A Radiomic Features–Based Pipeline for Accurate Bladder Cancer Staging

Document Type

Article

Source of Publication

Handbook of Texture Analysis: Generalized Texture for AI-Based Industrial Applications

Publication Date

1-1-2024

Abstract

The bladder cancer (BC) staging is considered the primary factor for determining the manner of the BC treatments, and early diagnosis for BC staging is vital to stop the advancing of BC. In this work, a multiparametric computer-aided diagnostic (MP-CAD) system is introduced to differentiate between BC staging, especially T1 and T1 BC stages, using T2-weighted and diffusion-weighted MRI. Our pipeline sequentially performs three steps. The first step is to use a fully connected convolutional neural network (CNN) for segmenting the bladder wall (BW) and pathology, a region of interest (ROI). The T1 and T2 BC stages’ visual appearances are almost the same. Therefore, the best challenge is to extract characteristic features that improve classification results. Then, the second step is extracting two types of features: radiomic features and descriptors of the pixel-wise apparent diffusion coefficient (ADC). The radiomic features are extracted from pathology using a T2W-MRI image, that is, the first-order statistical (histogram features) is created using the probability distribution function (PDF) and second-order statistical (GLCM). Finally, the discriminatory features are augmented, trained, and tested a support vector machine (SVM). Our pipeline has been tested using a leave-one-subject-out approach on 32 datasets. The overall accuracy, specificity, sensitivity, and area under the curve (AUC) of the receiver operating characteristics (ROC) are 70.43%, 60%, 81.82%, and 0.7182. The ROC curves performance for our approach shows that the fusion model is better than using individual MP-MRI.

ISBN

9780367486099

Publisher

CRC Press

First Page

178

Last Page

196

Disciplines

Medicine and Health Sciences

Keywords

Radiomic features, Bladder cancer, Staging, Pipeline, Medical imaging

Scopus ID

85199035931

Indexed in Scopus

yes

Open Access

no

Link to Full Text

Share

COinS