[Homer-users] Reply: questions about application of CW5

[thuppert]

This is in reply to a couple of questions that were asked with respect to
some of the data processing techiniques (etc) within HomER.

A couple of points first:

1)  Anyone on the homer-users list server can post questions or post
answers.  We would really like to encourage this to be a forum for the
community to discuss these types of issues.  A record of all these emails is
kept online (under view archives at:
http://www.nmr.mgh.harvard.edu/DOT/resources/homer.htm)




**********************************************************
1.The first is about motion artifact. In initial review, one of the
advantages of NIRS is insensitive to head motion. However, in our
experiments, we noticed NIRS was in fact sensitive to head motion. Some
researchers, who are the users of  ETG-100 (Hitachi) or similar system, used
certain method to correct short-term motion in their articles. Then how do
you treat with and solve the artifact problem? Is it appropriate to take the
same measure?


	Honestly, I'm not familar with what the Hitachi user's use to reduce motion
artifacts.  Here are a few of our suggestions:

         Physical solutions:  The best thing to do is to try to reduce the
motion artifacts as much as possible when you collect the data (a bit
obvious).  We find that aside from a fiber optic holder that is stable (i.e.
can be secured to the head tightly), the probe should be flexible and most
importantly comfortable for the subject.  Rigid probes tend to torque or
twist, which makes maintaining good fiber contact with the head very
difficult.  Also if the probe hurts, the person moves more!  We usually put
dark foam padding between the source-detector fibers to block direct light
as much as possible and secure the probe with lots of velcro.  A good idea
is to secure the fibers to a point on the head, that way, if the person
moves their head, there isn't any pulling on the probe and movement of the
fibers.

	   Data processing solutions:  Of course it won't be possible to always
remove motion.  One method we use in data processing (which is written into
HomER) is prinicple component filtering (sometimes called truncated SVD
filtering). This is described in the new User's guide, which will be posted
in the next few days.



**********************************************************
   2.The second is about the data processing. Because the baseline is
sometimes unstable, some researchers used some methods to correct the
baseline,e.g. linear fitting. Frankly,I don't completely understand  the
method. However, I think it's an important question in data processing. So I
want to know how you treat with this problem.


   There are a both low and high pass filters in HomER to correct for drifts
in baseline (and remove physiology- i.e. heart beat signals).  HomER also
has prinicple component filters to remove baseline physiology.  The
eigen-vectors from a purely baseline (rest) data file are calculated and
then projected from the functional data.  The newest version of Homer (to be
released this week) also allows linear regression against additionally
recorded physiology (i.e. ECG signals or pulse oximetry recordings).  Linear
modeling in HomER includes a DC (constant) term as well.



**********************************************************
  3.As you know, it isn't easy to collect perfect data because of poor
contaction. Usually, some channels have a lower SNR (signal-to-noise ratio).
So they have to be excluded to further analysis. It's pity! Are there
methods to improve the quality of signal by data processing. And what's the
criterion of low SNR, i.e. is there quantitative criterion to exclude the
low SNR channel?


	There is not much you can do with poor SNR data.  The TechEN instruments
use a demodulation  processing step to seperate source-detectors.  Each
laser is amplutude modulated with its own carrier frequency.  It might be
possible to rescue some poor data in this step by chosing the right low-pass
filters.

We usually use the ability to see the cardiac cycle in the data as a
criterion for usuable data.  If there is no cardiac visible in the data, it
probably is too low SNR (bad contact to the head).  You can prune channels
out of the analysis in HomER (based on raw signal intensity).


Hope that helps-



Ted Huppert, M.Sc.

PhD student-Harvard Univ.
Dept of Biophysics
Photon Migration Imaging lab
Mass General Hospital/CNY

Tele: (617)726-1223
Cell: (617) 869-1205

thuppert at nmr.mgh.harvard.edu