Automatic MRI Brain Segmentation:
Problems

These web pages describe various difficulties seen in segmenting MRI brain scans. These initial descriptions were written by Andy Worth (sorry it's incomplete). Your contributions are welcome and encouraged!

Definition of Segmentation

Here I want to stress that ``Segmentation'' is not only the decision to place each voxel into a general tissue class, but also the labeling of each voxel as belonging to a specific neuroanatomical structure. As described below, a major difficulty with this task is that there is often not enough information in the image to completely segment a structure.

Problems

Using automated routines to segment MRI brain images is difficult because of the following problems:

  • The brain must be located,
  • Variable imaging parameters,
  • Overlapping intensities,
  • Noise,
  • Partial voluming,
  • Gradients,
  • Motion,
  • Echos,
  • Many edges are not sharp,
  • Normal anatomical variation,
  • Non-intensity based borders,
  • Susceptibility Artifact.

    Go [back up], [Credits]
    Please email Andy if you have any questions, comments, suggestions regarding this discussion.