Noninvasive hepatic fibrosis staging using mr elastography: The usefulness of the bayesian prediction method

• Published in: Journal of magnetic resonance imaging 2017-08, Vol.46 (2), p.375-382
• Main Authors: Ichikawa, Shintaro, Motosugi, Utaroh, Enomoto, Nobuyuki, Matsuda, Masanori, Onishi, Hiroshi
• Format: Article
• Language: English
• Subjects: Adult; Aged; Aged, 80 and over; Bayes Theorem; Bayesian analysis; Elasticity Imaging Techniques; Female; Fibrosis; hepatic fibrosis; Humans; Image Interpretation, Computer-Assisted; Liver; Liver Cirrhosis - diagnostic imaging; Liver Cirrhosis - pathology; Liver diseases; magnetic resonance elastography; Magnetic Resonance Imaging; Male; Mathematical analysis; Middle Aged; Monte Carlo simulation; NMR; Nuclear magnetic resonance; Patients; posterior probability; Predictive Value of Tests; Probability; Reproducibility of Results; Resonance; Retrospective Studies; ROC Curve; Sensitivity and Specificity; Stiffness; the Bayesian method; Young Adult
• ISSN: 1053-1807; 1522-2586
• DOI: 10.1002/jmri.25551

Purpose To evaluate the usefulness of the Bayesian method for hepatic fibrosis staging with magnetic resonance elastography (MRE). Materials and Methods The sample of this retrospective study comprised patients with chronic liver disease (n = 309), in whom histopathological fibrosis staging and MRE using either a 1.5T (n = 214) or a 3T magnetic resonance imaging (MRI) system (n = 95) had been performed. The optimal cutoff stiffness value was determined and used to calculate the discrimination ability of fibrosis staging by the cutoff method. The Bayesian method calculated post‐MRE probability of each fibrosis stage, yielding MRE‐based fibrosis staging without a cutoff value as well as the confidence of staging. We compared the discrimination ability in all patients and in a subgroup of patients with high (≥90%) posterior probability. Results The discrimination ability for hepatic fibrosis staging was comparable between the Bayesian method and the cutoff method in all patients because the accuracy of staging with the Bayesian method and the cutoff method in all patients was not different (P = 1.0000). However, in patients with high posterior probability by the Bayesian method, the accuracy of staging with the Bayesian method was significantly improved compared with that of the cutoff method in all patients; for discriminating stage ≥F2 from F0–F1 (98.9% vs. 94.8%, P = 0.0069); for ≥F3 (99.6% vs. 92.6%, P < 0.0001); and for F4 (100% vs. 94.2%, P = 0.0002). Conclusion The Bayesian method has a highly accurate discrimination ability for noninvasive hepatic fibrosis staging using MRE, if the posterior probability is high. Level of Evidence: 3 Technical Efficacy: Stage 2 J. MAGN. RESON. IMAGING 2017;46:375–382