Premium
Automated Analysis of Multiparametric Magnetic Resonance Imaging/Magnetic Resonance Elastography Exams for Prediction of Nonalcoholic Steatohepatitis
Author(s) -
Dzyubak Bogdan,
Li Jiahui,
Chen Jie,
Mara Kristin C.,
Therneau Terry M.,
Venkatesh Sudhakar K.,
Ehman Richard L.,
Allen Alina M.,
Yin Meng
Publication year - 2021
Publication title -
journal of magnetic resonance imaging
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.563
H-Index - 160
eISSN - 1522-2586
pISSN - 1053-1807
DOI - 10.1002/jmri.27549
Subject(s) - magnetic resonance elastography , nonalcoholic fatty liver disease , magnetic resonance imaging , receiver operating characteristic , elastography , medicine , liver biopsy , nonalcoholic steatohepatitis , steatohepatitis , fatty liver , logistic regression , population , artificial intelligence , computer science , nuclear medicine , radiology , biopsy , pathology , ultrasound , disease , environmental health
Background Nonalcoholic fatty liver disease (NAFLD) affects 25% of the global population. The standard of diagnosis, biopsy, is invasive and affected by sampling error and inter‐reader variability. We hypothesized that widely available rapid MRI techniques could be used to predict nonalcoholic steatohepatitis (NASH) noninvasively by measuring liver stiffness, with magnetic resonance elastography (MRE), and liver fat, with chemical shift‐encoded (CSE) MRI. Besides, we validate an automated image analysis technique to maximize the utility of these methods. Purpose To implement and test an automated system for analyzing CSE‐MRI and MRE data coupled with model‐based prediction of NASH. Study Type Prospective. Subjects Eighty‐three patients with suspected NAFLD. Field Strength/Sequence A 1.5 T using a flow‐compensated motion‐encoded gradient echo MRE sequence and a multiecho CSE‐MRI sequence. Assessments The MRE and CSE‐MRI data were analyzed by two readers (5+ and 1 years of experience) and an automated algorithm. A logistic regression model to predict pathology‐diagnosed NASH was trained based on stiffness and proton density fat fraction, and the area under the receiver operating characteristic curve (AUROC) was calculated using 10‐fold cross validation for models based on both automated and manual measurements. A separate model was trained to predict the NASH severity score (NAS). Statistical Tests Pearson's correlation, Bland–Altman, AUROC, C‐statistic. Results The agreement between automated measurements and the more experienced reader ( R 2 = 0.87 for stiffness and R 2 = 0.99 for proton density fat fraction [PDFF]) was slightly better than the agreement between readers ( R 2 = 0.85 and 0.98). The model for predicting biopsy‐diagnosed NASH had an AUROC of 0.87. The NAS‐prediction model had a C‐statistic of 0.85. Data Conclusion We demonstrated a workflow that used a limited MRI acquisition protocol and fully automated analysis to predict NASH with high accuracy. These methods show promise to provide a reliable noninvasive alternative to biopsy for NASH‐screening in populations with NAFLD. Level of Evidence 2 Technical Efficacy Stage 2