Skip to main navigation Skip to search Skip to main content

When Experts Disagree: Characterizing imperfect segmentations from multiple annotators when ground truth is unavailable

  • Masha Geshvadi
  • , Gloria So
  • , David D. Chlorogiannis
  • , Colin Galvin
  • , Erickson Torio
  • , Amirali Azimi
  • , Yaw Tachie-Baffour
  • , Nazim Haouchine
  • , Alexandra Golby
  • , Mark Vangel
  • , William M. Wells
  • , Yan Epelboym
  • , Rose Du
  • , Funda Durupinar
  • , Sarah Frisken
  • University of Massachusetts Boston
  • Brigham and Women’s Hospital
  • University of Waterloo
  • Harvard University

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Manual vessel segmentations exhibit substantial inter- and intra-annotator variability, introducing uncertainty when such annotations are used as pseudo ground truth. In this work, we systematically investigate this variability in 2D Digital Subtraction Angiography images and characterize the uncertainty across multiple annotators. A dataset of 66 clinical images and approximately 2,000 derived image patches was analyzed using expert segmentations and additional ratings.

Original languageEnglish
Title of host publicationMedical Imaging 2026
Subtitle of host publicationImage Processing
EditorsJhimli Mitra, Yu Gan
PublisherSPIE
ISBN (Electronic)9781510697874
DOIs
StatePublished - Apr 3 2026
EventMedical Imaging 2026: Image Processing - Vancouver, Canada
Duration: Feb 15 2026Feb 19 2026

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
Volume13925
ISSN (Print)1605-7422
ISSN (Electronic)2410-9045

Conference

ConferenceMedical Imaging 2026: Image Processing
Country/TerritoryCanada
CityVancouver
Period2/15/262/19/26

ASJC Scopus Subject Areas

  • Electronic, Optical and Magnetic Materials
  • Atomic and Molecular Physics, and Optics
  • Biomaterials
  • Radiology Nuclear Medicine and imaging

Keywords

  • Digital Subtraction Angiography
  • Uncertainty Quantification
  • Vessel Segmentation

Cite this