Assessment of Pathological Grading of Bladder Cancer Using Texture Features from MRI

Abstract

Objective: To explore the value of texture features from MR in differentiating the pathological grade of urothelial carcinoma of the bladder. Methods: A total of 108 lesions from patients with bladder cancer confirmed by postoperative pathology were analyzed retrospectively, including 60 cases of high-grade urothelial carcinoma (HGUG) and 48 cases of low-grade urothelial carcinoma (LGUC). All patients underwent routine MRI examination with the same scanning parameters. Lesions on T2-weighted image were delineated on ITK-Snap software by two radiologists to extract the corresponding volumes of interest(VOI). A.K. (Artificial Intelligence Kit) software was used for texture extraction, and the least absolute shrinkage and selection operator (LASSO) regression was used to select image features with a method of 10 fold cross-validation, and to reduce the dimensionality. The spearman method was used afterwards to condense the image features by removing redundant. Logistic multiple regression correlation analysis was carried out, and the diagnostic efficiency of the characteristic parameters was analyzed by using the area under receiver operating characteristic curve (AUC). Result: 7 optimal features between HGUC group and LGUC group were screened, includeing two histogram features (Quantile0.025, skewness) and five GLCM features (GLCMEntropy_angle45_offset7, GLCMEntropy_angle0_offset7, ShortRunEmphasis_angle45_offset4, LongRunEmphasis_angle0_offset7, LongRunEmphasis_angle45_offset7), the AUC of the training group was 0.955, and the AUC of the testing group was 0. 894, both of which at good diagnostic decision point. Conclusion: Textural features from T2-weighted image can reflect the difference between low- and high-grade bladder cancer effectively.