Predicting Patients With Insufficient Liver Enhancement in the Hepatobiliary Phase Before the Injection of Gadoxetic Acid: A Practical Approach Using the Bayesian Method

J Magn Reson Imaging. 2020 Jan;51(1):62-69. doi: 10.1002/jmri.26760. Epub 2019 Apr 24.


Background: Gadoxetic acid-enhanced hepatobiliary phase (HBP) is useful in liver MRI, but sometimes shows insufficient liver enhancement. There is no established method to predict insufficient liver enhancement before the contrast injection.

Purpose: To reveal the utility of the Bayesian method for predicting patients with insufficient liver enhancement in the gadoxetic acid-enhanced HBP.

Study type: Retrospective.

Subjects: In all, 576 patients with chronic liver disease.

Field strength/sequence: 3T/3D gradient-echo T1 -weighted imaging and MR elastography (MRE).

Assessment: The patients were divided into two groups: insufficient and sufficient liver enhancement in HBP according to the liver-to-portal vein signal intensity ratio. Various parameters, including liver function tests and liver stiffness by MRE, were evaluated as predictors of insufficient liver enhancement.

Statistical tests: We used Chi-square/Student's t-test/logistic regression analysis to determine independent associates, and Bayes' theorem to estimate the probability of insufficient (or sufficient) liver enhancement. The feasibility of Bayesian prediction of insufficient liver enhancement was tested by leave-one-out cross-validation to calculate sensitivity and specificity for single variables and combinations of some variables in all patients and in a subpopulation showing a confidence level of >80%.

Results: Independent associates of insufficient liver enhancement in HBP included: serum albumin (odds ratio [OR] = 4.82, P < 0.001), total bilirubin (OR = 0.30, P < 0.00), platelet count (OR = 1.54, P < 0.00), and liver stiffness by MRE (OR = 0.59, P < 0.00). The accuracy of Bayesian prediction of insufficient liver enhancement was 80.9% (466/576) for the single parameter of albumin and 79.0% (455/576) for total bilirubin, and was increased to 85.2% (487/576) for a combination of albumin, total bilirubin, and liver stiffness. In patients who showed a confidence level of >80%, the accuracy was 89.0% (439/493) for the above combination.

Data conclusion: Bayesian prediction was useful to predict patients with insufficient enhancement by combining serum liver function tests and liver stiffness by MRE.

Level of evidence: 3 Technical Efficacy Stage: 5 J. Magn. Reson. Imaging 2020;51:62-69.

Keywords: cirrhosis; gadoxetate disodium; hepatobiliary phase; liver; magnetic resonance imaging.