Online Program Home
  My Program

Abstract Details

Activity Number: 549 - Recent Development of Statistical Learning Methods for Complex Biomedical Data
Type: Invited
Date/Time: Wednesday, August 2, 2017 : 2:00 PM to 3:50 PM
Sponsor: ENAR
Abstract #325044
Title: Optimal Individualized Treatment Strategy with Imaging Covariates
Author(s): Rui Song*
Companies: NC State University
Keywords:
Abstract:

Different from the standard treatment discovery framework which is used for finding single treatments for a homogenous group of patients, personalized medicine involves finding therapies that are tailored to each individual. In this paper, we consider two approaches to estimate the optimal treatment strategy that uses both scalar and imaging covariates. The first approach is model based and takes into account the smooth nature of most imaging data. We develop an efficient penalized total variation optimization to estimate the unknown slope function and other related parameters. We also establish the error bounds for the total variation slope estimator of imaging covariates and the coefficients of scalar covariates. The second approach is built upon convolutional neural network (CNN) which exploits the correlation between adjacent pixels in the two or three dimensional imaging space. We take this opportunity to employ deep learning to approximate the contrast function and assign future patients according to the sign of the estimated contrast function. Extensive simulations demonstrate that the two proposed methods have superior performance against other possible approaches.


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

Back to the full JSM 2017 program

 
 
Copyright © American Statistical Association