Multi-modal Brain Tumor Segmentation Based on Self-organizing Active Contour Model

Rui Liu, Jian Cheng, Xiaoya Zhu, Hao Liang, Zezhou Chen

Abstract: In this paper, an automatic and practical method based on active contour model (ACM) is proposed for multi-modal brain tumor segmentation. Firstly, we construct a concurrent self-organizing map (CSOM) networks. Then, applying the networks into a local region based ACM framework constructs a SOM based ACM, i.e. self-organizing active contour model (SOAC). Finally, by using SOAC, making tumor segmentation problems to be stated as a process of contour evolution. However, the segmentation task cannot be well performed for single-modal MRI images due to intensity similarities between brain normal tissues and lesions. For highlighting different tissues, between normal and abnormal, using multi-modal MRI information is an effective way to improve segmentation accuracy, obviously. Therefore, we introduce a global difference strategy, which creates a series of difference images from multi-modal MRI images, namely global difference images (GDI). By reorganizing MRI images and GDI, we propose an automatic segmentation method for brain tumor region extraction with multi-modal MRI images based on SOAC. The effectiveness of the method is tested on the real data from BRATS2013 and part of BRATS2015.