|
Activity Number:
|
562
- Regression Methods for Neuroimaging Data
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics in Imaging
|
|
Abstract #323508
|
|
|
Title:
|
Multivariate Residualization in Medical Imaging Analyses
|
|
Author(s):
|
Kevin Donovan* and Kristin Linn and Russell Shinohara
|
|
Companies:
|
University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
|
|
Keywords:
|
Imaging;
Confounding;
Multivariate;
Residualization
|
|
Abstract:
|
Confounding in medical imaging research is common, complicating association and prediction studies based on image data. Addressing this problem is compounded by the high dimensional and correlated nature of image data, making computationally efficient and robust methods to address confounding difficult to implement. Due to this multivariate nature, commonly used methods include functional regression and by-region univariate residualization. The utility of these methods is limited for data with many image regions due to computational and model complexity, and neglecting multivariate properties which may fail to remove confounding related to the joint distribution of these regions. We develop a multivariate residualization method using support vector regression to estimate the relationship between the image and the confounder and then computing the orthogonal projection of each subject’s image data onto this space. We illustrate this method’s performance in a set of simulation studies and apply it to data from the ADNI dataset with age as a confounder of interest.
|
Authors who are presenting talks have a * after their name.
