Interactive Semi-automated Method Using Non-negative Matrix Factorization and Level Set Segmentation for the BRATS Challenge

The 2016 BRATS includes imaging data on 191 patients diagnosed with low and high grade gliomas. We present a novel method for multimodal brain segmentation, which consists of (1) an automated, accurate and robust method for image segmentation, combined with (2) semi-automated and interactive multimodal labeling. The image segmentation applies Non-negative Matrix Factorization (NMF), a decomposition technique that reduces the dimensionality of the image by extracting its distinct regions. When combined with the level-set method (LSM), NMF-LSM has proven to be an efficient method for image segmentation. Segmentation of the BRATS images by NMF-LSM is computed by the Cheaha supercomputer at the University of Alabama at Birmingham. The segments of each image are ranked by maximal intensity. The interactive labeling software, which identifies the four targets of the challenge, is semi-automated by cross-referencing the normal segments of the brain across modalities.