机构：[1]Department of Radiology, West China Hospital of Sichuan University, Chengdu, China,四川大学华西医院[2]Department of Radiology, Sichuan Cancer Hospital and Institute, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Chengdu, China,四川省人民医院四川省肿瘤医院[3]Department of Research Collaboration, R&D center, Beijing Deepwise & League of PHD Technology Co., Ltd., Beijing, China,[4]Department of Pathology, West China Hospital of Sichuan University, Chengdu, China四川大学华西医院
PurposeTo assess the association of radiomics features based on multiparametric MRI (mpMRI) with the proportion of intraductal carcinoma of prostate (IDC-P) and validate the predictive models. Materials and MethodsWe retrospectively included pre-treatment MR images of prostate cancer (PCa) with IDC components of high proportion (>= 10%, hpIDC-P), low proportion (<10%, lpIDC-P), and pure acinar adenocarcinoma (PAC) from our institution for training and internal validation and cooperated cohort for external validation. Normalized images of T2WI, diffusion weighted imaging (DWI) and apparent diffusion coefficient (ADC) map, and dynamic contrast enhanced (DCE) sequences were used for radiomics modeling. The clinical model was built based on serum total prostate specific antigen (tPSA) and Gleason score (GS), and the integrated model was the combination of Rad-score and clinicopathological data. The discrimination ability was assessed by area under the receiver operating characteristic curve (ROC-AUC) in the internal and external validation sets and compared by DeLong test. ResultsOverall, 97 patients with hpIDC-P, 87 lpIDC-P, and 78 PAC were included for training and internal validation, and 11, 16, and 19 patients for external validation. The integrated model for predicting hpIDC-P got the best ROC-AUC of 0.88 (95%CI = 0.83-0.93) in internal and 0.86 (95%CI = 0.72-1.0) in external validation, which both outperformed clinical models (AUC=0.78, 95% CI = 0.72-0.85, AUC=0.69, 95% CI = 0.5-0.85, respectively) based solely on GS, and the radiomics model (AUC=0.85, 95% CI = 0.79-0.91) was slightly inferior to the integrated model and better than the clinical model in internal dataset. The integrated model for predicting lpIDC-P outperformed both radiomics and clinical models in the internal dataset, while slightly inferior to the integrated model for predicting hpIDC-P. ConclusionsRadiomics signature improved differentiation of both hpIDC-P and lpIDC-P versus PAC when compared with the clinical model based on Gleason score, and was validated in an external cohort.
基金：
This work was supported by the 1*3*5 Project for Disciplines of
Excellence, West China Hospital, Sichuan University (No.
ZY2017304),Science and Technology Support Program of
Sichuan Province (No. 22NSFSC117) and the sky imaging
research foundation (Z-2014-07-1912).
第一作者机构：[1]Department of Radiology, West China Hospital of Sichuan University, Chengdu, China,
共同第一作者：
通讯作者：
推荐引用方式(GB/T 7714)：
Yang Ling,Li Zhengyan,Liang Xu,et al.Radiomic Machine Learning and External Validation Based on 3.0 T mpMRI for Prediction of Intraductal Carcinoma of Prostate With Different Proportion[J].FRONTIERS IN ONCOLOGY.2022,12:doi:10.3389/fonc.2022.934291.
APA：
Yang, Ling,Li, Zhengyan,Liang, Xu,Xu, Jingxu,Cai, Yusen...&Song, Bin.(2022).Radiomic Machine Learning and External Validation Based on 3.0 T mpMRI for Prediction of Intraductal Carcinoma of Prostate With Different Proportion.FRONTIERS IN ONCOLOGY,12,
MLA：
Yang, Ling,et al."Radiomic Machine Learning and External Validation Based on 3.0 T mpMRI for Prediction of Intraductal Carcinoma of Prostate With Different Proportion".FRONTIERS IN ONCOLOGY 12.(2022)
