BrainWeb: About Data Download Formats

There are two main data formats available for the files downloaded from BrainWeb: MINC and raw. In any case, the data you download is not html, nor an executable, but binary data.

MINC format:

This format was developed here at the McConnell Brain Imaging Center (McBIC) at the Montreal Neurological Institute (MNI).

We make available several powerful packages for working with MINC files:

All these packages are freely available in source code form at http://packages.bic.mni.mcgill.ca. Please note that some of the tools, such as EMMA, are no longer actively supported. Some documentation about these tools is available at the BIC web site.

These tools were originally developed on an SGI/IRIX platform, however they have since been ported to many other UNIX-like systems (especially Linux and Mac OS X). Precompiled binaries for some common platforms are available at this site.

Windows 9x/NT versions of these packages are made available by Frank Morales, however these Windows ports are not supported by our lab.

Raw formats:

This data format is actually very simple: just a long sequence of integers, with no header. The header information is not downloaded, but just displayed at the top of the download page (however, files in MINC format contain both data and header information).

For example, the displayed header info:

    image dimensions: zspace yspace xspace
    dimension name         length         step        start
    --------------         ------         ----        -----
    zspace                     36            5          -72
    yspace                    217            1         -126
    xspace                    181            1          -90
should be interpreted as follows: Using this header info, you should be able to correctly read the raw volumes on any kind of computer platform out there (using C, Matlab, etc).

raw byte (unsigned)

One (unsigned) byte is used for each voxel, and the data is scaled such that it will use the entire 0...255 range of values. Exception: the "crisp" anatomical models are not scaled -- their byte values will be in the range 0...10.

raw short (12 bit)

A short (2 bytes) integer is used for each voxel, and the data is scaled such that it will use the range 0...4095 ("12 bit"). Same note about "crisp" anatomical models as above.

A possible problem with the raw short data format is the 'endianess' of your computer: since each image voxel is represented by a pair of 2 bytes, the order in which your computer stores these bytes in memory is very important. The BrainWeb data is produced on a "big-endian" server (most Unix machines, including Sun-s and SGI-s, are big-endian). If you use the data on an Intel PC (which is "little-endian") then you'll need to swap each pair of bytes to get the data right.