Tuesday, February 25, 2020
3:30 pm - 4:30 pm
Brian L. Strom Conference Room, Room 701 Blockley Hall
Abstract: Computed tomography (CT) is a medical imaging procedure that uses x-ray beams to produce detailed three-dimensional images of various anatomical areas, including the lungs. Patients with lung disease have inflammatory granulomas and/or air-sacs in regions of their lungs which are visible on CT. In this talk, I will present statistical methodologies and software that I developed to objectively identify population-level spatial patterns in lung CT data. I will start by discussing my fully-automated pre-processing pipeline for lung CT scans that resulted in the first publicly available standard lung template. This template provides a standardized coordinate system for lung imaging studies, making it possible to perform spatial comparisons across populations. Next, I will present my eigenvector spatial filtering (ESF) model, traditionally used in the geographical literature, and now adapted to imaging data, to perform whole-lung population-level inference. Compared to the traditional approach in the imaging field which fails to accurately account for the spatial autocorrelation between voxels, my ESF model results in lower false positive rates and higher statistical power under simulation.