Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster peaks in the melodic_IC.dscalar.nii ICA maps

[Glasser, Matthew]

I meant the volume to surface mapping, so it sounds like your plan should work 
fine.

Peace,

Matt.

From: Claude Bajada >
Date: Friday, August 4, 2017 at 5:17 AM
To: Matt Glasser >, Timothy 
Coalson >
Cc: "hcp-users@humanconnectome.org" 
>
Subject: Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster 
peaks in the melodic_IC.dscalar.nii ICA maps


Thanks for the reply Matt.

I didn't quite understand what you mean by: " I would do your mapping in 
individual subjects."

In case it was not clear, the tractography is of course performed in individual 
subject space using the MSMAll surfaces in the subject directory of the T1w 
folders. My result is data associated with those surfaces and I was planning on 
averaging the data across participants.

Claude

On 04.08.2017 00:28, Glasser, Matthew wrote:
Tractography has quite strong folding-related biases.  Using MSMAll might 
attenuate those some and focus more on the parts of tractography that aren’t 
biased by folding.  I would do your mapping in individual subjects.

Peace,

Matt.

From: 
>
 on behalf of Claude Bajada 
>
Date: Thursday, August 3, 2017 at 4:54 PM
To: Timothy Coalson >
Cc: "hcp-users@humanconnectome.org" 
>
Subject: Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster 
peaks in the melodic_IC.dscalar.nii ICA maps


Thanks Tim and Michael,

I am thinking about doing some work with some statistics derived from 
tractography (starting from the vertices on gray white matter interface and 
then projecting those statistics onto their respective vertices) My thought 
however, is whether the MSMAll (or functional registration) is the most 
appropriate registration for working with what is essentially structure. I 
suppose that one could argue that tracts should be more aligned to function 
than gyral anatomy but it is not obviously true. Would you recommend using a 
more "anatomical" group registration in this case (eg the 
${subject}.${side}.32k_fs_LR.surf.gii files over the MSMAll)?

Claude

On 03.08.2017 21:46, Timothy Coalson wrote:
On Thu, Aug 3, 2017 at 9:03 AM, Claude Bajada 
> wrote:

Hi all,

I am starting a new thread because while my question is related to the one ask, 
it is tangential.

Can I confirm that what you mean by not averaging surfaces is that one should 
not average the vertex points across gifti surfaces to create a so-called 
"average surface"

Can I ask then, is averaging the data associated with vertices from individual 
subjects and plotting the result on a template surface (eg colin or a just 
using an individual as a template) also problematic?

Ah, I missed this question on my first read.  You make a good point, that 
individual surfaces have some bias away from the group.  With a good 
registration, doing this could tell you interesting things about the subject 
(the relative size of a particular group-identified feature on this 
individual).  Strictly speaking, though, it will cause some bias in the display 
of group results, in terms of the size of features.

Displaying on group average surfaces may actually have some bias too - 
different regions will lose different amounts of folding detail (because of 
differences in variability), which also means losing surface area (and 
therefore features in high variability regions may *look* smaller than they 
should on group surfaces).  For processing group-average data, we compensate 
for this surface area loss with the use of vertex areas.  I'm not entirely sure 
whether the surface inflation method we use tries to keep the vertex areas 
relatively undistorted, but if so, then the group very_inflated surfaces may 
have the least bias from this effect (less folding present in the surfaces 
before averaging, so less surface area lost due to averaging).  Note that this 
bias would be only in visual size, not in intensity or sampling density.

Back on the single-subject template topic, as a practical matter, we can't yet 
segment individual cerebellums into surfaces reliably, so carefully acquired 
and processed single subjects such as colin are the best we can do at present 
for displaying the cerebellum as a surface.
Regards,
Claude

On 03.08.2017 02:05, Timothy Coalson wrote:
On Wed, Aug 2, 2017 at 6:02 PM, James Morrow 
> wrote:
Thanks Tim and Matt for the detailed responses.

I agree that mapping to volumes is 

Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster peaks in the melodic_IC.dscalar.nii ICA maps

[Claude Bajada]

Thanks for the reply Matt.

I didn't quite understand what you mean by: " I would do your mapping in 
individual subjects."


In case it was not clear, the tractography is of course performed in 
individual subject space using the MSMAll surfaces in the subject 
directory of the T1w folders. My result is data associated with those 
surfaces and I was planning on averaging the data across participants.


Claude


On 04.08.2017 00:28, Glasser, Matthew wrote:
Tractography has quite strong folding-related biases.  Using MSMAll 
might attenuate those some and focus more on the parts of tractography 
that aren’t biased by folding.  I would do your mapping in individual 
subjects.


Peace,

Matt.

From: > on behalf of Claude 
Bajada >

Date: Thursday, August 3, 2017 at 4:54 PM
To: Timothy Coalson >
Cc: "hcp-users@humanconnectome.org 
" >
Subject: Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates 
of cluster peaks in the melodic_IC.dscalar.nii ICA maps


Thanks Tim and Michael,

I am thinking about doing some work with some statistics derived from 
tractography (starting from the vertices on gray white matter 
interface and then projecting those statistics onto their respective 
vertices) My thought however, is whether the MSMAll (or functional 
registration) is the most appropriate registration for working with 
what is essentially structure. I suppose that one could argue that 
tracts should be more aligned to function than gyral anatomy but it is 
not obviously true. Would you recommend using a more "anatomical" 
group registration in this case (eg the 
${subject}.${side}.32k_fs_LR.surf.gii files over the MSMAll)?


Claude


On 03.08.2017 21:46, Timothy Coalson wrote:
On Thu, Aug 3, 2017 at 9:03 AM, Claude Bajada > wrote:


Hi all,

I am starting a new thread because while my question is related
to the one ask, it is tangential.

Can I confirm that what you mean by not averaging surfaces is
that one should not average the vertex points across gifti
surfaces to create a so-called "average surface"

Can I ask then, is averaging the data associated with vertices
from individual subjects and plotting the result on a template
surface (eg colin or a just using an individual as a template)
also problematic?

Ah, I missed this question on my first read. You make a good point, 
that individual surfaces have some bias away from the group.  With a 
good registration, doing this could tell you interesting things about 
the subject (the relative size of a particular group-identified 
feature on this individual).  Strictly speaking, though, it will 
cause some bias in the display of group results, in terms of the size 
of features.


Displaying on group average surfaces may actually have some bias too 
- different regions will lose different amounts of folding detail 
(because of differences in variability), which also means losing 
surface area (and therefore features in high variability regions may 
*look* smaller than they should on group surfaces).  For processing 
group-average data, we compensate for this surface area loss with the 
use of vertex areas.  I'm not entirely sure whether the surface 
inflation method we use tries to keep the vertex areas relatively 
undistorted, but if so, then the group very_inflated surfaces may 
have the least bias from this effect (less folding present in the 
surfaces before averaging, so less surface area lost due to 
averaging).  Note that this bias would be only in visual size, not in 
intensity or sampling density.


Back on the single-subject template topic, as a practical matter, we 
can't yet segment individual cerebellums into surfaces reliably, so 
carefully acquired and processed single subjects such as colin are 
the best we can do at present for displaying the cerebellum as a surface.


Regards,
Claude

On 03.08.2017 02:05, Timothy Coalson wrote:

On Wed, Aug 2, 2017 at 6:02 PM, James Morrow
> wrote:

Thanks Tim and Matt for the detailed responses.


I agree that mapping to volumes is sub-optimal. Our goal is
to identify coords to be used as targets for brain
stimulation with TMS. We need MNI coords for
neuronavigation. Given the extent of the TMS field, we have
some tolerance for imprecisions in the mapping.


I see.  We generally get asked these questions in the context of
fMRI analysis, hence our reluctance.

How does the neuronavigation go from MNI coordinates to subject
coordinates, do you happen to have a reasonable T1w MRI scan of
your 

Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster peaks in the melodic_IC.dscalar.nii ICA maps

[Claude Bajada]

Thanks Tim and Michael,

I am thinking about doing some work with some statistics derived from 
tractography (starting from the vertices on gray white matter interface 
and then projecting those statistics onto their respective vertices) My 
thought however, is whether the MSMAll (or functional registration) is 
the most appropriate registration for working with what is essentially 
structure. I suppose that one could argue that tracts should be more 
aligned to function than gyral anatomy but it is not obviously true. 
Would you recommend using a more "anatomical" group registration in this 
case (eg the ${subject}.${side}.32k_fs_LR.surf.gii files over the MSMAll)?


Claude


On 03.08.2017 21:46, Timothy Coalson wrote:
On Thu, Aug 3, 2017 at 9:03 AM, Claude Bajada > wrote:


Hi all,

I am starting a new thread because while my question is related to
the one ask, it is tangential.

Can I confirm that what you mean by not averaging surfaces is that
one should not average the vertex points across gifti surfaces to
create a so-called "average surface"

Can I ask then, is averaging the data associated with vertices
from individual subjects and plotting the result on a template
surface (eg colin or a just using an individual as a template)
also problematic?

Ah, I missed this question on my first read.  You make a good point, 
that individual surfaces have some bias away from the group.  With a 
good registration, doing this could tell you interesting things about 
the subject (the relative size of a particular group-identified 
feature on this individual).  Strictly speaking, though, it will cause 
some bias in the display of group results, in terms of the size of 
features.


Displaying on group average surfaces may actually have some bias too - 
different regions will lose different amounts of folding detail 
(because of differences in variability), which also means losing 
surface area (and therefore features in high variability regions may 
*look* smaller than they should on group surfaces).  For processing 
group-average data, we compensate for this surface area loss with the 
use of vertex areas.  I'm not entirely sure whether the surface 
inflation method we use tries to keep the vertex areas relatively 
undistorted, but if so, then the group very_inflated surfaces may have 
the least bias from this effect (less folding present in the surfaces 
before averaging, so less surface area lost due to averaging).  Note 
that this bias would be only in visual size, not in intensity or 
sampling density.


Back on the single-subject template topic, as a practical matter, we 
can't yet segment individual cerebellums into surfaces reliably, so 
carefully acquired and processed single subjects such as colin are the 
best we can do at present for displaying the cerebellum as a surface.


Regards,
Claude

On 03.08.2017 02:05, Timothy Coalson wrote:

On Wed, Aug 2, 2017 at 6:02 PM, James Morrow
> wrote:

Thanks Tim and Matt for the detailed responses.


I agree that mapping to volumes is sub-optimal. Our goal is
to identify coords to be used as targets for brain
stimulation with TMS. We need MNI coords for neuronavigation.
Given the extent of the TMS field, we have some tolerance for
imprecisions in the mapping.


I see.  We generally get asked these questions in the context of
fMRI analysis, hence our reluctance.

How does the neuronavigation go from MNI coordinates to subject
coordinates, do you happen to have a reasonable T1w MRI scan of
your subjects?  I don't know how big the TMS field is, and I
hadn't looked at the distance from subject to group average
surfaces before, but in one of the HCP subjects, I got a maximum
of 2cm distance from the group average surface (using
midthickness surfaces), which occurred in a few specific
locations, while 90% of the surface was 1cm distance or less.

Can I clarify – was the ICA run on the volumes and then later
mapped on to surfaces, or was it performed on the surface
data? If the former, are the original volumetric results for
the ICA of each subject available anywhere in .nii format?


Cheers,

James


*James Morrow*
Research assistant
Brain & Mental Health Laboratory

*Monash Institute of Cognitive and Clinical Neurosciences*
School of Psychological Sciences
Monash University
c/o MBI, 770 Blackburn Road
Clayton VIC 3800
Australia

T: 03 9902 9768
E: james.mor...@monash.edu 
www.med.monash.edu.au/psych/bmh/

Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster peaks in the melodic_IC.dscalar.nii ICA maps

[Timothy Coalson]

On Thu, Aug 3, 2017 at 9:03 AM, Claude Bajada 
wrote:

> Hi all,
>
> I am starting a new thread because while my question is related to the one
> ask, it is tangential.
>
> Can I confirm that what you mean by not averaging surfaces is that one
> should not average the vertex points across gifti surfaces to create a
> so-called "average surface"
>
> Can I ask then, is averaging the data associated with vertices from
> individual subjects and plotting the result on a template surface (eg colin
> or a just using an individual as a template) also problematic?
>
Ah, I missed this question on my first read.  You make a good point, that
individual surfaces have some bias away from the group.  With a good
registration, doing this could tell you interesting things about the
subject (the relative size of a particular group-identified feature on this
individual).  Strictly speaking, though, it will cause some bias in the
display of group results, in terms of the size of features.

Displaying on group average surfaces may actually have some bias too -
different regions will lose different amounts of folding detail (because of
differences in variability), which also means losing surface area (and
therefore features in high variability regions may *look* smaller than they
should on group surfaces).  For processing group-average data, we
compensate for this surface area loss with the use of vertex areas.  I'm
not entirely sure whether the surface inflation method we use tries to keep
the vertex areas relatively undistorted, but if so, then the group
very_inflated surfaces may have the least bias from this effect (less
folding present in the surfaces before averaging, so less surface area lost
due to averaging).  Note that this bias would be only in visual size, not
in intensity or sampling density.

Back on the single-subject template topic, as a practical matter, we can't
yet segment individual cerebellums into surfaces reliably, so carefully
acquired and processed single subjects such as colin are the best we can do
at present for displaying the cerebellum as a surface.

> Regards,
> Claude
>
> On 03.08.2017 02:05, Timothy Coalson wrote:
>
> On Wed, Aug 2, 2017 at 6:02 PM, James Morrow 
> wrote:
>
>> Thanks Tim and Matt for the detailed responses.
>>
>>
>> I agree that mapping to volumes is sub-optimal. Our goal is to identify
>> coords to be used as targets for brain stimulation with TMS. We need MNI
>> coords for neuronavigation. Given the extent of the TMS field, we have some
>> tolerance for imprecisions in the mapping.
>>
>
> I see.  We generally get asked these questions in the context of fMRI
> analysis, hence our reluctance.
>
> How does the neuronavigation go from MNI coordinates to subject
> coordinates, do you happen to have a reasonable T1w MRI scan of your
> subjects?  I don't know how big the TMS field is, and I hadn't looked at
> the distance from subject to group average surfaces before, but in one of
> the HCP subjects, I got a maximum of 2cm distance from the group average
> surface (using midthickness surfaces), which occurred in a few specific
> locations, while 90% of the surface was 1cm distance or less.
>
>
>> Can I clarify – was the ICA run on the volumes and then later mapped on
>> to surfaces, or was it performed on the surface data? If the former, are
>> the original volumetric results for the ICA of each subject available
>> anywhere in .nii format?
>>
>>
>> Cheers,
>>
>> James
>>
>> *James Morrow*
>> Research assistant
>> Brain & Mental Health Laboratory
>>
>> *Monash Institute of Cognitive and Clinical Neurosciences*
>> School of Psychological Sciences
>> Monash University
>> c/o MBI, 770 Blackburn Road
>> Clayton VIC 3800
>> Australia
>>
>> T: 03 9902 9768
>> E: james.mor...@monash.edu 
>> www.med.monash.edu.au/psych/bmh/
>>
>> 
>> 
>>
>> On 3 August 2017 at 07:00, Timothy Coalson  wrote:
>>
>>> As Matt said, "MNI coordinates" of functionally-aligned cortical surface
>>> features don't have much meaning, similar to how T1w-aligned MNI space
>>> volumes don't have good cortical functional alignment (except in a few
>>> low-variability regions).  We have shown that group average surface
>>> coordinates do not follow the MNI cortical ribbon (see attached image that
>>> simply shows the average white and pial contours on top of the MNI
>>> nonlinear template).  The ICA maps themselves are also more informative
>>> than just the vertices at their peaks.
>>>
>>> Surface group results should not be turned into volumetric files, group
>>> average surfaces do not have much folding left in them, and as such they do
>>> not match the MNI template anymore - this is due to folding
>>> incompatibilities between subjects, but also due to using functional
>>> surface registration instead of folding patterns.  Using the individual
>>> 

Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster peaks in the melodic_IC.dscalar.nii ICA maps

[Timothy Coalson]

To explain further, the MSMAll registered surfaces have vertex indices
matched by nearby functional features, not by their position on folding
patterns.  Creating a group-average MSMAll surface is essentially asking
the question "for this functional feature, where is it, on average".  Based
on what we know of how function doesn't stay in the same place with respect
to folding across subjects, and how folding patterns in different subjects
can differ considerably, we know that this will have a sizable point cloud
to take the average of, so we know the answer isn't going to align well to
the feature in individuals, but this is okay as long as we understand this,
and don't try to use the result inappropriately.  Even with freesurfer or
MSMSulc surface registration, there are still many folding patterns that
don't really fit anywhere in the group template, so you will inevitably
have a spread-out point cloud in many regions (this subject has a small
extra gyrus, which falls where that subject has a sulcus, and this subject
has a sulcus that turns upward instead of going straight, etc).

The issue comes when you try to use these average coordinates as if they
are in good alignment with the volume data (expecting them to follow the
~2.6mm thick cortical ribbon well) - they are not, because the
volume-aligned data doesn't agree on where that cortical function is across
subjects - different subjects have it at different coordinates.

For other types of data that are functionally relevant (for instance,
fMRI), averaging by vertices across subjects using MSMAll is asking things
like "for this functional feature, what is its average signal change during
task X".  This is perfectly fine, and is exactly the question you want to
be asking.  It all boils down to what you expect of the data after
averaging, and how well the subject data co-locates with your registration
features.

Tim


On Thu, Aug 3, 2017 at 9:11 AM, Harms, Michael  wrote:

>
>
> Creating an “average-surface” is fine, and we in fact to that and provide
> it as part of our “group-average” data (available for download from
> ConnectomDB).
>
>
>
> But, those average surfaces are just a “back-drop” for
> overlaying/visualizing the metric/cifti data, and aren’t intended to be
> used as a substrate for extracting coordinates.
>
>
>
> Cheers,
>
> -MH
>
>
>
> --
>
> Michael Harms, Ph.D.
>
> ---
>
> Conte Center for the Neuroscience of Mental Disorders
>
> Washington University School of Medicine
>
> Department of Psychiatry, Box 8134
>
> 660 South Euclid Ave.Tel: 314-747-6173
> <(314)%20747-6173>
>
> St. Louis, MO  63110  Email:
> mha...@wustl.edu
>
>
>
> *From: * on behalf of Claude
> Bajada 
> *Date: *Thursday, August 3, 2017 at 9:06 AM
> *To: *"hcp-users@humanconnectome.org" 
> *Subject: *[HCP-Users] Follow-up to: Obtaining the MNI coordinates of
> cluster peaks in the melodic_IC.dscalar.nii ICA maps
>
>
>
> Hi all,
>
> I am starting a new thread because while my question is related to the one
> ask, it is tangential.
>
> Can I confirm that what you mean by not averaging surfaces is that one
> should not average the vertex points across gifti surfaces to create a
> so-called "average surface"
>
> Can I ask then, is averaging the data associated with vertices from
> individual subjects and plotting the result on a template surface (eg colin
> or a just using an individual as a template) also problematic?
>
> Regards,
> Claude
>
> On 03.08.2017 02:05, Timothy Coalson wrote:
>
> On Wed, Aug 2, 2017 at 6:02 PM, James Morrow 
> wrote:
>
> Thanks Tim and Matt for the detailed responses.
>
>
>
> I agree that mapping to volumes is sub-optimal. Our goal is to identify
> coords to be used as targets for brain stimulation with TMS. We need MNI
> coords for neuronavigation. Given the extent of the TMS field, we have some
> tolerance for imprecisions in the mapping.
>
>
>
> I see.  We generally get asked these questions in the context of fMRI
> analysis, hence our reluctance.
>
>
>
> How does the neuronavigation go from MNI coordinates to subject
> coordinates, do you happen to have a reasonable T1w MRI scan of your
> subjects?  I don't know how big the TMS field is, and I hadn't looked at
> the distance from subject to group average surfaces before, but in one of
> the HCP subjects, I got a maximum of 2cm distance from the group average
> surface (using midthickness surfaces), which occurred in a few specific
> locations, while 90% of the surface was 1cm distance or less.
>
>
>
> Can I clarify – was the ICA run on the volumes and then later mapped on to
> surfaces, or was it performed on the surface data? If the former, are the
> original volumetric results for the ICA of each subject available anywhere
> in .nii 

Re: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster peaks in the melodic_IC.dscalar.nii ICA maps

[Harms, Michael]

Creating an “average-surface” is fine, and we in fact to that and provide it as 
part of our “group-average” data (available for download from ConnectomDB).

But, those average surfaces are just a “back-drop” for overlaying/visualizing 
the metric/cifti data, and aren’t intended to be used as a substrate for 
extracting coordinates.

Cheers,
-MH

--
Michael Harms, Ph.D.
---
Conte Center for the Neuroscience of Mental Disorders
Washington University School of Medicine
Department of Psychiatry, Box 8134
660 South Euclid Ave.Tel: 314-747-6173
St. Louis, MO  63110  Email: 
mha...@wustl.edu

From:  on behalf of Claude Bajada 

Date: Thursday, August 3, 2017 at 9:06 AM
To: "hcp-users@humanconnectome.org" 
Subject: [HCP-Users] Follow-up to: Obtaining the MNI coordinates of cluster 
peaks in the melodic_IC.dscalar.nii ICA maps


Hi all,

I am starting a new thread because while my question is related to the one ask, 
it is tangential.

Can I confirm that what you mean by not averaging surfaces is that one should 
not average the vertex points across gifti surfaces to create a so-called 
"average surface"

Can I ask then, is averaging the data associated with vertices from individual 
subjects and plotting the result on a template surface (eg colin or a just 
using an individual as a template) also problematic?
Regards,
Claude
On 03.08.2017 02:05, Timothy Coalson wrote:
On Wed, Aug 2, 2017 at 6:02 PM, James Morrow 
> wrote:
Thanks Tim and Matt for the detailed responses.

I agree that mapping to volumes is sub-optimal. Our goal is to identify coords 
to be used as targets for brain stimulation with TMS. We need MNI coords for 
neuronavigation. Given the extent of the TMS field, we have some tolerance for 
imprecisions in the mapping.

I see.  We generally get asked these questions in the context of fMRI analysis, 
hence our reluctance.

How does the neuronavigation go from MNI coordinates to subject coordinates, do 
you happen to have a reasonable T1w MRI scan of your subjects?  I don't know 
how big the TMS field is, and I hadn't looked at the distance from subject to 
group average surfaces before, but in one of the HCP subjects, I got a maximum 
of 2cm distance from the group average surface (using midthickness surfaces), 
which occurred in a few specific locations, while 90% of the surface was 1cm 
distance or less.

Can I clarify – was the ICA run on the volumes and then later mapped on to 
surfaces, or was it performed on the surface data? If the former, are the 
original volumetric results for the ICA of each subject available anywhere in 
.nii format?

Cheers,
James

James Morrow
Research assistant
Brain & Mental Health Laboratory

Monash Institute of Cognitive and Clinical Neurosciences
School of Psychological Sciences
Monash University
c/o MBI, 770 Blackburn Road
Clayton VIC 3800
Australia

T: 03 9902 9768
E: james.mor...@monash.edu
www.med.monash.edu.au/psych/bmh/

[mage removed by sender.]
[mage removed by sender.]

On 3 August 2017 at 07:00, Timothy Coalson 
> wrote:
As Matt said, "MNI coordinates" of functionally-aligned cortical surface 
features don't have much meaning, similar to how T1w-aligned MNI space volumes 
don't have good cortical functional alignment (except in a few low-variability 
regions).  We have shown that group average surface coordinates do not follow 
the MNI cortical ribbon (see attached image that simply shows the average white 
and pial contours on top of the MNI nonlinear template).  The ICA maps 
themselves are also more informative than just the vertices at their peaks.

Surface group results should not be turned into volumetric files, group average 
surfaces do not have much folding left in them, and as such they do not match 
the MNI template anymore - this is due to folding incompatibilities between 
subjects, but also due to using functional surface registration instead of 
folding patterns.  Using the individual surfaces to map to the volume and then 
averaging across them would spread your data out just as badly as doing 
volume-based analysis of cortex, so this is also highly discouraged.

Despite strongly advising you not to do what you outlined, I will tell you what 
commands you would need.  The -cifti-extrema command will output a map with 1s 
and -1s at each local extrema.  The surface it uses is for neighbor information 
and distance computation - this isn't as critical as coordinates are, as it 
merely sets the maximum possible density of extrema (what you could get from a 
very noisy map).  You will need to