A Deep Learning-Based Approach for Accurate Segmentation of Bladder Wall using MR Images

Abstract

In this paper, a deep learning-based convolution neural network (CNN) is developed for accurate segmentation of the bladder wall using T2-weighted magnetic resonance imaging (T2W-MRI). Our framework utilizes a dual pathway, two-dimensional CNN for pathological bladder segmentation. Due to large bladder shape variability across subjects and the existence of pathology, a learnable adaptive shape prior (ASP) model is incorporated into our framework. To obtain the goal regions, the neural network fuses the MR image data for the first pathway, and the estimated ASP model for the second pathway. To remove noisy and scattered predictions, the CNN soft output is refined using a fully connected conditional random field (CRF). Our pipeline has been tested and evaluated using a leave-one-subject-out approach (LOSO) on twenty MRI data sets. Our framework achieved accurate segmentation results for the bladder wall and tumor as documented by the Dice similarity coefficient (DSC) and Hausdorff distance (HD). Moreover, comparative results against other segmentation approaches documented the superiority of our framework to provide accurate results for pathological bladder wall segmentation.