Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 190 - Session on Semi-Supervised and Unsupervised Learning
Type: Contributed
Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
Sponsor: Biometrics Section
Abstract #312505
Title: Modeling Trajectory of Thalamic Atrophy from Brain MRI Scans of Multiple Sclerosis Patients Using Mixed Spline
Author(s): Steven Cen* and Mulugeta Gebregziabher and Daniel Pelletier and Christina Azevedo and Amirhossein Jaberzadeh
Companies: University of Southern California and Medical University of South Carolina and University of Southern California and University of Southern California and University of Southern California
Keywords: spline; mixed model; MRI; brain; aging; missing data

Disease duration is an important factor for multiple sclerosis (MS) patients to understand the natural history of MS, make appropriate treatment decisions, and aid in prognosis. However, true disease duration can be difficult to estimate because the clinical disease onset, defined as the age of the first clinical symptom, is probably years after the biological onset and the onset of tissue injury visible on structural magnetic resonance imaging (MRI). Traditionally, linear mixed modeling has been used to fit the trajectory of brain atrophy, but it is not an effective method for representing the complexity of aging data. We propose a mixed spline modeling approach to more accurately fit the trajectory of brain atrophy that could be used to estimate the age of onset of disease-related brain (thalamic) atrophy. The onset of brain tissue loss can be estimated as the age when the spline curve trajectory of an MS patient starts to depart from that of a normal aging spline curve trajectory. However, it is difficult to find sufficient longitudinal data over the entire lifespan to fit a mixed spline model. Thus, we have introduced a new concept of “fish bone” data structure and novel statistical approaches to construct pseudo-longitudinal data from a large cross-sectional normal aging data using imputation. In a simulation study, we have identified unrestricted B-Spline with TOEPLIZ as G-side matrix as the best mixed spline model. When applied to data from 470 MS cases and 1272 controls, a strong correlation was found between spline estimated vs. observed longitudinal values in independent validation data. Individual trajectory plots from the B-Spline mixed model showed a consistent pattern of similar trajectory curves for MS and normal aging at early ages, with gradual departure from each other over time.

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

Back to the full JSM 2020 program