Brain Tumor Segmentation
RESEARCH

Automated brain tumor segmentation. Blue= white matter. Green = grey matter. Gold= csf. Orange= edema. Red= tumor. The approach is performed in 3D. An axial slice is shown here. Supported by NIH-NIBIB R01-EB000219
3D visualization of another automatically segmented brain tumor (blue) shown in juxtapostion with a gadolinium-enhanced MR slice containing a cross-section of the tumor.

Automated tumor segmentation is difficult because one cannot always create a prior model of expected size, shape, location, or image intensity. Moreover, tumors produce "extra" tissue not present in atlases of normal patients, and may additionally change the image intensities of normal anatomical structures via infiltration or edema.

Our group is developing methods of automated brain tumor segmentation from multiple MR sequences. The approach employs

  • Use of multi-channel input
  • Mutual-information-based registration and atlas-based segmentation, with tissue classification into five groups (grey matter, white matter, csf, edema, and tumor)
  • 3D, level-set evolution with smoothness constraints
  • Analysis of brain asymmetry
Another problem is how to best to assess segmentation accuracy. We are developing methods of "inserting" tumors of realistic texture and with associated edema into healthy, multi-modal MR datasets. Since the locations and volumes of tumor and edema are known, these images can serve as a basis against which to evaluate various tumor segmentation methods.
REFERENCES

Prastawa M, Bullitt E, Moon N, Van Leemput K, Gerig G (2003) Automatic brain tumor segmentation by subject specific modification of atlas priors. Acad. Radiol. 10:1341-1348 <pdf>

Joshi S, Lorenzen P, Gerig G, Bullitt E (2003) Structural and Radiometric Asymmetry in Brain Images. MedIA 7:155-170 <pdf>.

Prastawa M, Bullitt E, Ho S, Gerig G (2004) A brain tumor segmentation framework based on outlier detection. MedIA 8:275-283 <pdf>.

Prastawa M, Bullitt E, Gerig G (2005) Synthetic ground truth for validation of brain tumor MRI segmentation. MICCAI 2005; LNCS 3749:26-33.<pdf>

 

updated Dec 2007