Biparametric MRI-based peritumoral radiomics for preoperative prediction of extracapsular extension in prostate cancer
10.3760/cma.j.cn112149-20240911-00556
- VernacularTitle:双参数MRI瘤周影像组学对前列腺癌包膜外侵犯的诊断价值
- Author:
Honghao XU
1
;
Qicong DU
;
Yuanhao MA
;
Xueyi NING
;
Baichuan LIU
;
Xu BAI
;
Di CHEN
;
Yun ZHANG
;
Zhe DONG
;
Chuang JIA
;
Xiaojing ZHANG
;
Xiaohui DING
;
Baojun WANG
;
Aitao GUO
;
Jian XUE
;
Xuetao MU
;
Huiyi YE
;
Haiyi WANG
Author Information
1. 解放军总医院第一医学中心放射诊断科，北京 100853
- Publication Type:Journal Article
- Keywords:
Prostate neoplasms;
Magnetic resonance imaging;
Extraprostatic extension;
Radiomics
- From:
Chinese Journal of Radiology
2025;59(9):1055-1062
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of biparametric-MRI (bpMRI) based peritumoral radiomics for preoperative prediction of extraprostatic extension (EPE) in prostate cancer (PCa).Methods:In this cross-sectional study, consecutive bpMRI of patients undergoing prostatectomy for PCa were retrospectively collected from the First Medical Center (center 1) and the Third Medical Center (center 2) of Chinese PLA General Hospital. A total of 274 patients were finally enrolled. Patients at center 1 from January 2020 to December 2022 were randomly divided into a training set (149 cases) and an internal validation set (63 cases) by stratified random sampling. Patients at center 2 from January 2023 to March 2024 were assigned to the external test set (62 cases). Patients were categorized into EPE-positive group and EPE-negative group according to pathological assessment postoperatively. In the training set, there were 49 cases in EPE-positive group and 100 cases in EPE-negative group. In the internal validation set, there were 26 cases in EPE-positive group and 37 cases in EPE-negative group. In the external test set, there were 22 cases in EPE-positive group and 40 cases in EPE-negative group. Axial T 2WI and apparent diffusion coefficient (ADC) images were manually annotated to obtain index lesion regions of interest (ROIs), with the peritumoral ROIs subsequently delineated by semi-automatic segmentation technique. Radiomics features were extracted from intra-tumoral, peri-tumoral, and intra-tumoral plus peri-tumoral ROIs. The training set data was employed to select and optimize features to build the radiomics models. The logistic regression analysis was used to develop radiomics, clinical, and integrated models. The predictive performance was assessed by the area under the receiver operating characteristic curve (AUC) in the external test set, and compared by the DeLong test. The sensitivity and specificity were compared by the exact McNemar test. Results:In the external test set, the peri-tumoral radiomics model based on bpMRI showed the highest performance in evaluating EPE, with an AUC of 0.739 (95% CI 0.611-0.842), which was identified as the optimal radiomics model. EPE grade ( OR=6.151, 95% CI 3.371-11.226, P<0.001) was incorporated into the clinical model, with an AUC of 0.780 (95% CI 0.657-0.875) in the external test set. The integrated model had an AUC of 0.817 (95% CI 0.698-0.904) in the external test set. There was no statistically significant difference in comparisons of AUCs among the three models (all P>0.05). The sensitivity of the integrated model (68.2%) showed no significant difference from those of the clinical model and the optimal radiomics model (77.3% and 86.4%, respectively; P=0.500 and P=0.289). However, the specificity of the integrated model (85.0%) was significantly higher than those of the clinical model (67.5%, P=0.016) and the optimal radiomics model (50.0%, P<0.001). Conclusion:A bpMRI-based peritumoral radiomics integrating clinical model demonstrates high performance for preoperative prediction of EPE in PCa.
