Prediction of myometrial invasion of bladder cancer based on texture analysis of the bladder wall at tumor base
10.3760/cma.j.cn131148-20220506-00323
- VernacularTitle:基于肿瘤基底部膀胱壁纹理分析预测膀胱癌肌层浸润性
- Author:
Chunlan QIN
1
;
Yong GAO
;
Xinhong LIAO
Author Information
1. 广西医科大学第一附属医院超声科，530021 南宁
- Keywords:
Texture analysis;
Ultrasonic omics;
Bladder wall;
Bladder cancer
- From:
Chinese Journal of Ultrasonography
2023;32(1):73-78
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To identify the value of ultrasound radiomic features extracted from the bladder wall at tumor base in predicting myometrial invasion of bladder cancer.Methods:A total of 175 cases with bladder cancer confirmed by pathology from January 2017 to February 2022 in the First Affiliated Hospital of Guangxi Medical University were retrospectively analyzed. They were divided into training set and testing set in a ratio of 7∶3. The MaZda texture analysis software was used to draw the region of interest (ROI) of the bladder wall and the tumor region for extracting texture features. The minimum absolute reduction and variable selection operator (LASSO) regression and 10-fold cross-validation were used to screen the features of training set for establishing the models. And the ROC curve was used to evaluate the efficiency of the models.Results:A total of 279 texture features were extracted from the ROI of the bladder wall and the tumor region, and 5 texture features were screened out for constructing omics scoring models by LASSO regression and 10-fold cross-test. The area under ROC curve (AUC)s used in training set and testing set of the bladder wall were 0.921 and 0.856, while the AUCs applied in training set and testing set of the tumor region were 0.849 and 0.704. Both in the training set and test set, the AUCs of the model of the bladder wall were higher than those of the model of the tumor region (all P<0.05). Conclusions:The omics scoring model based on the texture features of the bladder wall at tumor base can effectively identify muscle-invasive bladder cancer(MIBC) and non-muscle-invasive bladder cancer(NMIBC), and has better performance than the model based on the texture feature of the tumor region.
