T1w MRI Brain Tumor Image Segmentation using Active Contour Models and Convolution Neural Network Fusion

Abstract

Early detection of a brain tumour is a challenging task and allows the doctors to reduce the death rate. A brain tumour is first diagnosed by doctors; this process takes more time and needs a lot of experience. The Magnetic resonance imaging (MRI) is primary diagnostic tool to detect the tumour from abnormal brain image at an early stage. An automated segmentation method is introduced in this research article for efficiently segmenting and extracting tumour from MRI images. Pre-processing, clustering, fusion, and segmentation are the four key phase's uses in the proposed method. After pre-processing, the images are clustered by using K-mans and fuzzy c-means (FCM) algorithms. The clustered images are embedded by using the convolution neural network (CNN) data fusion method to obtain high level tumor information of the brain tumor. The fused images are segmented by active contour methods such as chan-vese (C-V) and level set method (LSM) to detect the exact tumor region. All experiments are performed on two brain tumour datasets: BRATS 2017 and Brain Web. The performance of the proposed flow work is measured by various similarity metrics. The Segmentation after fusion gives better results than segmentation alone. Overall, the CNN fusion-based C-V segmentation outperforms than LSM in order to detect the tumor from T1w brain MRI images with minimal loss of information.