B0738
Title: Subject-level weights for detecting brain volume differences
Authors: Christina Chen - University of Pennsylvania (United States) [presenting]
Matthew Tisdall - University of Pennsylvania (United States)
David Wolk - University of Pennsylvania (United States)
Sandhitsu Das - University of Pennsylvania (United States)
Paul Yushkevich - University of Pennsylvania (United States)
Russell Shinohara - University of Pennsylvania (United States)
Abstract: Multi-atlas image segmentation is a widely used approach in neuroimaging analyses that involves estimating the volume of a region of interest (ROI). However, current practices treat these images equally without incorporating any information about variations in segmentation precision among the study images or differences in the segmentation precision of different ROIs for each subject. A novel method is proposed that estimates the variances of ROI volume estimates for each subject due to multi-atlas segmentation and thus provides a way of reweighting these estimates to increase efficiency in downstream inference.