Imaging biomarkers in thyroid eye disease and their clinical associations

Shikha Chaganti, Katrina Nelson, Kevin Mundy, Robert Harrigan, Robert Galloway, Louise A. Mawn, Bennett Landman

Research output: Contribution to journalArticlepeer-review


The purpose of this study is to understand the phenotypes of thyroid eye disease (TED) through data derived from a multiatlas segmentation of computed tomography (CT) imaging. Images of 170 orbits of 85 retrospectively selected TED patients were analyzed with the developed automated segmentation tool. Twenty-five bilateral orbital structural metrics were used to perform principal component analysis (PCA). PCA of the 25 structural metrics identified the two most dominant structural phenotypes or characteristics, the "big volume phenotype" and the "stretched optic nerve phenotype," that accounted for 60% of the variance. Most of the subjects in the study have either of these characteristics or a combination of both. A Kendall rank correlation between the principal components (phenotypes) and clinical data showed that the big volume phenotype was very strongly correlated (p-value <0.05) with motility defects, and loss of visual acuity. Whereas, the stretched optic nerve phenotype was strongly correlated (p-value <0.05) with an increased Hertel measurement, relatively better visual acuity, and smoking. Two clinical subtypes of TED, type 1 with enlarged muscles and type 2 with proptosis, are recognizable in CT imaging. Our automated algorithm identifies the phenotypes and finds associations with clinical markers.

Original languageEnglish (US)
Article number044001
JournalJournal of Medical Imaging
Issue number4
StatePublished - Oct 1 2018


  • computed tomography
  • label fusion
  • multiatlas
  • orbit
  • principal component analysis
  • segmentation
  • thyroid eye disease

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging


Dive into the research topics of 'Imaging biomarkers in thyroid eye disease and their clinical associations'. Together they form a unique fingerprint.

Cite this