the mean raw signal or the trend that was computed to de-trend the data?

Thank you for your help.

Best,
Tamara

Hi Tamara,

(1) Yes, scanner drift is evaluated at the same level as in the later linear regression. Perhaps the estimated trend is cubic, for example.

(2) Outliers are not removed, they are merely counted, as a fraction of the estimated brain, per time point. The resulting outlier time series is used for later censoring (if requested), or just to make quality control images.

  • rick

Hi Rick,

Thank you for your response. Just to clarify, what does “trend” refer to? Does it refer to the mean raw signal or the trend that was computed to de-trend the data?

Thank you for your help.

Tamara

It is not just a mean, but whatever polynomial trend can be
fit to that data. What -polort term is actually used in your script?

  • rick

Great, thank you for clarifying. I see in the resulting script from uber_subjects that my polort is “3”. So a cubic term is used.

Thank you again.
Tamara