Premium
MRI‐Based Radiomics: Associations With the Recurrence‐Free Survival of Patients With Hepatocellular Carcinoma Treated With Conventional Transcatheter Arterial Chemoembolization
Author(s) -
Song Wenlong,
Yu Xiangling,
Guo Dajing,
Liu Huan,
Tang Zhuoyue,
Liu Xinjie,
Zhou Jun,
Zhang Haiping,
Liu Yangyang,
Liu Xi
Publication year - 2020
Publication title -
journal of magnetic resonance imaging
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.563
H-Index - 160
eISSN - 1522-2586
pISSN - 1053-1807
DOI - 10.1002/jmri.26977
Subject(s) - medicine , nomogram , proportional hazards model , hepatocellular carcinoma , transcatheter arterial chemoembolization , univariate , radiology , hazard ratio , multivariate statistics , lasso (programming language) , radiomics , oncology , nuclear medicine , confidence interval , statistics , mathematics , world wide web , computer science
Background Preoperative estimation of hepatocellular carcinoma (HCC) recurrence after conventional transcatheter arterial chemoembolization (c‐TACE) is crucial for subsequent follow‐up and therapy decisions. Purpose To evaluate the associations of radiomics models based on pretreatment contrast‐enhanced MRI, a clinical‐radiological model and a combined model with the recurrence‐free survival (RFS) of patients with HCC after c‐TACE, and to develop a radiomics nomogram for individual RFS estimations and risk stratification. Study Type Retrospective. Population In all, 184 consecutive HCC patients. Field Strength/Sequence 1.5T or 3.0T, including T 2 WI, T 1 WI, and contrast‐enhanced T 1 WI. Assessment All HCC patients were randomly divided into the training ( n = 110) and validation datasets ( n = 74). Radiomics signatures capturing intratumoral and peritumoral expansion (1, 3, and 5 mm) were constructed, and the radiomics models were set up using least absolute shrinkage and selection operator (LASSO) Cox regression. Clinical‐radiological features were identified by univariate and multivariate Cox regression. The clinical‐radiological model and the combined model fusing the radiomics signature with the clinical‐radiological risk factors were developed by a multivariate Cox proportional hazard model. A radiomics nomogram derived from the combined model was established. Statistical Tests LASSO Cox regression, univariate and multivariate Cox regression, Kaplan–Meier analysis were performed. The discrimination performance of each model was quantified by the C‐index. Results Among the different peritumoral expansion models, only the 3‐mm peritumoral expansion model (C‐index, 0.714) showed a comparable performance ( P = 0.4087) to that of the portal venous phase intratumoral model (C‐index, 0.727). The combined model showed the best performance and the C‐index was 0.802. Kaplan–Meier analysis showed that the cutoff values of the combined model relative to a median value (1.7426) perfectly stratified these patients into high‐risk and low‐risk subgroups. Data Conclusion The combined model is more valuable than the clinical‐radiological model or radiomics model alone for evaluating the RFS of HCC patients after c‐TACE, and the radiomics nomogram can be used to preoperatively and individually estimate RFS. Level of Evidence : 3 Technical Efficacy Stage : 4 J. Magn. Reson. Imaging 2020;52:461–473.