MRI-Based Radiomics: Nomograms predicting the short-term response after transcatheter arterial chemoembolization (TACE) in hepatocellular carcinoma patients with diameter less than 5 cm

• Published in: Abdominal imaging Vol. 46; no. 8; pp. 3772 - 3789
• Main Authors: Kuang, Yani, Li, Renzhan, Jia, Peng, Ye, Wenhai, Zhou, Rongzhen, Zhu, Rui, Wang, Jian, Lin, Shuangxiang, Pang, Peipei, Ji, Wenbin
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2021; Springer Nature B.V
• Subjects: Carcinoma, Hepatocellular - diagnostic imaging; Carcinoma, Hepatocellular - therapy; Chemoembolization; Chemoembolization, Therapeutic; Gastroenterology; Hepatobiliary; Hepatocellular carcinoma; Hepatology; Hospitals; Humans; Image contrast; Image enhancement; Imaging; Liver cancer; Liver Neoplasms - diagnostic imaging; Liver Neoplasms - therapy; Magnetic Resonance Imaging; Medical imaging; Medicine; Medicine & Public Health; Nomograms; Patients; Performance prediction; Radiology; Radiomics; Regression analysis; Retrospective Studies; Training; HCC progression; MRI-based nomogram; TACE; Prediction ability; Radiomics
• ISSN: 2366-004X; 2366-0058; 2366-0058
• DOI: 10.1007/s00261-021-02992-2

Purpose To construct MRI radiomics nomograms that can predict short-term response after TACE in HCC patients with diameter less than 5 cm. Methods MRI images and clinical data of 153 cases with tumor diameter less than 5 cm before TACE from 3 hospitals were collected retrospectively and divided into 1 internal training set and 1 external validation set. The T2-weighted imaging (T2WI) and dynamic contrast-enhanced MRI arterial phase (DCE-MR AP) images were studied. Multivariable logistic regression was used to construct Radiomics models, Clinics models, and Nomograms based on T2WI and DCE-MR AP, respectively. The receiver characteristic curve (ROC) was used to evaluate the predictive performance of each model. Results In this study, 113 eligible cases in Hospital 1 were collected as the training set, and 40 eligible cases in other hospitals were used as the verification set. 11 T2WI features and 11 DCE-MRI AP features with the most predictive value were finally screened. 3 models based on T2WI and 3 models based on DCE-MRI AP were established, respectively. The area under curve (AUC) value of Nomogram based on T2WI of training set and validation set was 0.83 and 0.81, respectively. The AUC value of the models based on T2WI and models based on AP was almost equal, and Nomograms were the most effective models among all three types of models. Conclusion MRI-based Nomogram has greater predictive efficacy to predict the response after TACE than Radiomics and Clinics models alone, and the efficacy of T2WI-based models and DCE-MRI AP-based models was almost equal.