Multi-Feature Guided Brain Tumor Segmentation Based on Magnetic Resonance Images

Ye AI  Feng MIAO  Qingmao HU  Weifeng LI  

IEICE TRANSACTIONS on Information and Systems   Vol.E98-D   No.12   pp.2250-2256
Publication Date: 2015/12/01
Publicized: 2015/08/25
Online ISSN: 1745-1361
DOI: 10.1587/transinf.2015EDP7083
Type of Manuscript: PAPER
Category: Pattern Recognition
tumor segmentation,  feature fusion,  graph cut,  MRI,  

Full Text: PDF>>
Buy this Article

In this paper, a novel method of high-grade brain tumor segmentation from multi-sequence magnetic resonance images is presented. Firstly, a Gaussian mixture model (GMM) is introduced to derive an initial posterior probability by fitting the fluid attenuation inversion recovery histogram. Secondly, some grayscale and region properties are extracted from different sequences. Thirdly, grayscale and region characteristics with different weights are proposed to adjust the posterior probability. Finally, a cost function based on the posterior probability and neighborhood information is formulated and optimized via graph cut. Experiment results on a public dataset with 20 high-grade brain tumor patient images show the proposed method could achieve a dice coefficient of 78%, which is higher than the standard graph cut algorithm without a probability-adjusting step or some other cost function-based methods.