Automated Analysis of Multiparametric Magnetic Resonance Imaging/Magnetic Resonance Elastography Exams for Prediction of Nonalcoholic Steatohepatitis.

摘要: Background Nonalcoholic fatty liver disease (NAFLD) affects 25% of the global population. The standard diagnosis, biopsy, is invasive and affected by sampling error inter-reader variability. We hypothesized that widely available rapid MRI techniques could be used to predict nonalcoholic steatohepatitis (NASH) noninvasively measuring stiffness, with magnetic resonance elastography (MRE), fat, chemical shift-encoded (CSE) MRI. Besides, we validate an automated image analysis technique maximize utility these methods. Purpose To implement test system for analyzing CSE-MRI MRE data coupled model-based prediction NASH. Study type Prospective. Subjects Eighty-three patients suspected NAFLD. Field strength/sequence A 1.5 T using a flow-compensated motion-encoded gradient echo sequence multiecho sequence. Assessments were analyzed two readers (5+ 1 years experience) algorithm. logistic regression model pathology-diagnosed NASH was trained based on stiffness proton density fat fraction, area under receiver operating characteristic curve (AUROC) calculated 10-fold cross validation models both manual measurements. separate severity score (NAS). Statistical tests Pearson's correlation, Bland-Altman, AUROC, C-statistic. Results agreement between measurements more experienced reader (R2 = 0.87 R2 0.99 fraction [PDFF]) slightly better than 0.85 0.98). predicting biopsy-diagnosed had AUROC 0.87. NAS-prediction C-statistic 0.85. Data conclusion demonstrated workflow limited acquisition protocol fully high accuracy. These methods show promise provide reliable noninvasive alternative biopsy NASH-screening in populations Level evidence 2 TECHNICAL EFFICACY STAGE: 2.