VBM8-­‐Toolbox

 Manual  
INTRODUCTION  AND  OVERVIEW   2  

GETTING  STARTED   3  
Download  and  Installation   3  
Starting  the  Toolbox   3  
Basic  VBM  analysis  (overview)   4  

BASIC  VBM  ANALYSIS  (DETAILED  DESCRIPTION)   6  

Modules   6  
First  Module:  Estimate  and  write   6  
Second  Module:  Display  one  slice  for  all  images   8  
Third  Module:  Check  sample  homogeneity  using  covariance   8  
Fourth  Module:  Smooth   9  

Building  the  statistical  model   10  

Two-­‐sample  T-­‐Test   11  
Using  the  Full  Factorial  Model  (for  a  2x2  Anova)   12  
Multiple  Regression  (Correlation)   13  
Using  the  Full  Factorial  Model  (for  an  Interaction)   14  
Estimating  the  statistical  model   15  
Defining  Contrasts   15  

SPECIAL  CASES   17  

VBM8  for  longitudinal  data   17  

Change  Settings  for  Preprocessing   18  
Preprocessing  of  longitudinal  data   18  
Statistical  Analysis  of  longitudinal  data  in  one  group   19  
Statistical  Analysis  of  longitudinal  data  in  two  groups   20  

Altered  Workflows  for  VBM-­analyses   23  

Adapting  the  workflows   24  

Customized  Tissue  Probability  Maps   24  
Customized  DARTEL-­‐template   25  

ADDITIONAL  INFORMATION  ON  NATIVE,  NORMALIZED  AND  MODULATED  VOLUMES   27  

NAMING  CONVENTION  OF  OUTPUT  FILES   28  

TECHNICAL  INFORMATION   29  

REFERENCES   30  

  1  
  
Introduction  and  Overview  
 
This  manual  is  intended  to  help  any  user  to  perform  a  VBM  analysis  using  the  VBM8  Toolbox.  It  covers  
different  aspects  of  this  type  of  analysis  –  starting  with  the  very  basics  and  ending  up  with  additional  
information  on  how  to  deal  with  different  data.  Basically  it  may  be  divided  into  four  main  sections:  
 
• Naturally,   a   quick   guide   of   how   to   get   started   is   given   at   the   beginning.   This   section   provides  
information   how   to   download   and   install   the   software   and   start   the   Toolbox.   Furthermore,   a   short  
overview  on  the  steps  of  a  VBM  analysis  is  given.  
 
• A  detailed  description  of  a  basic  VBM  analysis  is  subsequently  given,  which  will  guide  the  user  step  
by   step   through   the   whole   process   –   from   preprocessing   to   the   selection   of   contrasts.   This  
description  should  provide  all  necessary  information  to  analyze  most  studies  successfully.  
 
• There   are   a   few   special   cases   of   VBM   analyses,   for   which   the   basic   analysis   workflow   has   to   be  
adapted.  These  cases  are  longitudinal  studies  and  studies  in  children  or  special  patient  populations.  
Relevant   changes   to   a   basic   VBM   analysis   are   described   here   and   a   description   of   how   to   apply  
these   changes   is   provided.   Importantly,   only   the   changes   are   described   –   steps   like   for   example  
quality  control  or  smoothing  are  the  same  as  in  the  basic  analysis  and  not  described  a  second  time.  
 
• The   manual   closes   with   information   on   native,   normalized   and   modulated   volumes,   which  
determines   how   the   results   may   be   interpreted.   Furthermore   an   overview   of   the   naming  
conventions  used  as  well  as  technical  information  is  given.  
 
 
 

  2  
Getting  Started  
DOWNLOAD  AND  INSTALLATION  
• The   VBM8   Toolbox   runs   within   SPM8.   That   is,   SPM8   needs   to   be   installed   and   added   to   your  
Matlab   search   path   before   the   VBM8   Toolbox   can   be   installed   (see  
http://www.fil.ion.ucl.ac.uk/spm/  and  http://en.wikibooks.org/wiki/SPM).    
• Download   (http://dbm.neuro.uni-­‐jena.de/vbm8/)   and   unzip   the   VBM8   Toolbox.   You   will   get   a  
folder  named  “vbm8”,  which  contains  various  matlab  files  and  compiled  scripts.  Copy  the  folder  
“vbm8”  into  the  SPM8  “toolbox”  folder.    
 
 
STARTING  THE  TOOLBOX  
• Start  Matlab  
• Start  SPM8  (i.e.,  type  “spm  fmri”)  
• Select  “vbm8”  from  the  SPM  menu  (see  Figure  1).  You  will  find  the  drop-­‐down  menu  between  the  
“Display”  and  the  “Help”  button.  This  will  open  the  VBM8  Toolbox  (in  SPM’s  2nd  window).  You  
can  access  the  VBM8  Toolbox  menu  by  clicking  on  “VBM8”  (located  at  the  upper  left  corner  of  
this  window;  see  Figure  2).  
 
 
 

 
Figure  1:  SPM  menu   Figure   2:   SPM’s   2nd   window   with   the   VBM8  
Toolbox  menu  
 
 
 
  3  
BASIC  VBM  ANALYSIS  (OVERVIEW)  
The   VBM8   Toolbox   comes   with   different   modules,   which   may   be   used   for   an   analysis.   Usually,   a   VBM  
analysis  comprises  the  following  steps:  
 
(a)  Preprocessing:  
1. T1  images  are  normalized  to  a  template  space  and  segmented  into  gray  matter  (GM),  white  
matter   (WM)   and   cerebrospinal   fluid   (CSF).   The   preprocessing   parameters   can   be   adjusted   via  
the  module  “Estimate  and  write”.  
2. After   the   preprocessing   is   finished,   a   quality   check   is   highly   recommended.   This   can   be  
achieved  via  the  modules  “Display  one  slice  for  all  images”  and  “Check  sample  homogeneity  
using  covariance”.  Both  options  are  located  under  “VBM8  Check  data  quality”.  
3. Before  entering  the  GM  images  into  a  statistical  model,  image  data  need  to  be  smoothed.  Of  
note,  this  step  is  not  implemented  into  the  VBM8  Toolbox  but  achieved  via  the  standard  SPM  
module  “Smooth”.  
 
(b)  Statistical  analysis:  
4. The   smoothed   GM   images   are   entered   into   a   statistical   analysis.   This   requires   building   a  
statistical   model   (e.g.,   T-­‐Tests,   ANOVAs,   multiple   regressions).   This   is   done   by   the   standard  
SPM  modules  “Specify  2nd  Level”.  
5. The  statistical  model  is  estimated.  This  is  done  by  the  standard  SPM  module  “Estimate”.  
6. After   estimating   the   statistical   model,   contrasts   will   be   defined   to   get   the   results   of   the  
analysis.  This  is  done  by  the  standard  SPM  module  “Results”.  
 
 
The  sequence  of  “preprocessing    quality  check    smoothing    statistical  analysis”  remains  the  
same  for  every  VBM  analysis,  even  when  different  steps  are  adapted  (see  “special  cases”).  
 
 
 

  4  
A  few  words  about  the  Batch  Editor…  
− As   soon   as   you   select   a   module   from   the   VBM8   Toolbox   menu,   a   new   window   (the   Batch  
Editor)  will  open.  The  Batch  Editor  is  the  environment  where  you  will  set  up  your  analysis  (see  
Figure  3).  For  example,  an  “<-­‐X”  indicates  where  you  need  to  select  files  (e.g.,  your  image  files,  
the  template,  etc.).  Other  parameters  have  either  default  settings  (which  can  be  modified)  or  
require  input  (e.g.,  choosing  between  different  options,  providing  text  or  numeric  values,  etc.).  
− Once  all  missing  parameters  are  set,  a  green  arrow  will  appear  on  the  top  of  the  window  (the  
current  snapshots  in  Figure  3  show  the  arrow  still  in  gray).  Click  this  arrow  to  run  the  module  
or   select   “File      Run   Batch”.   It   is   very   useful   to   save   the   settings   before   you   run   the   batch  
(click  on  the  disk  symbol  or  select  “File    Save  Batch”).  
− Of   note,   you   can   always   find   helpful   information   and   parameter-­‐specific   explanations   at   the  
bottom  of  the  Batch  Editor  window.1  
− All  settings  can  be  saved  either  as  .mat  file  or  as  .m  script  file  and  reloaded  for  later  use.  The  
.m  script  file  has  the  advantage  to  be  editable  with  a  text  editor.  
 

Figure   3:   The   Batch   Editor   is   the   environment   where   the   analysis   is   set   up.   Left:   For   all   settings  
marked   with   “<-­‐X”,   files   have   to   be   selected   (“Select   Files”).   Right:   Parameters   can   be   edited   and  
adapted  (“Edit  Value”).  
 

                                                                                                           
1
 Additional  VBM8-­‐related  information  can  be  found  by  selecting  “VBM  Tools  website“  in  the  VBM8  Toolbox  menu.  This  
will  open  a  website.  Here,  look  for  “VBM  subpages”  on  the  right.  

Of   note,   for   debugging,   the   relevant   information   can   be   printed   by   selecting   “Print   VBM   debug   information”   from   the  
VBM8  Toolbox  menu.  

  5  
Basic  VBM  analysis  (detailed  description)  
1. Select  your  working  directory.  
Before   running   anything,   it   is   recommended   to   always   set   SPM’s   working   directory:  
“Utilities      CD”   (right   to   the   “Help”   button   in   the   SPM   main   menu).   Select   your   working  
directory  and  click  “Done”  to  change  the  directory.  
2. Open  the  VBM  Toolbox.  
3. Select  the  module  you  want  to  run.  
4. Specify  the  parameters  (see  below).  
5. Run  the  module.    
 
 
 
 
 

Modules  
 
FIRST  MODULE:  ESTIMATE  AND  WRITE  
 
 
VBM8    Estimate  and  write  
  Parameters:  
o Volumes  <-­‐X    Select  Files    [select  raw  data]    Done  
- Select   one   volume   for   each   subject.   As   the   Toolbox   does   not   support   multispectral  
data   yet   (i.e.,   different   imaging   methods   for   the   same   brain,   such   as   T1-­‐,   T2-­‐,  
diffusion-­‐weighted   or   CT   images),   it   is   recommended   to   choose   a   T1-­‐weighted  
image.  
- Importantly,  the  images  need  to  be  in  the  same  orientation  as  the  template  and  
priors;  you  can  double-­‐check  and  correct  via  using  “Display”  in  the  SPM  menu.  By  
default  the  MNI  template  is  used  (located  in  your  SPM  folder  “spm8    templates  
  T1”)  
o Estimation  Options    [use  defaults  or  modify]  
- The  defaults  provide  a  solid  starting  point  for  the  analysis.  If  you  prefer  to  use  
your  own  customized  Tissue  Probability  Maps  (TPMs),  you  can  specify  /  select  
them   here.   However,   the   new   segmentation   approach   in   VBM8   does   not  
need   tissue   priors   for   the   segmentation   anymore.   Customized   TPMs   will   be  
only  used  for  the  spatial  normalization.  Thus,  creating  your  own  TPMs  might  
be   only   appropriate   for   children’s   data   if   they   largely   deviate   from   the  

  6  
 standard   MNI   template.   In   order   to   create   your   own   TPMs   the   Template-­‐O-­‐
Matic  (TOM8)  Toolbox  can  be  used.  
o Extended  Options    [use  defaults  or  modify]  
- Again,   the   defaults   provide   a   solid   starting   point.   The   high-­‐dimensional  
DARTEL   normalization   is   used   as   the   default   spatial   normalization.  
Alternatively,  the  low-­‐dimensional  SPM8  spatial  normalization  can  be  chosen.  
Furthermore,  remaining  non-­‐brain  tissue  can  be  removed  more  thoroughly  by  
setting   the   option   “Clean   up   any   partitions”   to   “Thorough   Cleanup”.   This   is  
particularly   useful   for   very   atrophic   brains   (e.g.,   as   occurring   in   Alzheimer’s  
disease).   The   parameters   for   two   de-­‐nosing   methods   can   be   also   changed.  
The  optimal  weighting  for  the  SANLM  filter  is  internally  estimated.  The  MRF  
weighting   is   not   necessary   to   change,   because   the   SANLM   filter   will   have   a  
much  larger  de-­‐nosing  effect.  Values  of  “0”  will  deselect  both  filters.  
o Writing  Options    [use  defaults  or  modify]  
- For  GM,  WM,  and  CSF  image  volumes  see  page  14:  “Additional  Information  
on  native,  normalized  and  modulated  normalized  volumes”.  Note:  The  default  
option  “Modulated  normalized  –  non  linear  only”  will  result  in  an  analysis  of  
relative  differences  in  regional  GM  volume,  corrected  for  individual  brain  size.  
- A  bias  corrected  image  volume,  in  which  MRI  inhomogeneities  and  noise  are  
removed,   can   be   written   in   normalized   or   native   space.   This   is   useful   for  
quality   control   and   also   to   create   an   average   image   of   all   normalized   T1  
images  in  order  to  display  /  overlay  the  results.  Note:  For  a  basic  VBM  analysis  
use  the  defaults.  
- A   partial   volume   effect   (PVE)   label   image   volume   can   also   be   written   in  
normalized  or  native  space  or  as  a  DARTEL  export  file.  This  is  useful  for  quality  
control   and   also   for   future   applications   using   this   image   to   reconstruct  
surfaces.  Note:  For  a  basic  VBM  analysis  use  the  defaults.  
- The   Jacobian   determinant   for   each   voxel   can   be   written   in   normalized   space.  
This   information   can   be   used   to   do   a   Tensor-­‐Based   Morphometry   (TBM)  
analysis.  Note:  For  a  basic  VBM  analysis  this  is  not  needed.  
- Finally,   deformation   fields   can   be   written.   This   option   is   useful   to   re-­‐apply  
normalization   parameters   to   other   images   or   particular   regions   of   interest.  
Note:  For  a  basic  VBM  analysis  this  is  not  needed.  
 
After  setting  all  parameters,  you  will  be  able  to  save  and  run  the  module:  
- File    Save  Batch  [optionally  save  your  model  and  selections  as  a  *.m  script  file;  the  next  time  
you   can   simply   load   this   *.m   script   file   with   “Load   Batch”,   and   adjust   /   modify   respective   sections  
if  necessary].  
-­‐   File    Run  Batch  [the  outcomes  will  be  written  to  the  same  directory  like  the  original  data].  As  
per   defaults,   the   outcomes   will   be   the   bias   corrected   normalized   volumes   (wm*)   and   the   tissue  
segments   (i.e.,   the   modulated   normalized   gray   matter   (m0wrp1*)   and   white   matter   (m0wrp2)  
segments  of  each  volume).  If  you  selected  the  low-­‐dimensional  spatial  normalization  approach  the  
modulated   images   will   be   named   m0wp1*   for   gray   matter   and   m0wp2*   for   white   matter.   All  
  7  
 possible   naming   is   summarized   under   “Naming   convention   of   output   files”   (p.   17)   and   in   the  
graphics  window  of  SPM,  when  the  SPM8  Toolbox  is  started.  
 
 
SECOND  MODULE:  DISPLAY  ONE  SLICE  FOR  ALL  IMAGES  
 
 
VBM8    Check  data  quality    Display  one  slice  for  all  images  
  Parameters:  
o Volumes  <-­‐X    Select  Files    [select  the  new  files]    Done  
- Select  the  newly  written  data  [e.g.  the  “wm*”  files,  which  are  the  normalized  bias  
corrected  volumes].  This  tool  will  display  one  horizontal  slice  for  each  subject,  thus  
giving  a  good  overview  if  the  segmentation  and  normalization  procedures  yielded  
reasonable  results.  For  example,  if  the  native  volume  had  artifacts  or  if  the  native  
volumes   had   a   wrong   orientation,   the   results   may   look   odd.   Solutions:   Use   “Check  
Reg”  from  the  SPM  main  menu  to  make  sure  that  the  native  images  have  the  same  
orientation   like   the   MNI   Template   (“SPM      templates      T1”).   Adjust   if   necessary  
using  “Display”  from  the  SPM  main  menu.  
o Proportional  scaling  [use  defaults  or  modify]  
- Check  “yes”,  if  you  display  T1  volumes.  
Show  slice  in  mm    [use  defaults  or  modify]  

- This   module   displays   horizontal   slices.   This   default   setting   provides   a   good  
overview.  
 
- File    Save  Batch    
- File    Run  Batch  [the  outcomes  will  be  displayed  in  SPM’s  graphic  window]  
 
 
THIRD  MODULE:  CHECK  SAMPLE  HOMOGENEITY  USING  COVARIANCE  
 
 
VBM8    Check  data  quality    Check  sample  homogeneity  using  covariance  
  Parameters:  
o Volumes  <-­‐X    Select  Files    [select  gray  matter  volumes]    Done  
- Select  the  newly  written  data  [e.g.  the  “m0wrp1*”  files,  which  are  the  normalized  
(wr)  GM  segments  (p1)  modulated  for  the  non-­‐linear  components  (m0)].  This  tool  
visualizes   the   covariance   between   the   volumes   using   a   boxplot   and   covariance  
matrices.   Thus,   it   will   help   identifying   outliers.   Any   outlier   should   be   carefully  

  8  
 inspected   for   artifacts   or   pre-­‐processing   errors   using   “Check   Reg”   from   the   SPM  
main  menu.  
o Proportional  scaling    [use  defaults  or  modify]  
- Check  “yes”,  if  you  display  T1  volumes.  
o Show  slice  in  mm    [use  defaults  or  modify]  
- “0mm”  means  the  center  slice  along  the  origin  of  the  MNI  template.    
o Nuisance    [enter  nuisance  variables  if  applicable]  
- For  each  nuisance  variable  which  you  want  to  remove  from  the  data  prior  to  
calculating   the   covariance,   select   “New:   Nuisance”   and   enter   a   vector   with  
the  respective  variable  for  each  subject  (e.g.  age  in  years).  All  variables  have  
to   be   entered   in   the   same   order   as   the   respective   volumes.   You   can   also   type  
“spm_load”  to  upload  a  *txt  file  with  the  covariates  in  the  same  order  as  the  
volumes.  
- File    Save  Batch    
- File    Run  Batch    
- A   boxplot   window   and   covariance   matrices   will   open,   which   depict   the  
covariance  between  the  volumes;  outliers  can  be  displayed  in  SPM’s  graphic  
window.   The   covariance   matrix   shows   the   covariance   between   all   volumes.  
High   covariance   values   mean   that   your   data   are   more   similar.   The   boxplot  
summarizes   all   covariance   values   for   each   subject   and   shows   the  
homogeneity   of   your   sample.   A   small   overall   covariance   in   the   boxplot   not  
always  means  that  this  volume  is  an  outlier  or  contains  an  artifact.  If  there  are  
no  artifacts  in  the  image  and  if  the  image  quality  is  reasonable  you  don’t  have  
to  exclude  this  volume  from  the  sample.  This  tool  is  intended  to  utilitize  the  
process  of  quality  checking  and  there  is  no  clear  criteria  defined  to  exclude  a  
volume   only   based   on   the   overall   covariance   value.   However,   volumes   with  
an  overall  covariance  below  two  standard  deviations  are  indicated  and  should  
be  checked  very  carefully.  
 
 
 
 
 
 
 
FOURTH  MODULE:  SMOOTH  
 
 
SPM  menu    Smooth  
  Parameters:  
o Images  to  Smooth  <-­‐X    Select  Files    [select  grey  matter  volumes]    Done  

  9  
 - Select   the   newly   written   data   [e.g.   the   “m0wrp1”   files,   which   are   the  
normalized   (wr)   grey   matter   segments   (p1)   modulated   for   the   non-­‐linear  
components  (m0)].  
o FWHM    [use  defaults  or  modify]  
- 8-­‐12mm   kernels   are   widely   used   for   VBM.   To   use   this   setting   select   “edit  
value”  and   type   “8   8  8”   (or   “12  12  12”,  respectively)  for  a  kernel  with  8mm  
(with  12mm)  FWHM.  
o Data  Type    [use  defaults  or  modify]  
o Filename  Prefix    [use  defaults  or  modify]  
 
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [the  outcomes  will  be  written  to  the  same  directory  like  the  original  data]  
 
 
 
 
 

Building  the  statistical  model  

Although  there  are  many  potential  designs  offered  in  the  2nd-­‐level  analysis  I  recommend  to  use  the  
“Full   factorial”   design   because   it   covers   most   statistical   designs.   For   cross-­‐sectional   VBM   data   you  
have  usually  1..n  samples  and  optionally  covariates  and  nuisance  parameters:  
 
Number  of  factor  levels   Number  of  covariates   Statistical  Model  
   1      0   one-­‐sample  t-­‐test  
   1      1   single  regression  
   1   >1   multiple  regression  
   2      0   two-­‐sample  t-­‐test  
>2      0   Anova  
Ancova   (for   nuisance   parameters)   or  
>1   >0  
Interaction  (for  covariates)  
 

  10  
  
 
TWO-­‐SAMPLE  T-­‐TEST  
 
 
SPM  menu    Specify  2nd-­‐level  
  Parameters:  
o Directory  <-­‐X    Select  Files    [select  the  working  directory  for  your  analysis]    Done  
o Design    “Two-­‐sample  t-­‐test”  
 Group   1   scans      Select   Files      [select   the   smoothed   grey   matter   volumes   for  
group   1;   following   this   script  
these   will   be   the   “sm0wp1”   files]   *
You  could  specify  one  or  many  
  Done   covariates  (i.e.,  partial  out  the  
 Group   2   scans      Select   Files      variance  of  specific  factors  when  
[select   the   smoothed   grey   matter   looking  at  group  differences)  
volumes  for  group  2]    Done   • Covariates    New  Covariate  
• Vector  <-­‐X    enter  the  values  of  
 Independence    Yes   the  covariates  (e.g.,  age  in  years)  
 Variance    Equal  or  Unequal   in  the  same  order  as  the  
 Grand  mean  scaling    No   respective  file  names  or  type  
“spm_load”  to  upload  a  *.txt  file  
 ANCOVA    No  
with  the  covariates  in  the  same  
*
o Covariates   order  as  the  volumes  
• Name  <-­‐X    Specify  Text  (e.g.,  
o Masking   “age”)  
 Threshold   Masking      Absolute   • Interactions    None  
  [specify  value  (e.g.,  “0.1”)]   • Centering    No  centering  
 Implicit  Mask    Yes  
 Explicit  Mask    <None>  
o Global  Calculation    Omit  

o Global  Normalization  
 Overall  grand  mean  scaling    No  
o Normalization    None  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [this  will  create  an  “SPM.mat”  file  in  your  working  directory]  
 
 
 
 

  11  
  
USING  THE  FULL  FACTORIAL  MODEL  (FOR  A  2X2  ANOVA)  
 
 
SPM  menu    Specify  2nd-­‐level  
  Parameters:  
o Directory  <-­‐X    Select  Files    [select  the  working  directory  for  your  analysis]    Done  
o Design    “Full  Factorial”  
 Factors    “New:  Factor;  New:  Factor”  
Factor  
- Name    [specify  text  (e.g.,  ”sex”)]  
- Levels    2  
- Independence    Yes  
- Variance    Equal  or  Unequal  
- Grand  mean  scaling    No  
- ANCOVA    No  
Factor  
- Name    [specify  text  (e.g.,  “handedness”)]  
- Levels    2  
- Independence    Yes  
- Variance    Equal  or  Unequal  
- Grand  mean  scaling    No  
- ANCOVA    No  
 Specify  Cells    “New:  Cell;  New:  Cell;  New:  Cell;  New:  Cell”  
Cell  
- Levels    [specify  text  (e.g.,  “1  1”)]  
- Scans      [select   files   (e.g.,   the   smoothed   GM   volumes   of   the   left-­‐handed  
males)]  
Cell  
- Levels    [specify  text  (e.g.,  “1  2”)]  
- Scans      [select   files   (e.g.,   the   smoothed   GM   volumes   of   the   right-­‐handed  
males)]  
Cell  
- Levels    [specify  text  (e.g.,  “2  1”)]  
- Scans      [select   files   (e.g.,   the   smoothed   GM   volumes   of   the   left-­‐handed  
females)]  
Cell  
- Levels    [specify  text  (e.g.,  “2  2”)]  
- Scans      [select   files   e.g.,   the   smoothed   GM   volumes   of   the   right-­‐handed  
females)]  
o Covariates*  
o Masking  
 Threshold  Masking    Absolute    [specify  value  (e.g.,  “0.1”)]  
  12  
  Implicit  Mask    Yes  
 Explicit  Mask    <None>  
o Global  Calculation    Omit  
o Global  Normalization  
 Overall  grand  mean  scaling    No  
o Normalization    None  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [this  will  create  an  “SPM.mat”  file  in  your  working  directory]  
 
 
MULTIPLE  REGRESSION  (CORRELATION)  
 
 
SPM  menu    Specify  2nd-­‐level  
  Parameters:  
o Directory  <-­‐X    Select  Files    [select  the  directory  for  your  analysis]    Done  
o Design      “Multiple  Regression”  
 Scans    [select  files  (e.g.,  the  smoothed  GM  volumes  of  all  subjects)]    Done  
Covariates    “New:  Covariate”  

Covariate  
- Vector    [enter  the  values  in  the  same  order  as  the  respective  file  names  of  
the  smoothed  GM  images]  
- Name    [specify  test  (e.g,  “age”)]  
- Centering    No  centering  
- Intercept    Include  Intercept  
o Covariates*  
o Masking  
 Threshold  Masking    Absolute    [specify  value  (e.g.,  “0.1”)]  
 Implicit  Mask    Yes  
 Explicit  Mask    <None>  
o Global  Calculation    Omit  
o Global  Normalization  
 Overall  grand  mean  scaling    No  
o Normalization    None  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [This  will  create  an  “SPM.mat”  file  in  your  selected  folder]  

  13  
  
USING  THE  FULL  FACTORIAL  MODEL  (FOR  AN  INTERACTION)  
 
 
SPM  menu    Specify  2nd-­‐level  
  Parameters:  
o Directory  <-­‐X    Select  Files    [select  the  working  directory  for  your  analysis]    Done  
o Design    “Full  Factorial”  
 Factors    “New:  Factor”  
Factor  
- Name    [specify  text  (e.g.,  ”sex”)]  
- Levels    2  
- Independence    Yes  
- Variance    Equal  or  Unequal  
- Grand  mean  scaling    No  
- ANCOVA    No  
 Specify  Cells    “New:  Cell;  New:  Cell”  
Cell  
- Levels    [specify  text  (e.g.,  “1”)]  
- Scans    [select  files  (e.g.,  the  smoothed  GM  volumes  of  the  males)]  
Cell  
- Levels    [specify  text  (e.g.,  “2”)]  
- Scans    [select  files  (e.g.,  the  smoothed  GM  volumes  of  the  females)]  
o Covariates    “New:  Covariate”  
 Covariate  
- Vector    [enter  the  values  in  the  same  order  as  the  respective  file  names  of  
the  smoothed  GM  images]  
- Name    [specify  test  (e.g,  “age”)]  
- Interactions    With  Factor  1  
- Centering    No  centering  
o Masking  
 Threshold  Masking    Absolute    [specify  value  (e.g.,  “0.1”)]  
 Implicit  Mask    Yes  
 Explicit  Mask    <None>  
o Global  Calculation    Omit  
o Global  Normalization  
 Overall  grand  mean  scaling    No  
o Normalization    None  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [this  will  create  an  “SPM.mat”  file  in  your  working  directory]  

  14  
  
ESTIMATING  THE  STATISTICAL  MODEL  
 
 
SPM  menu    Estimate  
  Parameters:  
o Select  SPM.mat  <-­‐X    Select  Files    [select  the  SPM.mat  which  you  just  built]    Done  
o Method    “Classical”  
- File    Save  Batch    
- File    Run  Batch  [This  will  create  an  “SPM.mat”  file  in  your  selected  folder]  
 
 
 
 
 
 
 
 
DEFINING  CONTRASTS  
 
 
 
SPM   menu      Results      [select   the   SPM.mat   file]      Done   (this   opens   the   Contrast   Manager)     
Define  new  contrast  (i.e.,  choose  “t-­‐contrast”  or  “F-­‐contrast”;  type  the  contrast  name  and  specify  the  
contrast  by  typing  the  respective  numbers,  as  shown  below):  
T-­‐contrasts:  
a. Simple  group  difference  
  ⇒  Use  SPM.mat  from  model  “2  sample  T-­‐test”  
• For  Group  A  >  Group  B:  specify  “1  -­‐1”  
• For  Group  A  <  Group  B:  specify  “-­‐1  1”  
 
b. 2x2  ANOVA  
  ⇒  Use  SPM.mat  from  model  “2X2  ANOVA”  
• For  left-­‐handed  males  >  right-­‐handed  males:  specify  “1  -­‐1  0  0”  
• For  left-­‐handed  females  >  right-­‐handed  females:  specify  “0  0  1  -­‐1”  
• For  left-­‐handed  males  >  left-­‐handed  females:  specify  “1  0  -­‐1  0”  
• For  right-­‐handed  males  >  right-­‐handed  females:  specify  “0  1  0  -­‐1”  
etc.  
• For  males  >  females:  specify  “1  1  -­‐1  -­‐1”  
• For  left-­‐handers  >  right-­‐handers:  specify  “1  -­‐1  1  -­‐1”  
 

  15  
 c. Multiple  Regression  (Correlation)  
  ⇒  Use  SPM.mat  from  model  “CORRELATION”  
• For  positive  correlation:  specify  “1”  
• For  negative  correlation:  specify  “-­‐1”  
d. Interaction  
  ⇒  Use  SPM.mat  from  model  “INTERACTION”  
• For  regression  slope  Group  A  >  Group  B:  specify  “0  0  1  -­‐1”  
• For  regression  slope  Group  A  <  Group  B:  specify  “0  0  -­‐1  1”  
    Done  
F-­‐contrasts:  
If  you  would  like  to  use  the  old  SPM2  F-­‐contrast  “Effects  of  interest”  the  respective  contrast  vector  is:  
eye(n)-­‐1/n  
where  n  is  the  number  of  columns  of  interest.  This  F-­‐contrast  is  often  helpful  for  plotting  parameter  
estimates  of  effects  of  interest.  
 
Getting  Results:  
SPM  menu    Results    [select  a  contrast  from  Contrast  Manager]    Done  
• Mask  with  other  contrasts    No  
• Title  for  comparison:  [use  the  pre-­‐defined  name  from  the  Contrast  Manager  or  change  it]  
• P  value  adjustment  to:  
o None  (uncorrected  for  multiple  comparisons),  set  threshold  to  0.001  
o FDR  (false  discovery  rate),  set  threshold  to  0.05,  etc.  
o FWE  (family-­‐wise  error),  set  threshold  to  0.05,  etc.  
• Extent  threshold:  [either  use  “none”  or  specify  the  number  of  voxels2)  
 
 

                                                                                                           
2
 In  order  to  empirically  determine  the  extent  threshold  (rather  than  saying  100  voxels  or  500  voxels,  which  is  completely  
arbritrary),  simply  run  this  first  without  specifying  an  extent  threshold.  This  will  give  you  an  output  (i.e.,  the  standard  SPM  
glass  brain  with  significant  effects).  When  you  click  “Table”  (SPM  main  menu)  you  will  get  a  table  with  all  relevant  values  
(MNI   coordinates,   p-­‐values,   cluster   size   etc).   Below   the   table   you   will   find   additional   information,   such   as   “Expected  
Number  of  Voxels  per  Cluster”.  Remember  this  number  (this  is  your  empirically  determined  extent  threshold).  Re-­‐run  SPM  
  Results  etc.  and  specify  this  number  when  asked  for  the  “Extent  Threshold”.  There   is   also   a   hidden   option   in   “VBM8    
Data  presentation    Threshold  and  transform  spmT-­‐maps”  to  define  the  extent  threshold  in  terms  of  a  p-­‐value  or  to  use  
the  “Expected  Number  of  Voxels  per  Cluster”.  

  16  
Special  Cases  
 

VBM8  for  longitudinal  data  

 
BACKGROUND  
The   majority   of   VBM   studies   are   based   on   cross-­‐sectional   data,   where   one   image   is   acquired   for   each  
subject.  However,  in  order  to  track  e.g.  learning  effects  over  time  longitudinal  designs  are  necessary,  
where   additional   time-­‐points   are   acquired   for   each   subject.   The   analysis   of   these   longitudinal   data  
requires  a  customized  processing,  that  considers  the  characteristics  of  intra-­‐subject  analysis.  While  for  
cross-­‐sectional   data   images   can   be   processed   independently   for   each   subject   longitudinal   data   has   to  
be   registered   to   the   baseline   image   (or   mean   image)   for   each   subject.   Furthermore,   spatial  
normalization  is  estimated  for  the  baseline  image  only  and  applied  to  all  images  (Figure  4).  Additional  
attention   is   then   needed   for   the   setup   of   the   statistical   model.   The   following   section   will   therefore  
describe  data  preprocessing  and  model  setup  for  longitudinal  data.  
 

 
 
Fig   4.:   Flow   diagram   for   processing   longitudinal   data   with   VBM8.   This   figure   demonstrates    
the   steps   for   processing   longitudinal   data.   After   an   initial   realignment,   the   mean   of   the    
realigned   images   is   calculated   (mean)   and   used   as   reference   image   in   a   subsequent    
realignment.     The   realigned   images   (rix)   are   then   corrected   for   signal   inhomogeneities   with    
regard   to   the   reference   mean   image.   Spatial   normalization   parameters   are   estimated   in    
the  next  step  using  the  segmentations  of  the  mean  image.  These  normalization  parameters    
are   applied   to   the   segmentations   of   the   bias-­‐corrected   images   (p1mrix).   The   resulting    
normalized  segmentations  (wp1mrix)  are  finally  again  realigned.    

  17  
Preprocessing  of  longitudinal  data  -­‐  overview  
The   VBM8   Toolbox   supplies   a   batch   for   longitudinal   study   design.   Here,   for   each   subject   the  
respective  images  need  to  be  selected.  Intra-­‐subject  realignment,  bias  correction,  segmentation,  and  
normalization   are   calculated   automatically.   Preprocessed   images   are   written   as   wp1mr*   and   wp2mr*  
for   grey   and   white   matter   respectively.   To   define   the   segmentation   and   normalization   parameters,  
the  defaults  in  cg_vbm8_defaults.m  are  used.  
 
 
 
CHANGE  SETTINGS  FOR  PREPROCESSING  
 
 
Change  your  working  directory  to  “/toolbox/vbm8”  in  your  SPM  directory:    
  select  “Utilities    cd”  in  the  SPM  menu  and  change  to  the  respective  folder.  
Then  type  “open  cg_vbm8_defaults.m”  in  your  matlab  command  window.  The  file  will  open  in  the  
editor.  If  you  are  unsure  how  to  change  the  values,  open  the  module  “estimate  and  write”  in  the  
batch  editor  for  reference.  
 
The  most  interesting  parameters  to  change  here  may  be:  
• DARTEL  or  spm8  default  normalization:  
  Line  65    set  to  “0”  for  spm8  default  normalization  
 
• The  Tissue  Probability  Maps:  
Line  16    replace  “{fullfile(spm('dir'),'toolbox','Seg','TPM.nii')}”  with  the  path  to  your  Maps.  
 
It  is  always  a  great  idea  to  memorize  or  record  the  changes  and  change  all  values  back  to  default  after  
the  Analysis.  
 
 
 
PREPROCESSING  OF  LONGITUDINAL  DATA  
 
 
VBM8    Process  longitudinal  data  
  Parameters:  
o Data  <-­‐X    New:  Subject  Subject  Longitudinal  data  for  one  subject  Select  Files  
  [select  raw  data]    Done  
- Select  all  volumes  for  each  subject.  As  the  Toolbox  does  not  support  multispectral  
data   yet   (i.e.,   different   imaging   methods   for   the   same   brain,   such   as   T1-­‐,   T2-­‐,  
diffusion-­‐weighted   or   CT   images),   it   is   recommended   to   choose   a   T1-­‐weighted  
image.  
- Select  “New:  Subject”  to  add  data  for  a  new  subject  
The  data  for  each  subject  should  be  listed  as  one  “subject”  in  the  Batch  Editor,  i.e.  
there  are  as  many  subjects  listed  as  included  in  the  analysis.  

  18  
  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [this  will  create  an  “SPM.mat”  file  in  your  working  directory]  
For  the  naming  conventions  of  all  written  files  see  p.29  “Naming  convention  of  output  files”.  The  final  
GM  segments  are  wp1mr*,  the  final  WM  segments  are  named  wp2mr*.  
 
Statistical  analysis  of  longitudinal  data  -­‐  overview  
The  main  interest  in  longitudinal  studies  is  the  common  change  of  tissue  volume  over  time  in  a  group  
of   subjects   or   the   difference   in   these   changes   between   two   or   more   groups.   The   setup   of   the  
statistical  model  needed  to  assess  these  questions  will  be  described  on  two  examples.  First,  the  case  
of   only   one   group   of   4   subjects   with   2   time   points   each   (e.g.   normal   aging)   is   presented.  
Subsequently,   the   case   of   two   groups   of   subjects   with   4   time   points   per   subject   will   be   described.  
These   examples   should   cover   most   analyses   –   the   number   of   time   points   /   groups   just   have   to   be  
adapted.  
 
 
STATISTICAL  ANALYSIS  OF  LONGITUDINAL  DATA  IN  ONE  GROUP  
 
 
SPM  menu    Specify  2nd-­‐level  
  Parameters:  
o Directory  <-­‐X    Select  Files    [select  the  working  directory  for  your  analysis]    Done  
o Design    “Flexible  Factorial”  
 Factors    “New:  Factor;  New:  Factor”  
Factor  
- Name    [specify  text  (e.g.,  ”subject”)]  
- Independence    Yes  
- Variance    Equal  or  Unequal  
- Grand  mean  scaling    No  
- ANCOVA    No  
Factor  
- Name    [specify  text  (e.g.,  “time”)]  
- Independence    No  
- Variance    Equal  or  Unequal  
- Grand  mean  scaling    No  
- ANCOVA    No  
 Specify   Subjects   or   all   Scans   &   Factors      “Subjects”      “New:   Subject;   New:  
Subject;  New:  Subject;  New:  Subject;”  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  first  Subject)]  
- Conditions    “1  2”  [for  two  time  points]  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  second  Subject)]  
  19  
 - Conditions    “1  2”  [for  two  time  points]  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  third  Subject)]  
- Conditions    “1  2”  [for  two  time  points]  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  fourth  Subject)]  
- Conditions    “1  2”  [for  two  time  points]  
 Main  effects  &  Interaction    “New:  Main  effect”  
Main  effect  
- Factor  number    2  
Main  effect  
- Factor  number    1  
o Covariates  (see  “basic  VBM  analysis  (detailed  description)”)  
o Masking  
 Threshold  Masking    Absolute    [specify  value  (e.g.,  “0.1”)]  
 Implicit  Mask    Yes  
 Explicit  Mask    <None>  
o Global  Calculation    Omit  
o Global  Normalization  
 Overall  grand  mean  scaling    No  
o Normalization    None  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [this  will  create  an  “SPM.mat”  file  in  your  working  directory]  
 
 
STATISTICAL  ANALYSIS  OF  LONGITUDINAL  DATA  IN  TWO  GROUPS  
 
 
 
SPM  menu    Specify  2nd-­‐level  
  Parameters:  
o Directory  <-­‐X    Select  Files    [select  the  working  directory  for  your  analysis]    Done  
o Design    “Flexible  Factorial”  
 Factors    “New:  Factor;  New:  Factor;  New:  Factor”  
Factor  
- Name    [specify  text  (e.g.,  ”subject”)]  
- Independence    Yes  
- Variance    Equal  
- Grand  mean  scaling    No  

  20  
 - ANCOVA    No  
Factor  
- Name    [specify  text  (e.g.,  ”group”)]  
- Independence    Yes  
- Variance    Unequal  
- Grand  mean  scaling    No  
- ANCOVA    No  
Factor  
- Name    [specify  text  (e.g.,  “time”)]  
- Independence    No  
- Variance    Equal  
- Grand  mean  scaling    No  
- ANCOVA    No  
 Specify   Subjects   or   all   Scans   &   Factors      “Subjects”      “New:   Subject;   New:  
Subject;  New:  Subject;  New:  Subject;”  
Subject  
- Scans      [select   files   (the   smoothed   GM   volumes   of   the   1st   Subject   of   first  
group)]  
- Conditions    “  [1  1  1  1;  1  2  3  4]’  “  [first  group  with  four  time  points]  
            Do  not  forget  the  additional  single  quote!  
Subject  
- Scans      [select   files   (the   smoothed   GM   volumes   of   the   2nd   Subject   of   first  
group)]  
- Conditions    “  [1  1  1  1;  1  2  3  4]’  “  [first  group  with  four  time  points]  
Subject  
- Scans      [select   files   (the   smoothed   GM   volumes   of   the   3rd   Subject   of   first  
group)]  
- Conditions    “  [1  1  1  1;  1  2  3  4]’  “  [first  group  with  four  time  points]  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  4th  Subject  of  second  
group)]  
- Conditions    “  [2  2  2  2;  1  2  3  4]’  “  [second  group  with  four  time  points]  
Subject  
- Scans      [select   files   (the   smoothed   GM   volumes   of   the   1st   Subject   of   second  
group)]  
- Conditions    “  [2  2  2  2;  1  2  3  4]’  “  [second  group  with  four  time  points]  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  2nd  Subject  of  second  
group)]  
- Conditions    “  [2  2  2  2;  1  2  3  4]’  “  [second  group  with  four  time  points]  
Subject  
- Scans    [select  files  (the  smoothed  GM  volumes  of  the  3rd  Subject  of  second  
group)]  
- Conditions    “  [2  2  2  2;  1  2  3  4]’  ”  [second  group  with  four  time  points]  
Subject  

  21  
 - Scans    [select  files  (the  smoothed  GM  volumes  of  the  4th  Subject  of  second  
group)]  
- Conditions    “  [2  2  2  2;  1  2  3  4]’  ”  [second  group  with  four  time  points]  
 Main  effects  &  Interaction    “New:  Interaction;  New:  Main  effect”  
Interaction  
- Factor  numbers    2  3  [Interaction  between  group  and  time]  
Main  effect  
- Factor  number    1  
o Covariates  (see  “basic  VBM  analysis  (detailed  description)”)  
o Masking  
 Threshold  Masking    Absolute    [specify  value  (e.g.,  “0.1”)]  
 Implicit  Mask    Yes  
 Explicit  Mask    <None>  
o Global  Calculation    Omit  
o Global  Normalization  
 Overall  grand  mean  scaling    No  
o Normalization    None  
- File    Save  Batch  [this  will  save  the  parameters  as  *.m  script  file]  
- File    Run  Batch  [this  will  create  an  “SPM.mat”  file  in  your  working  directory]  
 
 

  22  
Altered  Workflows  for  VBM-­‐analyses  
Background  
For   most   analyses   the   VBM   Toolbox   will   supply   all   tools   needed.   That   is,   as   the   new   segmentation  
algorithm  is  not  dependent  on  Tissue  Probability  Maps  (TPMs)  anymore  and  as  predefined  DARTEL-­‐
templates   for   healthy   adults   exist,   most   questions   can   be   assessed   using   the   toolbox   settings.  
However,   for   some   special   cases   such   as   analyses   in   children   or   special   patient   populations,   the  
toolbox  settings  might  not  be  optimal.  For  these  cases  the  VBM8  Toolbox  provides  an  integration  into  
the   SPM8   environment   that   can   be   used   to   optimize   the   preprocessing.   In   the   following,   we   will  
present  strategies  how  to  deal  with  these  special  cases.  
 
 
Standard  VBM  preprocessing:  Input,  Output  and  where  to  modify  
The   first   module   of   the   VBM8   Toolbox   (“Estimate   and   Write”)   processes   all   preprocessing   steps  
except   for   the   smoothing.   Basically,   it   takes   structural   volumes   and   TPMs   as   input.   It   will   then  
segment  the  data,  apply  a  registration  to  MNI  Space  (either  rigid  or  affine)  and  subsequently  a  non-­‐
linear  deformation.  The  non-­‐linear  deformation  parameters  can  be  calculated  via  the  low  dimensional  
SPM   default   approach   or   the   high   dimensional   DARTEL   algorithm   and   the   predefined   templates.  
Figure  5  depicts  this  preprocessing  workflow  and  highlights  possibilities  where  to  modify.  
 

 
 
  Fig.  5:  Flow-­‐chart  of  the  preprocessing  steps  within  the  module  “Estimate  and  Write”.  
  Marked  in  red  are  those  steps,  where  the  preprocessing  can  be  customized.  Per  
  default,  the  built-­‐in  DARTEL  normalization  works  with  the  VBM8  DARTEL  templates  of  
  550  healthy  adult  control  subjects.  Affine  registered  tissue  segments  can  be  used  to  
  create  customized  DARTEL-­‐templates,  which  can  then  be  used  to  replace  the  default  
  DARTEL  template.  

  23  
Adapting  the  workflows  
 
Customized  Tissue  Probability  Maps  -­‐  overview  
For  data  on  children  it  will  be  a  good  idea  to  create  customized  TPMs,  which  reflect  age  and  gender  of  
the   population.   The   TOM8   Toolbox   (available   via:   https://irc.cchmc.org/software/tom.php)   provides  
the   means   to   customize   these   TPMs.   To   learn   more   about   the   TOM   toolbox,   see   also  
http://dbm.neuro.uni-­‐jena.de/software/tom/.  
 
 
 
CUSTOMIZED  TISSUE  PROBABILITY  MAPS  
 
 
About  the  TOM8  Toolbox:  
  select  Module  “create  new  template”  
    select  “TOM.mat”  (you  will  have  to  download  this  file  together  with  the  toolbox)  
    write  priors/template  as  single  file  
  for  all  others  use  default  settings  or  modify.  For  “Age”  either  a  vector  or  a  mean  age  
(when  using  the  average  approach)  must  be  specified.  

Implementation  into  VBM8:  

Module:  “Estimate  and  write”    
”Estimation  Options”.    
  Tissue  Probability  Maps  (  Select  your  customized  TPMs  here)  
 
 
 
 
 
Customized  DARTEL-­‐template  -­‐  overview  
For  all  cases  that  include  at  least  50-­‐100  subjects  a  customized  DARTEL  template  can  be  created.  That  
is,  grey  matter  and  white  matter  tissue  segments  of  all  subjects  are  used  to  create  a  mean  template  of  
the   study   sample.   As   the   VBM8   toolbox   writes   all   files   needed   to   create   these   templates   (“DARTEL  
export”),   this   requires   only   two   additional   steps.   In   order   to   use   these   newly   created   DARTEL-­‐
Templates  with  the  vbm8  Toolbox,  an  affine  registration  of  the  DARTEL  export  should  be  used.  From  
these   affine   registered   segments   customized   DARTEL   templates   can   then   be   created   and   used   with  
the  vbm8  module  “Write  already  estimated  segmentations”.    
 
 
 
  24  
  
CUSTOMIZED  DARTEL-­‐TEMPLATE  
 
 
Three   steps   are   needed   to   create   normalized   tissue   segments   with   customized   DARTEL   Templates.  
These  steps  can  be  batched  using  dependencies  in  the  Batch  Editor.  The  last  step  can  be  rerun  using  
the   customized   templates,   if   additional   output   files   are   needed.   In   the   first   step   the   T1   images   are  
segmented,   and   the   tissue   segments   normalized   to   the   Tissue   Probability   Maps   using   an   affine  
transformation.  Start  with  selecting  the  module  “Estimate  and  Write”.  
 
 
Module:  “Estimate  and  Write”:  
    for  all  options  except  “writing  options”  use  settings  like  for  a  “standard”  VBM  analysis.  
”Writing  Options”  
      “Grey  Matter”  “DARTEL  export”    “affine”  
      “White  Matter”  “DARTEL  export”    “affine”  
 
 
These   settings   will   produce   the   volumes   “rp1*-­‐affine.nii”   and   “rp2*-­‐affine.nii”,   which   are   the   grey  
(rp1)   and   white   (rp2)   matter   segments   after   affine   registration.   The   following   modules   can   be   chosen  
directly  in  the  batch  editor  (SPM    Tools    DARTEL  Tools    Run  DARTEL  (create  Templates),  and  
SPM    Tools    VBM8    VBM8:  Write  already  estimated  segmentations).  It  makes  sense  to  add  and  
specify   these   modules   together   with   the   “Estimate   and   Write”   module   within   the   Batch   Editor   and   to  
set  dependencies.  
 
 
Module  “Run  DARTEL  (create  Templates)”  
    Images    select  two  times  “new:  Images”  
      Images:    select  the  “rp1*-­‐affine.nii”  files  or  create  a  dependency.  
      Images:    select  the  “rp2*-­‐affine.nii”  files  or  create  a  dependency.  
    all  other  options:  use  defaults  or  modify  
 
The  thus  created  customized  DARTEL  Templates  are  now  used  in  the  vbm8  Toolbox:  
 
Module  “Write  already  estimated  segmentations”  
  Parameters:  
o Volumes   <-­‐X      Select   the   original   T1   images   like   in   the   first   module   “Estimate   and  
Write”.  

  25  
 o Extended  Options    “High-­‐dimensional:DARTEL”  (default)    Select  DARTEL  Template  
- if  you  use  dependencies  in  the  Batch  Editor  create  a  dependency  on  the  first  
Template  in  the  batch  editor.  Templates  2-­‐6  will  be  used  automatically.  
- If  you  have  already  created  your  DARTEL  Templates,  chose  the  fist  Template  
(named   “Template_1”   per   default).   Attention:   If   you   want   to   rename   your  
template,   you   may   either   add   a   prefix   to   the   default   name,   or   make   sure   that  
the   template   number   is   embedded   in   the   name   between   two   underscores  
(e.g.  “abc_1_xyz”).  
- For  all  other  options  use  the  same  settings  as  in  the  first  module,  or  modify.  
o Writing  Options    select  the  output  files  just  like  in  any  standard  VBM  analysis:  
- For  GM,  WM,  and  CSF  image  volumes  see  page  14:  “Additional  Information  
on  native,  normalized  and  modulated  normalized  volumes”.  
- A  bias  corrected  image  volume,  in  which  MRI  inhomogeneities  and  noise  are  
removed,  can  be  written  in  normalized  or  native  space.  
- A   partial   volume   effect   (PVE)   label   image   volume   can   also   be   written   in  
normalized  or  native  space  or  as  a  DARTEL  export  file.  
- The  Jacobian  determinant  for  each  voxel  can  be  written  in  normalized  space.    
- Finally,   deformation   fields   can   be   written.   This   option   is   useful   to   re-­‐apply  
normalization  parameters  to  other  images  or  particular  regions  of  interest.    
 
 
- File    Save  Batch  
 
-­‐   File    Run  Batch  
 
 
How  to  proceed:  
All  steps  described  above  are  just  an  adaption  of  the  VBM8  Toolbox  module  “Estimate  and  Write”.  A  
subsequent  smoothing  is  necessary  before  statistics  can  be  calculated.  As  always  it  is  a  good  idea  to  
save  the  applied  modules  and  to  perform  quality  control.  Here,  the  modules  “Display  one  slice  for  all  
images”  and  “Check  sample  homogeneity  using  covariance”  from  the  VBM8  Toolbox  will  be  helpful.  
 

  26  
Additional  information  on  native,  normalized  and  modulated  volumes  
When  preprocessing  the  images  (see  “First  Module:  Estimate  and  write”,  on  pages  5-­‐6),  the  decision  
about  the  normalization  parameters  will  determine  the  interpretation  of  the  analysis  outcomes:  
 
“Native  space”  produces  tissue  class  images  in  spatial  correspondence  to  the  original  data.  Although  
this   could   be   useful   for   estimating   global   tissue   volumes   (e.g.,   GM+WM+CSF=TBV)   it   is   not   suitable   to  
conduct  VBM  analyses  due  to  the  missing  voxel-­‐wise  correspondence  across  brains).  Of  note,  if  one  is  
interested   in   these   global   tissue   volumes   in   native   space   (“raw   values”),   it   won’t   be   necessary   to  
actually   output   the   tissue   class   images   in   native   space.   The   “Estimate   and   write”-­‐function  
automatically   generates   a   text   file   for   each   subject   (*_seg8.txt),   which   contains   the   raw   values   for  
GM,   WM,   and   CSF.   The   subject-­‐specific   values   can   be   combined   (i.e.,   integrated   into   a   single   text   file)  
by   using   the   function   “Calculate   raw   volumes   for   GM/WM/CSF”:   VBM8      Tools      Calculate   raw  
volumes  for  GM/WM/CSF  
 
“Normalized”  produces  tissue  class  images  in  spatial  correspondence  to  the  template.  This  is  useful  
for  VBM  analyses  (e.g.,  “concentration”  of  gray  matter;  Good  et  al.  2001;  Neuroimage).  
 
“Modulated  normalized”  gives  two  different  options:  
• Affine+non-­‐linear   produces   tissue   class   images   in   alignment   with   the   template,   but   multiplies  
(“modulates”)   the   voxel   values   by   the   Jacobian   determinant   (i.e.,   linear   and   non-­‐linear  
components)   derived   from   the   spatial   normalization.   This   is   useful   for   VBM   analyses   and   allows  
comparing   the   absolute   amount   of   tissue   (e.g.,   “volume”   of   gray   matter;   Good   et   al.   2001;  
Neuroimage).  
• Non-­‐linear   only   produces   tissue   class   images   in   alignment   with   the   template,   but   multiplies   the  
voxel   values   by   the   non-­‐linear   components   only.   This   is   useful   for   VBM   analyses   and   allows  
comparing  the  absolute  amount  of  tissue  corrected  for  individual  brain  sizes.  Of  note,  this  option  
is   similar   to   using   “Affine+non-­‐linear”   (see   above)   in   combination   with   “global   normalization”  
(when   later   building   the   statistical   model   using   the   traditional   PET   designs).   That   is,   when  
building   the   statistics,   one   would   specify   “Global   normalization”      “Overall   grand   mean   scaling  
–   No”      “Normalization   –   Proportional”.   It   is   also   similar   to   using   “Affine+non-­‐linear”   (see  
above)  and  including  the  numeric  brain  volume  (as  given  in  the  *_seg8.txt  file  for  each  subject)  
as   covariate   (when   later   building   the   statistical   model).   Although   all   3   approaches   allow  
comparing   tissue   volumes   while   correcting   for   individual   brain   size,   it   is   recommended   to   use  
the  option  “non-­‐linear  only”  as  it  applies  the  correction  directly  to  the  data,  rather  than  to  the  
statistical  model.    
 
For  further  explanation  see  also:    
  http://dbm.neuro.uni-­‐jena.de/vbm/segmentation/modulation  
 
 
 

  27  
Naming  convention  of  output  files  
 
 
segmented  images:   bias  corrected  images:  
m[0]w[r]p[0123]*   wm[r]*  
m   -­‐  modulated   m   -­‐  bias  corrected  
m0   -­‐  modulated  non-­‐linear  only   w   -­‐  warped  
w   -­‐  warped   r   -­‐  dartel  warped    
r     -­‐  dartel  warped      
p   -­‐  segmented   estimated  raw  volumes:  
0   -­‐  PVE  label   pxxx_seg.txt  
1   -­‐  GM    
2   -­‐  WM  
3   -­‐  CSF  
_affine  -­‐  affine  registered  only  
 
 
 
 
For  longitudinal  data  (default  settings):  
 
segmented  images:   bias  corrected  images:   mean   images   (useful   for  
wp[12]mr*   mr*   overlaying  results):  
w   -­‐  warped   m   -­‐  bias  corrected   wmrmean*  
p   -­‐  segmented   r   -­‐  realigned     w   -­‐  warped  
1   -­‐  GM     m   -­‐  bias  corrected  
2   -­‐  WM     r   -­‐  realigned  
m   -­‐  bias  corrected   estimated  raw  volumes:   mean   -­‐  mean  
r   -­‐  realigned   pmrxxx_seg.txt    
  estimated  raw  volumes  (mean):  
pmeanxxx_seg.txt  
 
 

  28  
Technical  information  
This  toolbox  is  an  extension  of  the  “New  Segment  Toolbox”  in  SPM8,  but  uses  a  completely  different  
segmentation  approach.3    
 
1. The  segmentation  approach  is  based  on  an  adaptive  Maximum  A  Posterior  (MAP)  technique  
without   the   need   for   a   priori   information   about   tissue   probabilities.   That   is,   the   Tissue  
Probability   Maps   are   not   used   constantly   in   the   sense   of   the   classical   unified   segmentation  
approach,  but  just  for  spatial  normalization3.  The  following  MAP  estimation  is  adaptive  in  the  
sense  that  local  variations  of  the  parameters  (i.e.,  means  and  variance)  are  modelled  as  slowly  
varying   spatial   functions   (Rajapakse   et   al.   1997).   This   not   only   accounts   for   intensity  
inhomogeneities  but  also  for  other  local  variations  of  intensity.    
2. Additionally,   the   segmentation   approach   uses   a   Partial   Volume   Estimation   (PVE)   with   a  
simplified  mixed  model  of  at  most  two  tissue  types  (Tohka  et  al.  2004).  We  start  with  an  initial  
segmentation   into   three   pure   classes:   gray   matter   (GM),   white   matter   (WM),   and  
cerebrospinal   fluid   (CSF)   based   on   the   above   described   MAP   estimation.   The   initial  
segmentation  is  followed  by  a  PVE  of  two  additional  mixed  classes:  GM-­‐WM  and  GM-­‐CSF.  This  
results  in  an  estimation  of  the  amount  (or  fraction)  of  each  pure  tissue  type  present  in  every  
voxel   (as   single   voxels   –   given   by   their   size   –   probably   contain   more   than   one   tissue   type)   and  
thus  provides  a  more  accurate  segmentation.  
3. Furthermore,  we  apply  two  denoising  methods.  The  first  method  is  a  spatially  adaptive  non-­‐
local  means  (SANLM)  denoising  filter  (Manjon  et  al.  2010).  This  filter  will  remove  noise  while  
preserving  edges  and  is  implemented  as  preprocessing  step.  The  second  method  is  a  classical  
Markov   Random   Field   (MRF)   approach,   which   incorporates   spatial   prior   information   of  
adjacent  voxels  into  the  segmentation  estimation  (Rajapakse  et  al.  1997).  
4. Another   important   extension   to   the   SPM8   segmentation   is   the   integration   of   the   Dartel  
normalisation  (Ashburner  2007)  into  the  toolbox.  If  high-­‐dimensional  spatial  normalisation  is  
chosen,   an   already   existing   Dartel   template   in   MNI   space   will   be   used.   This   template   was  
derived   from   550   healthy   control   subjects   of   the   IXI-­‐database   (http://www.brain-­‐
development.org)   and   is   provided   in   MNI   space4   for   six   different   iteration   steps   of   Dartel  
normalisation.   Thus,   for   the   majority   of   studies   the   creation   of   sample-­‐specific   Dartel  
templates  is  not  necessary  anymore5.    
 

                                                                                                           
3
 The  classic  SPM8  segmentation  is  still  used  in  addition,  but  only  to  initially  remove  non-­‐brain  tissue  from  the  image.  
4
 Thus,  no  additional  MNI  normalization  is  necessary.  
5
  For   studies   investigating   data   of   children   I   still   recommend   creating   a   customized   Dartel   template.   Of   note,   for   this  
option   a   representative   sample   with   a   sufficient   number   of   subjects   is   required   (n>50-­‐100).   Alternatively,   if   a   sufficient  
sample   size   cannot   be   achieved,   the   low-­‐dimensional   SPM8   normalization   approach   combined   with   customized   Tissue  
Probability  Maps  (e.g.  from  the  TOM8  toolbox)  can  be  selected.  

  29  
References  
 
Good,  C.D.  et  al.  (2001):  A  voxel-­‐based  morphometric  study  of  ageing  in  465  normal  adult  human  brains.  Neuroimage.  
14(1  Pt  1):21-­‐36.  
Rajapakse,  J.C.  et  al.  (1997):  Statistical  Approach  to  Segmentation  of  Single-­‐Channel  Cerebral  MR  Images.  IEEE  Trans.  Med.  
Imag.  16(2):176-­‐186.    
Tohka,   J.   et   al.   (2004):   Fast   and   robust   parameter   estimation   for   statistical   partial   volume   models   in   brain   MRI.  
Neuroimage  23(1):84-­‐97.  
Ashburner,  J.  (2007):  A  fast  diffeomorphic  image  registration  algorithm.  Neuroimage  38(1):95-­‐113.  
Manjon,  J.V.  et  al.  (2010).  Adaptive  Non-­‐Local  Means  Denoising  of  MR  Images  With  Spatially  Varying  Noise  Levels  Journal  
of  Magnetic  Resonance  Imaging,  31:  192-­‐203.  
 
 
2010/04/04             Florian  Kurth  [email protected]    
            Eileen  Luders  [email protected]  
            Christian  Gaser  christian.gaser@uni-­‐jena.de  

  30