Detection of Gliomas in Spinal Cord Using U-Net++ Segmentation with Xg Boost Classification

Abstract

The most prevalent primary brain tumours are gliomas. According to recommendations of the World Health Organization (WHO), they are divided into 4 classes (Grade I-II-III-IV). This paper proposes novel technique in early detection of Gliomas in spinal cord based on segmentation and classification techniques by DL methods. Here input image has been pre-processed for noise removal, image resizing and smoothening of image. Then this processed image has been segmented utilizing U-Net++ architecture in which the skull or vertebral column parts has been segmented and classified using XG_Boost architecture. Our method's effectiveness on a dataset of 3064 MRI image slices from 233 patients that is publically available is compared with previously published classical ML as well as DL techniques. In comparison, our methodology remarkably outperformed the other methods utilising the same database, with a tumour classification accuracy of 0.965.