Meeting Banner
Abstract #2208

Signal Intensity and Texture Feature Analysis in Contrast-Enhanced Liver MRI for Chronic Liver Disease Diagnosis

Jihun Oh1, Diego Martin2, Xiaoping P. Hu3

1School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, United States; 2Department of Medical Imaging, University of Arizona, Tucson, AZ, United States; 3Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, GA, United States

This paper describes our work of using features derived from contrast-enhanced liver MR images for providing a quantitative assessment of chronic liver disease severity. We first examined the mean slope of contrast uptake in hepatobiliary phase and demonstrated that it is significantly correlated with fibrosis score. We also examined several texture measures in equilibrium phase using Gabor filtering and grey level co-occurrence matrix and built a supervised maximum a posteriori classifier using these features to predict the disease severity. The classifier was evaluated by cross-validation and shown to be highly robust in predicting fibrosis score.

Abstract PDF Poster PDF

Keywords

able accuracy accurate acquisition agent aided aiding allows among aorta assessing assigned biomedical blood causing chronic cirrhosis class classification classifier clinical clinicians comprehensively computation computed computer consequently contrast correlated criterion cross dashed decision degree degrees derived destruction detection diagnosis diagnostic discrete disease dissimilarity distribution document early edge electrical engineering enhanced enhancement entire entropy equilibrium error establish examines feature features fibrosis field filter fisher focuses function hepatic highly independently indicates influence inhibition injection institute intensity involves known lacking larger leave linear liver longitudinal making matching matrix measures medical minutes multivariate needed neighboring normalized occurrence open orientation patient patients patterns peak performance pixels posteriori predict predicted prediction process processing progress progressive property quantitative radius ranked reduces reflected regeneration related relied representation rest reticular school score scores separability severity shorten significantly slope source spatial staging structure subtracted suggesting supervised taken technology texture textures tissue transactions transplantation trapezoidal treatment triangle tumors utility validation variance various