[Homer-users] RE: Principal Component Analysis in HomER

[thuppert]

The PCA filters in HOMER remove the strongest N components depending on the
value entered (i.e. the dominant components of the spatial covariance are
removed).

You are absolutely correct.  PCA is designed to remove global changes to the
data.  If you apply this to breath-holding (etc), than it may have
undesirable effects (namely, remove the signal of interest).  PCA works best
for NIRS probes that are much larger than the area of functional activity.

The 3 PCA filters in HOMER differ only by what data is used to compute
spatial covariance.  The first filter uses ALL data (inc all wavelengths),
this will remove non-spectroscopic signals (i.e. motion artifacts).  The
first filter may be appropriate for this data.  The 2nd and 3rd filters use
optical density changes (per wavelength) or concentration (per Hb species),
these will remove global changes.  These are NOT appropriate for this type
of data.  (note- these filters can use separate baseline data to define
spatial covariance- its unclear if baseline systemic fluctuations are
(approx) orthogonal to breath-hold changes etc- the results are unclear at
best).  My advice is always to use the MINIMUM signal processing possible.


Note- PCA can also be used the other way to remove the lowest components of
the signal (often called a truncated SVD).  This has the effect of
regularizing the spatial structure of the data to enforce more global
changes and remove localized "noise".  This is not implemented in HOMER.
Although, the results would be very similar to regularization of the image
reconstructions.



Ted Huppert, M.Sc.

PhD Candidate
Harvard University:
Graduate Programs in Biophysics
Photon Migration Imaging lab
Massachusetts General Hospital
Tel: (617)726-9338

thuppert at nmr.mgh.harvard.edu








  -----Original Message-----
  From: Santosh Hariharan [mailto:santosh.hariharan at gmail.com]
  Sent: Monday, November 27, 2006 7:30 PM
  To: thuppert at nmr.mgh.harvard.edu
  Subject: Principal Component Analysis in HomER


  Dear Sir,

                 I am a graduate student of University of Texas, Arlington.
I have enrolled for my masters in BioMedical Engineering.I am using homer
for processing data. Just had a small query regarding PCA filtering
implemented in HomER.
  When we use the PCA filter(All the three kind provided in HomER), does it
filter or remove the higher eigen value principal component (Number
depending on the value we enter in the edit field)?

                I am asking this because, the data i have collected is for
breath holding, rebreathing, hyperventilation, which are also global
phenomenon.

  --
  Regards,
  Santosh Hariharan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/homer-users/attachments/20061128/90ea4179/attachment.html