actually I need thickness data out of it. Is .mgh format documented somewhere?

On Thu, 24 Apr 2008, Pádraig Kitterick wrote:

How much detail do you need frmo the mgh file? I wrote a quick Python script which reads in at least the volume data but I think I just passed over most of the other stuff in there...

Yaroslav Halchenko wrote:

...

Does anyone know any handy module to at least read in .mgh files from Python? Or should I simply try mri_convert them into .minc and use netcdf module within python?

...

Thanks in advance for the ideas!

Yaroslav,

The mgh format is pseudo-documented here:

https://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/MghFormat

But the thickness data is not stored in mgh format, but rather in another proprietary format, with documentation here:

http://wideman-one.com/gw/brain/fs/surfacefileformats.htm

Your best option is to look at the matlab scripts in the $FREESURFER_HOME/matlab directory. In particular, read_curv.m, which I have attached to the email (along with worker routine fread3.m).

Krish Subramaniam, a new software engineer here at MGH, cc'd in this email, has just started a project where some of the major IO routines will be written in python, so maybe you can coordinate with him.

Nick

On Thu, 2008-04-24 at 11:49 -0400, Yaroslav Halchenko wrote:

actually I need thickness data out of it. Is .mgh format documented somewhere?

On Thu, 24 Apr 2008, Pádraig Kitterick wrote:

...

How much detail do you need frmo the mgh file? I wrote a quick Python script which reads in at least the volume data but I think I just passed over most of the other stuff in there...

Yaroslav Halchenko wrote:

...

Does anyone know any handy module to at least read in .mgh files from Python? Or should I simply try mri_convert them into .minc and use netcdf module within python?

...

...

Thanks in advance for the ideas!

Pádraig is correct in stating that thickness data can be stored in an mgh file. But the default output of the freesurer recon-all stream does not store them in that format: it will output ?h.thickness files, which are the proprietary format.

However, those can be converted to .mgh like this:

mris_convert -c lh.thickness lh.inflated lh.thickness.mgh

or to ascii like this:

mris_convert -c lh.thickness lh.inflated lh.thickness.asc

But I doubt that helps too much, since I'm guessing you want to avoid having to use the freesurfer binaries and want to just read the files directly.

Nick

On Thu, 2008-04-24 at 17:09 +0100, Pádraig Kitterick wrote:

MGH is just a generic volume-based format so I think thickness data will be stored in a 1xNumVerticesx1x1 volume. Here is a python function which reads in the volume data. It's basically a translation of the readmgh function from the Freesurfer matlab toolbox. You can modify it to return more parts of the file data if necessary...

---

import struct import numpy

def read_mgh(filename): fp = open(filename,'rb') intsize = struct.calcsize('>i') shortsize = struct.calcsize('>h') floatsize = struct.calcsize('>f') charsize = struct.calcsize('>b')

 v = struct.unpack('>i',fp.read(intsize))[0]
 ndim1 = struct.unpack('>i',fp.read(intsize))[0]
 ndim2 = struct.unpack('>i',fp.read(intsize))[0]
 ndim3 = struct.unpack('>i',fp.read(intsize))[0]
 nframes = struct.unpack('>i',fp.read(intsize))[0]
 vtype = struct.unpack('>i',fp.read(intsize))[0]
 dof = struct.unpack('>i',fp.read(intsize))[0]

 UNUSED_SPACE_SIZE = 256
 USED_SPACE_SIZE = (3*4) + (4*3*4) # space for ras transform
 unused_space_size = UNUSED_SPACE_SIZE - 2

 ras_good_flag = struct.unpack('>h',fp.read(shortsize))[0]
 if ras_good_flag:
     # We read these in but don't process them
     # as we just want to move to the volume data
     delta = struct.unpack('>fff',fp.read(floatsize*3))
     Mdc = struct.unpack('>fffffffff',fp.read(floatsize*9))
     Pxyz_c = struct.unpack('>fff',fp.read(floatsize*3))

     unused_space_size = unused_space_size - USED_SPACE_SIZE

 for i in range(unused_space_size):
     struct.unpack('>b',fp.read(charsize))[0]

 nv = ndim1 * ndim2 * ndim3 * nframes
 vol = 

numpy.fromstring(fp.read(floatsize*nv),dtype=numpy.float32).byteswap()

 nvert = max([ndim1,ndim2,ndim3])
 vol = numpy.reshape(vol,(ndim1,ndim2,ndim3,nframes),order='F')
 vol = numpy.squeeze(vol)
 fp.close()

 return vol

---

Padraig.

Yaroslav Halchenko wrote:

...

actually I need thickness data out of it. Is .mgh format documented somewhere?

On Thu, 24 Apr 2008, Pádraig Kitterick wrote:

...

How much detail do you need frmo the mgh file? I wrote a quick Python script which reads in at least the volume data but I think I just passed over most of the other stuff in there...

Yaroslav Halchenko wrote:

...

Does anyone know any handy module to at least read in .mgh files from Python? Or should I simply try mri_convert them into .minc and use netcdf module within python?

...

...

Thanks in advance for the ideas!

Hi Yaroslav and Pádraig,

Please follow the link pasted below to get an alpha version of pymgh. It requires the latest Python and Numpy.

https://www.nmr.mgh.harvard.edu/facility/filedrop/showgroup/10224/1/2a5da2a1... (expires in 30 days )

Make sure PYTHONPATH has the directory which contains 'pymgh'. Found by typing,

...

...

import sys sys.path

As far as the usage goes, pymgh basically instantiates a "MGH" class ( which contains details about the header as well as the data ) when you call the load() method. There is a class save() method as well as a module save() method ( static function) The following examples will throw more light.

from pymgh import mgh m = mgh.MGH() m.load('test.mgz') # this loads information from test.mgz onto m. Both .mgh and .mgz work

m.header # see the header information

m.vol # the N-D numpy volume of the same datatype as mri_type

v = m.vol * 3 # some basic manipulation

m.save(filename = 'modtest.mgz', volume = v ) # save the new volume with same headers onto a different file

There is also a module method mgh.save() which requires all keyword arguments. More documentation on functions are found in their docstrings and I am afraid it's the only documentation available now. pymgh mostly follows the structure in the wiki page describing MGHformat. And the resulting Numpy volume has the exact shape as the MATLAB volume loaded by load_mgh.m

This package is fresh, so expect surprises. I request you to go to the tests/ directory and run the test_mgh_io.py before everything. It has a few basic sanity tests.

cd tests/ python test_mgh_io.py -v 2

Contact me ( krish @ nmr... ) if you want to report a bug / suggestions etc.

Thanks Krish

On Apr 24, 2008, at 1:23 PM, Yaroslav Halchenko wrote:

...

well, I can leave with 1 time conversion ;) Having direct interface from python is a neat thing to have though. by ldd on mris_convert and its size (3MB) I guess freesurfer statically compiles in IO functionality in. It would be great if there just was a dynamic library which does IO. Then it could be quite easy to create python wrapper which would use it more or less directly...

...