External Email - Use Caution        

Thanks, Raj.

I’d argue that, if the volumes of the nuclei were completely driven by the global volume of the amygdala, e.g., in an extreme case, V_nuc = alpha  x  V_whole, then the effect size of the nucleus at hand would be the same as for the whole amygdala. Whether such effect size is higher or lower than the “real” effect size in presence of perfect segmentation… depends on the nucleus ;-)

Doug: any thoughts?

Cheers,

/Eugenio

 

-- 

Juan Eugenio Iglesias

ERC Senior Research Fellow

Centre for Medical Image Computing (CMIC)

University College London

http://www.jeiglesias.com

http://cmictig.cs.ucl.ac.uk/

 

 

From: "Dr Rajendra Morey, M.D." <rajendra.morey@duke.edu>
Date: Monday, 1 October 2018 at 21:53
To: "Iglesias Gonzalez, Eugenio" <e.iglesias@ucl.ac.uk>, Freesurfer support list <Freesurfer@nmr.mgh.harvard.edu>
Subject: RE: [Freesurfer] amygdala segmentation atlas influence

 

Hi Eugenio

Thanks for your response it is helpful. In the dataset that I just analyzed with about 375 scans I suspect that the segmentation for some nuclei is heavily driven by the atlas. This is resulting in relatively low variance in both the patient group and the control group. The implication is that the between-group statistical test is overly sensitive and therefore produces higher than the expected chance   of a significant result. Is this a reasonable concern.

 

The counterargument is that the means of the two groups are closer to each other than the true means because of a heavy reliance on the atlas.  The implication here is that the between group statistical test is going to be more conservative than the true difference. Or does it mean the significance  of the test is not affected but that the effect size is artificially low. Let me know your thoughts!

 

Thanks

raj

                                                                                                                                                                                                                                           

From: Iglesias Gonzalez, Eugenio [mailto:e.iglesias@ucl.ac.uk]
Sent: Sunday, September 30, 2018 10:44 AM
To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>
Cc: Dr Rajendra Morey, M.D. <rajendra.morey@duke.edu>
Subject: Re: [Freesurfer] amygdala segmentation atlas influence

 

Dear Raj,

A few years ago, we published a paper on how Monte Carlo sampling could be used to quantify this. Unfortunately, the method was incredibly slow and difficult to tune, and the performance gain was tiny, so we never put it in FreeSurfer. But this is a direction I’m still interested in, and which I’m planning to revisit in the future.

Cheers,

/Eugenio

 

-- 

Juan Eugenio Iglesias

ERC Senior Research Fellow

Centre for Medical Image Computing (CMIC)

University College London

http://www.jeiglesias.com

http://cmictig.cs.ucl.ac.uk/

 

 

From: <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of Emily K Clarke <ekc23@duke.edu>
Reply-To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>
Date: Wednesday, 26 September 2018 at 15:15
To: "freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edu>
Cc: "Dr Rajendra Morey, M.D." <rajendra.morey@duke.edu>
Subject: [Freesurfer] amygdala segmentation atlas influence

 

        External Email - Use Caution        

Hello all,

I have run the amygdala segmentation in my datasets to obtain volumes of nuclei. I am aware from the fswiki that the volumes should be interpreted with caution because of the reliance on the atlas. When running FS, is there a way to get some metric about how much the segmentation relied on the atlas relative information from the T1 being segmented. If this metric is not available to the user, is some version of it calculated by the software and therefore available internally. Thanks in advance for your help!

raj