Clinic-radiological features and radiomics signatures based on Gd-BOPTA-enhanced MRI for predicting advanced liver fibrosis

• Published in: European radiology 2023-01, Vol.33 (1), p.633-644
• Main Authors: Zheng, Wanjing, Guo, Wei, Xiong, Meilian, Chen, Xiaodan, Gao, Lanmei, Song, Yang, Cao, Dairong
• Format: Article
• Language: English
• Subjects: Chronic infection; Diagnostic Radiology; Fibrosis; Gadolinium; Humans; Image enhancement; Imaging; Internal Medicine; Interventional Radiology; Liver; Liver Cirrhosis - diagnostic imaging; Magnetic Resonance; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Medical diagnosis; Medical imaging; Medicine; Medicine & Public Health; Neuroradiology; Radiology; Radiomics; Retrospective Studies; Signatures; Training; Ultrasound; Widening
• ISSN: 1432-1084; 0938-7994; 1432-1084
• DOI: 10.1007/s00330-022-08992-0

Objectives To develop and validate a combined model based on Gd–BOPTA-enhanced MRI to identify advanced liver fibrosis. Methods A total of 102 patients with chronic HBV infection were divided into a training cohort ( n = 80) and a time-independent testing cohort 1 ( n = 22). In the training cohort, radiomics signatures were extracted from the hepatobiliary phase. Model 1 was constructed with clinic-radiological factors using multivariable logistic regression to predict advanced liver fibrosis, and model 2 incorporated radiomics signatures based on model 1. The diagnostic performances were compared with serum fibrosis tests and FibroScan tests using area under curve (AUC) in testing cohort 1. Another 45 patients with other causes were collected in testing cohort 2 for further validation. Results Model 1 showed age (OR = 1.079) and periportal space widening (OR = 7.838) were the independent factors for predicting advanced fibrosis. After integrating radiomics signatures, model 2 enabled more accurately than model 1 in training cohort (0.940 vs. 0.802, p = 0.003). In testing cohort 1, model 2 demonstrated a superior AUC compared with model 1 (0.900 vs. 0.813， p = 0.131), FibroScan test (0.900 vs. 0.733, p = 0.193), and serum fibrosis tests (APRI and Fib-4 was 0.667 and 0.791). In testing cohort 2, model 2 incorporating radiomics signatures showed satisfactory performance (0.874 vs. 0.757， p = 0.010) compared with model 1. Conclusions Radiomics signatures derived from Gd-BOPTA–enhanced HBP images may offer complementary information to the clinic-radiological model for predicting advanced liver fibrosis. Key Points • Linear or reticular hyperintensity on T2WI, periportal space widening, and diffuse periportal enhancement on HBP can be useful for predicting advanced liver fibrosis. • Clinic-radiological features such as patient age and periportal space widening are the two independent factors predicting advanced fibrosis. • Radiomics signatures derived from Gd-BOPTA-enhanced HBP images offer complementary information to the clinic-radiological model for predicting advanced liver fibrosis.