Predicting gastro-oesophageal variceal bleeding in hepatitis B-related cirrhosis by CT radiomics signature. Issue 12 (December 2019)

• Record Type: Journal Article
• Authors: Yang, J.Q.; Zeng, R.; Cao, J.M.; Wu, C.Q.; Chen, T.W.; Li, R.; Zhang, X.M.; Ou, J.; Li, H.J.; Mu, Q.W.
• Is Part Of: Clinical radiology. Volume 74:Issue 12(2019)
• Page Start: 976.e1
• Page End: 976.e9
• Publication Date: 2019-12
• Subjects: Medical radiology -- Periodicals; Radiotherapy -- Periodicals; Radiotherapy -- Periodicals; Radiology -- Periodicals; Societies, Medical -- Periodicals; Medical radiology; Radiotherapy; Electronic journals; Periodicals; 616.0757
• DOI: 10.1016/j.crad.2019.08.028
• Languages: English
• ISSNs: 0009-9260

Abstract

Abstract : AIM: To develop liver a computed tomography (CT) radiomics model to predict gastro-oesophageal variceal bleeding (GVB) secondary to hepatitis B-related cirrhosis. MATERIALS AND METHODS: Electronic medical records and image data of liver triple-phase contrast-enhanced CT examinations of 295 patients with hepatitis B-related cirrhosis were collected retrospectively from two hospitals. Two hundred and thirty-six and 59 patients were enrolled randomly into the training and validation cohorts, respectively; and 75 in the training cohort and 16 in the validation cohort endured GVB while the others did not during follow-up period. Radiomics features of the liver were extracted from the portal venous phase images, and clinical features came from medical records. The tree-based method and univariate feature selection were used to select useful features. The radiomics model, clinical model, and integration of radiomics and clinical models were built using the useful image features and/or clinical features. Predicting performance of three models was evaluated with the area under receiver-operating characteristic curve (AUC), accuracy, and F-1 score. RESULTS: Twenty-one useful radiomics features and/or three clinical features were selected to build prediction models that correlated with GVB. AUC of integration of radiomics and clinical models was larger than of clinical or radiomics models for the training cohort (0.83±0.09 versus 0.64±0.08 or 0.82±0.10) and the validation cohort (0.64 versus 0.61 or 0.61). Integration of radiomics and clinical models obtained good performance in predicting GVB for both the training and validation cohorts (accuracy: 0.76±0.07 and 0.73, and F-1 score: 0.77±0.09 and 0.72, respectively). CONCLUSION: Integration of the radiomics and clinical models may be a non-invasive method to predict GVB. Highlights: Quantitative methods for predicting gastroesophageal variceal bleeding secondary to cirrhosis are lacking. Multivariate models based on radiomics and clinical features predict gastroesophageal variceal bleeding. Model by integrating radiomics and clinical features better predicts gastroesophageal variceal bleeding.