An optimized automatic prediction of stage and grade in bladder cancer based on U-ResNet

Article type: Research Article

Authors: Cao, Xin-Zi | Luo, Sheng-Zhou | Li, Jing-Cong | Pan, Jia-Hui^{; *}

Affiliations: School of Software, South China Normal University, Guangzhou, China

Correspondence: [*] Corresponding author. Jia-Hui Pan, School of Software, South China Normal University, Guangzhou, China. E-mail: [email protected].

Abstract: The grade and stage of bladder tumors is an essential key for diagnosing and treating bladder cancer. This study proposed an automated bladder tumor prediction system to automatically assess the bladder tumor grade and stage automatically on Magnetic Resonance Imaging (MRI) images. The system included three modules: tumor segmentation, feature extraction and prediction. We proposed a U-ResNet network that automatically extracts morphological and texture features for detecting tumor regions. These features were used in support vector machine (SVM) classifiers to predict the grade and stage. Our proposed method segmented the tumor area and predicted the grade and stage more accurately compared to different methods in our experiments on MRI images. The accuracy of bladder tumor grade prediction was about 70%, and the accuracy of the data set was about 77.5%. The extensive experiments demonstrated the usefulness and effectiveness of our method.

Keywords: Bladder tumor segmentation, U-ResNet network, grade and stage, feature extraction, support vector machine

DOI: 10.3233/JIFS-210263

Journal: Journal of Intelligent & Fuzzy Systems, vol. 40, no. 6, pp. 12139-12150, 2021