Fully automated brain tumor segmentation using two MRI modalities

Abstract

An algorithm is presented for fully automated brain tumor segmentation from only two magnetic resonance image modalities. The technique is based on three steps: (1) alternating different levels of automatic histogram-based multi-thresholding step, (2) performing an effective and fully automated procedure for skull-stripping by evolving deformable contours, and (3) segmenting both Gross Tumor Volume and edema. The method is tested using 19 hand-segmented real tumors which shows very accurate results in comparison to a very recent method (STS) in terms of the Dice coefficient. Improvements of 5% and 20% respectively for segmentation of edema and Gross Tumor Volume have been recorded. © 2013 Springer-Verlag.