Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 384 - Recent Statistical Advances in Diagnostic Medicine
Type: Invited
Date/Time: Wednesday, August 5, 2020 : 1:00 PM to 2:50 PM
Sponsor: Section on Medical Devices and Diagnostics
Abstract #309198
Title: Nonparametric Optimal Rules for Combining Multiple Quantitative Imaging Biomarkers
Author(s): Lu Mao*
Companies: University of Wisconsin-Madison
Keywords: Classification; EM algorithm; Isototonic regression; Machine learning; Magnetic resonance

Widely used in diagnostic medicine, Quantitative imaging biomarkers (QIB) are measurements derived from medical images using well-defined computer algorithms. For example, hepatic stiffness and relative fat fraction are quantified from magnetic resonance (MR) electrography of the liver for detection of nonalcoholic steato-hepatitis. Different QIBs may carry information on different aspects of the disease status. As a result, they may yield a more accurate diagnostic test when properly combined than used individually. To find the optimal rule of combination, we develop a nonparametric model under the sole assumption that the risk score is monotone with respect to each biomarker in question. The nonparametric, monotonicity-constrained risk score is estimated by an EM-type algorithm and is then used as an ordinal diagnostic test. Simulation studies show that our approach outperforms standard parametric models or machine learning techniques for classification based on the multiple QIBs. As an illustration, we apply the proposed methodology to a real study on nonalcoholic fatty liver disease patients undergoing MR electrography.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program