A Cad System For Accurate Diagnosis Of Bladder Cancer Staging Using A Multiparametric MRI

Abstract

In this paper, a computer-aided diagnostic (CAD) system is developed using a multiparametric magnetic resonance imaging (MPMRI) (T2-MRI and DW-MRI) to differentiate between BC staging, especially T1 and T2. The segmentation of the bladder wall (BW) and the localization of the whole BC area (A t ) and its extent inside the wall (A w ) is first performed. Secondly, a set of functional, texture, and morphological features are estimated. Due to the massive difference between the wall and bladder lumen cells, A t is split into nested equidistance contours (i.e., iso-contours), and features are estimated for each iso-contours. The functional features are based on the cumulative distribution function (CDF) statistical measures for the estimated apparent diffusion coefficient (ADC) from DWMRI. Texture features, namely radiomic features, are derived from T2W-MRI, A t both carcinoma intensity and gradient images for each iso-contours from A t , T2-MRI. Besides, morphological features are also incorporated to describe the tumors' geometric from T2W-MRI, A w . Finally, the estimated iso-features are augmented and used to train and test neural networks classifier as well as a statistical machine learning (ML) classifier. The system has been tested using a leave-one-subject-out approach on 42 data sets. The overall accuracy, sensitivity, specificity, and area under the curve (AUC) of the receiver operating characteristics (ROC) are 92.86%, 97.05%, 100%, and 0.9705, respectively. We introduce the diagnostic accuracy of individual MRI modality for our proposal to highlight the advantage of fusion multiparametric iso-features that is confirmed by the ROC analysis. Furthermore, the accuracy of two different techniques statistical ML classifiers (i.e., random forest (RF) and support vector machine (SVM)) and end-to-end convolution neural networks (i.e., ResNet50) is compared against our pipeline. Keywords: Support vector machines ,Sensitivity, Magnetic resonance imaging , Neural networks , Receivers , Bladder ,Proposals.