ucnia-splash-2.jpg
Simulated Brain Tumor MRI Database PDF Print E-mail

Marcel Prastawa, Elizabeth Bullitt, Guido Gerig

Validation of segmentations of brain MRI require the use of a commmon database with known ground truth. For example, the validation database for the normal adult brain MRI provided by BrainWeb. However, there is a lack of ground truth validation data for pathological brain MRI such as those presenting tumor and edema.

We have made available synthetic brain MR images presenting tumor, that is generated by inserting pathology into a healthy brain MRI with known ground truth. The tumor and edema structures are generated following a physical model, which shows deformation due to tumor mass effect and infiltration due to both tumor and edema (if present). The synthetic brain tumor MR images generated using our method shows similar segmentation challenges as real brain tumor MRI.

An overview of the simulation process can be seen in this diagram.

Publications

The database is generated from the normal BrainWeb data described in:

  • Aubert-Broche B, Griffin M, Pike GB, Evans AC, Collins DL. Twenty new digital brain phantoms for creation of validation image data bases. IEEE Trans Med Imaging. 2006 Nov;25(11):1410-6.

  • C.A. Cocosco, V. Kollokian, R.K.-S. Kwan, A.C. Evans. BrainWeb: Online Interface to a 3D MRI Simulated Brain Database. NeuroImage, vol.5, no.4, part 2/4, 1997.

The simulation method used to generate the pathological ground truth and MR images is documented in the following publications:

  • Marcel Prastawa, Elizabeth Bullitt, and Guido Gerig. Simulation of Brain Tumors in MR Images for Evaluation of Segmentation Efficacy. [Draft]

  • Marcel Prastawa, Elizabeth Bullitt, and Guido Gerig. Synthetic Ground Truth for Validation of Brain Tumor MRI Segmentation. Medical Image Computing and Computer Assisted Intervention (MICCAI) 2005. Lecture Notes in Computer Science (LNCS) 3749, Pages 26-33. [PDF | © Springer-Verlag]

Validation Database

Five sample datasets generated using our method are available for download below. Each compressed file contains the synthetic MRI, the probabilistic ground truth, the manually defined initial tumor seed, and the XML parameter file used as inputs to our software for simulating the dataset.

All images are stored using the Meta image file format, which can be read using ITK. In the Meta format, images are stored as a sequence of text header information and raw (gzipped) binary data.

Software

The simulation software that generates pathological ground truth from the adult healthy ground truth (e.g., the BrainWeb data) is available for download for the Linux and Windows platforms. The software requires an input directory that describes a healthy anatomy (anatomical probabilities, mesh, diffusion tensor image, etc). An example input directory that is generated from subject 04 of the BrainWeb Twenty Normals Database is available for download.

The program reads in an XML file that describes the simulation parameters. A tutorial XML file with comments is available as a starting point for generating your own simulations.
[Tutorial XML]

The simulation program is multithreaded and may take up to 3 GB of memory. We recommend running the program on a machine with at least 16 cores and 4 GB of memory.

Acknowledgements

This work was supported by the NIH NIBIB under grant R01 EB000219.