Segmentation of Whole Tumor Using Localized Active Contour and Trained Neural Network in Boundaries

One of the purposes from segmenting the brain tissues is to separate the damaged tissue in the patient's brain. In fact, brain tissue segmentation is one of the essential steps in the detection and treatment of brain abnormalities. This time-consuming task is usually performed by clinical experts who are not errorless. The proposed method in this paper is to automate the brain tumor segmentation with the aim of making the segmentation process more complete and closer to the clinical treatments. We propose a novel method that is a combination of neural networks and active contours to automatically segment the gliomas in MRI multi-modalities brain images. The proposed algorithm is trained locally by using a neural network at random points in tumor boundary patches, then, by combining the modality of the MRI images and the active contours, the complete tumor is segmented. The obtained results as well as the evaluation criteria such as DICE coefficient, show that the proposed model is highly competitive in comparison with the state of the art segmentation methods.