[IGSTK-Developers] FW: error prediction in rigid body point based registration: comments from Jay West

Thu Feb 9 14:40:21 EST 2006

David and others:

Here are some comments from Jay West at Accuray on the Wiki page. Jay is on
my program committee at SPIE so I will see him this week and thank him for
the comments.

David, can you update the page at your convenience to reflect these
comments? I am sure if you have some brief questions you can ask Jay West
directly, but I doubt he has time for long discussions. But to me this is
the beauty of open source as everything is open for discussion, comment, and
improvement.

Kevin

  _____  

From: Jay West [mailto:jwest at accuray.com] 
Sent: Thursday, February 09, 2006 2:32 PM
To: Kevin Cleary
Subject: Re: error prediction in rigid body point based registration

Hi Kevin,

Sorry for the delay - as always, it's a struggle to find free time here at
Accuray. Here are my comments on the Wiki page:

Definitions: 1: "homologus" should read; homologous

Definitions: 3: "homologus" should read; homologous

Definitions: 4 should read something like "Expected target registration
error: statistical expected value of TRE, derived from distributions based
on theoretical derivations or past observation."

Motivation: "root means square" should read; "root mean square". Possibly
reference Sibson '79 paper for claim that FRE is independent of fiducial
configuration.

Motivation: "derived approximate expression" should read "derived an
approximate expression"

Motivation: last line - "function of target registration error" should read;
"function of FLE". 

Motivation: last line - "the the fiducial points" should read; "the fiducial
points"

David's suggestion: I agree that it's important to handle the composition of
transforms. There is no method of doing this specified in the suggestion - I
suggest referencing the TMI paper on this by Calvin and me, published in
2004: "Designing optically tracked instruments for image-guided surgery",
Jay B. West and Calvin R. Maurer, Jr., vol 23 (5), 533-545, May 2004.

What information is needed to compute TRE: Assuming you're working in three
dimensions, seems to be intuitive that whatever formulation you use, you
need 

(a) The point at which you're calculating, in some arbitrary coordinate
system (three parameters)

(b) The principal axes of the fiducial configuration: the page claims this
can be done with 3 parameters - I would use 6 (two vectors, the third being
the cross product of the first two) for simplicity.

(c) The centroid of the fiducial configuration (three parameters)

(d) The value of the moments of the fiducial configuration about each axis
(three parameters)

Hence you need 12 parameters, including the three which are arguments to the
computation function. 

The answer to (1) is that yes, all rigid body registration errors can be
described with the same basic formulations, assuming obviously that you're
subject to the restrictions and happy with the assumptions made in
Fitzpatrick et al. 

For (2), yes, see the IEEE TMI paper I mentioned above.

Simplification: I agree with the mathematical logic here. However, in many
cases, especially those with moments about the axes that vary greatly, this
will result in a gross overestimation of TRE at many places. I'm not really
familiar with IGSTK, but I don't intuitively see why maintaining ten
floating point numbers and hence giving accurate TRE estimated everywhere
should be an enormous overhead. 

James' Suggestion: I agree that the name should be something like
"PredictTargetRegistrationError", or "ExpectedTargetRegistrationError". 

Stephen's Suggestion: Yes, the TRE work assumes a rigid registration
transform. It also assumes that the registration method used is minimization
of rms FRE between corresponding fiducial points. If either of these
assumptions is violated, the TRE formula no longer holds. For deformable
registration methods, it's not enough to simply approximate TRE by looking
at the registration transform as a locally rigid transform and derive TRE
point-by-point in this way. The trouble is that the transform is derived in
a different way, and so the TRE approximation is not valid. 

Jay

-----Original Message-----
From: Kevin Cleary 
Sent: Friday, January 06, 2006 6:48 PM
To: Jay West
Subject: RE: error prediction in rigid body point based registration

Jay - 

No particular deadline, but certainly by the SPIE conference would be useful
as several of the developers will be there

I do not know this person, but I do know the person who heads the biophysics
group and would be happy to check with him if you like

Kevin

  _____  

From: Jay West [mailto:jwest at accuray.com] 
Sent: Thursday, January 05, 2006 11:46 AM
To: Kevin Cleary
Subject: Re: error prediction in rigid body point based registration

Kevin,

Sorry for the delay - it's still on my list of things to do. Is there a
deadline for this that I need to work to?

BTW, I'm about to interview a guy called Zegang Dong for a medical physicist
position here. His PhD is

from Georgetown Biophysics dept. - do you know him, and if so, would you
recommend him?

Thanks

Jay

-----Original Message-----
From: Kevin Cleary 
Sent: Thursday, January 05, 2006 8:41 AM
To: Jay West
Subject: RE: error prediction in rigid body point based registration

Jay

Happy New Year! Any chance you can take a look at this Wiki page and give us
some feedback?

We have been getting things organized for SPIE and I think it will be a good
meeting

Kevin

------------------------------------------------------------------
Kevin Cleary, Ph.D.                        Work phone: 202-687-8253
Associate Professor                        Work fax: 202-784-3479
Deputy Director  

Imaging Science and Information Systems (ISIS) Center
Department of Radiology                    Pager: 202-901-2033
Georgetown University Medical Center       Cell phone: 202-294-3409
2115 Wisconsin Avenue, Suite 603           Home phone: 301-299-0788
Washington, DC, 20007                      Home fax: 301-299-0789

ISIS center:  <http://www.isis.georgetown.edu/> www.isis.georgetown.edu
Research group:  <http://www.caimr.georgetown.edu/> www.caimr.georgetown.edu
WashCAS:  <http://www.washcas.org/> www.washcas.org
Email:  <mailto:cleary at georgetown.edu> cleary at georgetown.edu
-------------------------------------------------------------------

  _____  

From: Jay West [mailto:jwest at accuray.com] 
Sent: Sunday, December 11, 2005 7:49 PM
To: Kevin Cleary
Subject: Re: error prediction in rigid body point based registration

Hi Kevin:

Yes, I'd be happy to take a look at the page and let you know my take on it.

Look forward to seeing you at SPIE in Feb.

Regards

Jay

-----Original Message-----
From: Kevin Cleary 
Sent: Thursday, December 08, 2005 11:26 AM
To: Jay West
Subject: error prediction in rigid body point based registration

Jay:

Hope all is well at Accuray. I want to ask a favor of you. I am the
principal investigator of an NIH funded open source software project to
develop a software toolkit for image-guided surgery. As one of the parts of
this project, we are implementing a point based registration class. Would
you be so kind to take a look at the Wiki page on this and give me some
feedback? This is no hurry and I would be glad to give you some more
background on the project any time. We are planning a small demo at SPIE.

http://public.kitware.com/IGSTKWIKI/index.php/Error_Prediction_in_Rigid-Body
_Point-Based_Registration    

Thanks

Kevin

------------------------------------------------------------------
Kevin Cleary, Ph.D.                        Work phone: 202-687-8253
Associate Professor                        Work fax: 202-784-3479
Deputy Director  

Imaging Science and Information Systems (ISIS) Center
Department of Radiology                    Pager: 202-901-2033
Georgetown University Medical Center       Cell phone: 202-294-3409
2115 Wisconsin Avenue, Suite 603           Home phone: 301-299-0788
Washington, DC, 20007                      Home fax: 301-299-0789

ISIS center:  <http://www.isis.georgetown.edu/> www.isis.georgetown.edu
Research group:  <http://www.caimr.georgetown.edu/> www.caimr.georgetown.edu
WashCAS:  <http://www.washcas.org/> www.washcas.org
Email:  <mailto:cleary at georgetown.edu> cleary at georgetown.edu
-------------------------------------------------------------------

WARNING CONFIDENTIAL INFORMATION:

The information contained in the e-mail may contain confidential and
privileged information and is intended solely for the use of the intended
recipient(s).  Access for any review, re-transmission, dissemination or
other use of, or taking of any action in regard and reliance upon this
e-mail by persons or entities other than the intended recipient(s) is
unauthorized and prohibited.  If you are not the intended recipient, please
contact the sender by reply e-mail and destroy all copies of the original
message and any attachments.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/igstk-developers/attachments/20060209/ff2bd37a/attachment-0001.html>