Deep learning network with Euclidean similarity factor for Brain MR Tumor segmentation and volume estimation

Abstract

The major goal of this paper is to isolate tumor region from nontumor regions and the estimation of tumor volume. Accurate segmentation is not an easy task due to the varying size, shape and location of the tumor. After segmentation, volume estimation is necessary in order to accurately estimate the tumor volume. By exactly estimating the volume of abnormal tissue, physicians can do excellent prognosis, clinical planning and dosage estimation. This paper describes a new Euclidean Similarity factor (ESF) based active contour model with deep learning for segmenting the tumor region into complete, core and enhanced tumor portions. Initially, the ESF considers the spatial distances and intensity differences of the region automatically to detect the tumor region. It preserves the image details but removes the noisy details. Then, the 3D Convolutional Neural Network (3D CNN) segments the tumor by automatically extracting spatiotemporal features. Finally, the extended shoelace method estimates the volume of the tumor accurately for [Formula: see text]-sided polygons. The simulation result achieves a high accuracy of 92% and Jaccard index of 0.912 and computes the tumor volume with effective performance than existing approaches.