UCL logo




 

 

 

How it works

This is an overview of the three stages involved in brain volume estimation using spm_segment. For instructions on using the program, see here.

Lesion Contouring

To accurately estimate brain volume for patients with multiple sclerosis, an image mask of the lesion volume must first be created. At present, no freely available software can accurately identify MS lesions. Hence semi-automated lesion contouring by a skilled operator is required if the lesion volume to be processed correctly. This can be time consuming (20 minutes for a 156 slice axial dataset), however it is a routine part of many MS studies.
Omitting this step will introduce errors in the final volume estimates, as much of the MS lesion volume will be mis-classified as grey matter due to its low signal intensity in a T1 weighed image.

Brain Segmention

The SPM brain segmentation algorithm is used to generate maps representing the probability of a particular voxel belonging to each of the tissue types (GM, WM, CSF). It must be noted that different versions of SPM do not produce identical results. SPM99 has been used by our group for many studies and is well validated. For a fuller discussion, see here. In SPM there are a number of parameters controlling the segmentation. The options used by spm_segment can be found here. These parameters were optimised for our images at 1.5T. For 3T images you might want to change the bias regularisation.

Binary Classification of Tissue

To calculate a volume, binary tissue classification masks are calculated from the SPM probability maps. Provided that the voxel size is small, the number of voxels in each mask, multiplied by the voxel size will give a good estimate of the tissue volumes. Tissue masks are calculated through the following steps:

  • Exclude voxels which lie within the lesion mask.
  • Exclude voxels for which the probability of being GM, WM or CSF is less than a specified threshold.
  • Classify each remaining voxel as belonging to the tissue type with the highest probability.
  • Calculate volumes for GM, WM, CSF and lesions by multiplying the number of voxels in each mask by the voxel size in millilitres

The following options customise the binary classification:

  • 'SPM Version' - see 'Which SPM Version'
  • 'Modality' - MRI modality of input images, default is T1 weighted
  • 'Threshold' - A probability threshold that must be exceeded for classification. Values less than 25% have no effect.
  • 'Use Brainstem Cutoff' - use if non-brain tissue in the neck region is classified as brain, or if the volume of brainstem included varies between segmentations. Typically the difference is less that 1ml and this step is not necessary, see here.

Results

The Results are saved in a file containing volumes in millilitres for GM, WM, CSF and lesion volumes.


Segmented Brain Image

Last modified by Jon Jackson on Mon Dec 8 17:13:19 GMT 2008


NMR Research Unit - Institute of Neurology - Queen Square House - Queen Square - London - WC1N 3BG - UK - Telephone: +44 (0)20 7829 8771 - Fax +44 (0)20 7278 5616 - Copyright © 1999-2009 UCL


Search by Google