Approximate time needed to complete the tutorial: 15 min.
TUTORS: To test your cleaning function please see this page
Pre-processing is a fundamental steps in neurophysiological data analysis. It allows to clean the data from contaminating artifacts of several nature before performing your analysis.
From the previous pages you have learned how to classify an artifact both in single channels and in IC components. Still, to be able to detect all types of artifacts requires a lot of training and experience.
For this reason we developed a validation test that will help to diagnose your cleaning with respect to an automatic procedure. This test will not only provide an answer to the quality of the cleaning procedure you followed, but will also help to clarify possible doubts and issues you encountered in the entire process.
So, after saving your data as indicated previously, you are strongly invited to contact a tutor who will come at your desk to run the validation test which includes a) test on the cleaned signal lengths (indicative of the noisy interval you have discarded), b) test on the bad channels, c) test on the biomarkers.
Hereafter an example of the outcome of the validation test is reported:
Check on signal length:
--- Original duration 300 seconds (5 minutes).
--- Automatic procedure: final duration = 300 seconds (100% of the original signal length)
--- Student procedure: final duration = 295.9 seconds (98.63% of the original signal length)
Check on removed channels:
--- Eyes Channels: [ 8 14 21 25 125 126 127 128 ]
--- Automatic procedure: Total Bad Channels 3
: Bad Channels picked by faster [ 25 44 99 ]
--- Student procedure: Total Bad Channels (including eye channels) = 9
: Eye Channels picked by the student [ 25 ]
: Bad Channels picked by the student and by faster [ 25 44 99 ]
: Bad Channels picked by the student not by faster [ 22 46 56 98 114 122 ]
Check on absolute amplitude values ...
--- Percent Error plot: (Student-Automatic)/Automatic*100
You will discuss more in details, together with the tutor, the meaning of the outcome of the validation test and the percent error on the biomarker computation, shown in the figure above.
Automatic methods for EEG pre-processing are becoming more and more popular because, thanks to sophisticated algorithms, they allow to reduce man-time during data analysis, especially when large database need to be cleaned and screened. At the same time, you should be aware that these methods are not yet perfect and they cannot substitute the analytic sight of a human eye. During the validation test, it could happen, indeed, that your cleaning procedure results even more detailed than the one performed by the automatic algorithm.
Artifacts can be categorized in
Stereotyped artifacts, such as Eye movements (EOG = O(10² μV); EEG = O(10 μV)), Eye Blinks
Non-stereotyped artifacts, including Movements of the electrodes on the scalp, Current drift, Spurious electrical activity picked by the EEG amplifier, Muscle movements (EMG = O(10² μV); EEG = O(10 μV)))
Pre-processing consists into different filtering processes, in space and time, that can be synthesized into three main steps :
Bad channels detection
Transient artifacts removal
Independent components rejection, when Independent Component Analysis is used for detect, separate and remove activity in EEG records from artifacts.
When you have discussed the cleaning result with a tutor, please save the figure as a pdf file and e-mail it to Klaus. Name the files “Group xx. Student yyyyyy. pdf”
tutorial/test_your_cleaning.txt · Last modified: 2016/11/21 22:16 by Simon Houtman