Brain Tumour Segmentation from Multispectral MR Image Data Using Ensemble Learning Methods

Abstract

The number of medical imaging devices is quickly and steadily rising, generating an increasing amount of image records day by day. The number of qualified human experts able to handle this data cannot follow this trend, so there is a strong need to develop reliable automatic segmentation and decision support algorithms. The Brain Tumor Segmentation Challenge (BraTS), first organized seven years ago, provoked a strong intensification of the development of brain tumor detection and segmentation algorithms. Beside many others, several ensemble learning solutions have been proposed lately to the above mentioned problem. This study presents an evaluation framework developed to evaluate the accuracy and efficiency of these algorithms deployed in brain tumor segmentation, based on the BraTS 2016 train data set. All evaluated algorithms proved suitable to provide acceptable accuracy in segmentation, but random forest was found the best, both in terms of precision and efficiency.