A study on histogram normalization for brain tumor segmentation from multispectral MR image data

Győrfi, Ágnes and Karetka-Mezei, Zoltán and Iclanzan, David and Kovács, Levente and Szilágyi, László (2019) A study on histogram normalization for brain tumor segmentation from multispectral MR image data. In: Iberoamerican Congress on Pattern Recognition (CIARP 2019), 28-31 Oct 2019, La Habana, Cuba. (In Press)

Text paper082.pdf Restricted to Registered users only Download (237kB)

Abstract

Absolute values in magnetic resonance image data do not say anything about the investigated tissues. All these numerical values are relative, they depend on the imaging device and they may vary from session to session. Consequently, there is a need for histogram normalization before any other processing is performed on MRI data. The Brain Tumor Segmentation (BraTS) challenge organized yearly since 2012 contributed to the intensification of the focus on tumor segmentation techniques based on multi-spectral MRI data. A large subset of methods developed within the bounds of this challenge declared that they rely on a classical histogram normalization method proposed by Nyúl et al in 2000, which supposed that the corrected histogram of a certain organ composed of normal tissues only should be similar in all patients. However, this classical method did not count with possible lesions that can vary a lot in size, position, and shape. This paper proposes to perform a comparison of three sets of histogram normalization methods deployed in a brain tumor segmentation framework, and formulates recommendations regarding this preprocessing step.

Actions (login required)

Edit Item