The Value of Ensemble Learning Model Based on Conventional Non-Contrast MRI in the Pathological Grading of Cervical Cancer

作者全名："He, Zhimin; Lv, Fajin; Li, Chengwei; Liu, Yang; Xiao, Zhibo"

作者地址："[He, Zhimin; Lv, Fajin; Li, Chengwei; Liu, Yang; Xiao, Zhibo] Chongqing Med Univ, Dept Radiol, Affiliated Hosp 1, Chongqing 400016, Peoples R China; [He, Zhimin] First Hosp Putian City, Dept Radiol, Putian 351100, Fujian, Peoples R China"

通信作者："Xiao, ZB (通讯作者)，Chongqing Med Univ, Dept Radiol, Affiliated Hosp 1, Chongqing 400016, Peoples R China."

来源："COMPUTATIONAL MATHEMATICS MODELING IN CANCER ANALYSIS, CMMCA 2023"

ESI学科分类：　

WOS号：WOS:001116900900004

JCR分区：　

影响因子：　

年份：2023

卷号：14243

期号：　

开始页：31

结束页：41

文献类型：Proceedings Paper

关键词：Cervical cancer; Grade; Radiomics; Ensemble learning; Magnetic resonance imaging

摘要："Purpose: To investigate the value of an stacking ensemble learning model based on conventional non-enhanced MRI sequences in the pathological grading of cervical cancer. Methods: We retrospectively included 98 patients with cervical cancer (54 well/moderately differentiated and 44 poorly differentiated). Radiomics features were extracted from T2WI Axi and T2WI Sag. Feature selection was performed by intra-class correlation coefficients (ICC), t-test, least absolute shrinkage and selection operator (LASSO). Logistic Regression (LR), Support Vector Machine (SVM), k-Nearest Neighbor (kNN), and Extreme Gradient Boosting (XGB) were used as the first-layer base classifier, and LR as the second-layer meta-classifier in stacking ensemble learning model. The model performance was evaluated by the area under the curve (AUC) and accuracy. Results: In the basic classifiers, the XGB model showed the best performance, the average AUC was 0.74(0.69,0.76) and the accuracy was 0.73. It was followed by SVM, LR and KNN models, and the average AUC were 0.73(0.66,0.80), 0.71(0.62,0.78) and 0.66(0.61,0.72), respectively. The performance of stacking ensemble model showed effective improvement, with an average AUC of 0.77(0.67,0.84), and the accuracy was 0.83. Conclusions: The ensemble learning model based on conventional non-enhanced MRI sequences could identify poorly differentiated cervical cancer from well/moderately differentiated cervical cancer, and can provide more references for preoperative non-invasive assessment of cervical cancer."

基金机构：　

基金资助正文：