Detection of brain tumor invasion within edema using multiparametric imaging and computational intelligence

Todd R Jensen, Marquette University


Accurate discrimination of brain tumor from other tissues using noninvasive imaging is a high priority for the planning of surgical and radiotherapy treatments. Prior to treatment, the area of contrast enhancement on T1 -weighted magnetic resonance (MR) images is typically identified as tumor, and regions of hyperintensity on T2 -weighted or FLAIR MR images outside the tumor area are termed peritumoral edema. Since evidence of tumor invasion in the peritumoral edema has been reported, radiation treatment frequently targets both the tumor and edematous regions but in doing so may also be targeting some areas unnecessarily or suboptimally. Therefore, the objective of this study was to establish the feasibility of computer-aided detection of tumor invasion within edema using morphological, diffusion-weighted, and perfusion-weighted MRI in order to facilitate more effective dose distribution. Classifiers can be trained to recognize specific tissue types based on patterns determined by features such as signal values in MR images. Classifiers of different designs were trained to discriminate tumor and edema, tested on edematous regions of meningiomas and contrast-enhancing areas of glioblastomas, and assessed for their ability to detect nonenhancing tumor invasion. Training data taken from 10 studies included edema voxels from the peritumoral edematous regions of meningiomas and tumor voxels from the contrast-enhancing areas of glioblastomas. Meninigiomas were used as naturally occurring controls because, unlike gliomablastomas, they are typically benign and do not have invasive tumor cells within the peritumoral edema. Features for each voxel included signal values from FLAIR and T1 -weighted images, apparent diffusion coefficient (ADC) derived from diffusion-weighted imaging, mean diffusivity (MD) and fractional anisotropy (FA) derived from diffusion tensor imaging, and relative cerebral blood volume (rCBV) measures calculated from perfusion-weighted imaging. It was found that the combination of features derived from morphological, diffusion-weighted and perfusion-weighted MR imaging methods contained the information necessary for detection of tumor invasion within the peritumoral edema that was not previously identified on morphological, diffusion-weighted, and perfusion-weighted images or feature maps.

Recommended Citation

Jensen, Todd R, "Detection of brain tumor invasion within edema using multiparametric imaging and computational intelligence" (2006). Dissertations (1962 - 2010) Access via Proquest Digital Dissertations. AAI3210965.