Cascaded 3D V-Net for Fully Automatic Segmentation and Classification of Brain Tumor Using Multi-channel MRI Brain Images

Abstract

Listen

Translate article

Recently, the incidence of Brain tumor (BT) has become highly common worldwide. BT is a life threatening ailment and its early identification is imperative for saving human life. BT segmentation and categorization are crucial tasks involved in BT recognition. Many existing approaches have been developed for detecting BTs. But, the existing models do not focus on segmenting and categorizing diverse categories of BTs. Further, these existing models are incapable of processing 3D images. Hence, to address these requirements, the work proposed a cascaded 3D V-Net framework that aims at segmenting and categorizing three distinct BT categories like non-enhancing and necrotic tumor core (NET/NCT), peritumoral edema (ED) along with enhancing tumor (ET) from 3D magnetic resonance imaging (MRI) brain input images. This work adopts a BRATS 2020 MRI image database for experimentation. The developed 3D V-Net framework’s performance is assessed using an existing framework through considering accuracy, sensitivity, precision and IoU metrics for manifesting the proposed V-Net model’s superiority in BT segmentation and classification performance. And the presented 3D V-Net framework supersedes the existing model framework in BT categorization performance through exhibiting 99.58% accuracy.