Investigating Immune Checkpoint Inhibitor–Induced Atherosclerosis Using Automated CT–PET Analysis | TU Wien

Information

Publication Type: Master Thesis
Workgroup(s)/Project(s): not specified
Date: 2026
TU Wien Library: AC18107974
Second Supervisor: Eduard Gröller
Open Access: yes
First Supervisor: Renata Georgia Raidou
Pages: 90
Keywords: PET-CT Analysis, segmentation, machine learning

Abstract

A clinical hypothesis claims that cancer patients treated with immune checkpoint inhibitor (ICI) therapy are at increased risk of developing atherosclerosis. To test this, clinicians use PET/CT imaging together with manual sampling procedures that are labor-intensive, operator-dependent, and hard to reproduce at scale. In this thesis, we develop a fully automated pipeline that replaces manual region-of-interest selection with objective, full-volume extraction of arterial FDG uptake, i.e., we combine deep-learning-based anatomical segmentation with a set of morphological filters designed together with medical experts. We apply it to a longitudinal cohort of 64 ICI-treated patients, each with paired pre- and post-therapy scans, and we test the hypothesis across five complementary filtering methods. In contrast to previous manual studies, including previous work on the same dataset, we find no statistically significant increase in aortic uptake after therapy. However, we find a statistically significant increase in the thyroid gland, a process known and reported in the literature as ICI-induced thyroiditis, indicating that the pipeline still detects real treatment effects, so the null is not just an artifact of a weak method. More generally, our results show the value of reproducible full-volume quantification in cardio-oncology imaging.

Additional Files and Images

thesis

Weblinks

BibTeX

@mastersthesis{eseroglou-2026-iic,
  title =      "Investigating Immune Checkpoint Inhibitor–Induced
               Atherosclerosis Using Automated CT–PET Analysis",
  author =     "Christoforos Eseroglou",
  year =       "2026",
  abstract =   "A clinical hypothesis claims that cancer patients treated
               with immune checkpoint inhibitor (ICI) therapy are at
               increased risk of developing atherosclerosis. To test this,
               clinicians use PET/CT imaging together with manual sampling
               procedures that are labor-intensive, operator-dependent, and
               hard to reproduce at scale. In this thesis, we develop a
               fully automated pipeline that replaces manual
               region-of-interest selection with objective, full-volume
               extraction of arterial FDG uptake, i.e., we combine
               deep-learning-based anatomical segmentation with a set of
               morphological filters designed together with medical
               experts. We apply it to a longitudinal cohort of 64
               ICI-treated patients, each with paired pre- and post-therapy
               scans, and we test the hypothesis across five complementary
               filtering methods. In contrast to previous manual studies,
               including previous work on the same dataset, we find no
               statistically significant increase in aortic uptake after
               therapy. However, we find a statistically significant
               increase in the thyroid gland, a process known and reported
               in the literature as ICI-induced thyroiditis, indicating
               that the pipeline still detects real treatment effects, so
               the null is not just an artifact of a weak method. More
               generally, our results show the value of reproducible
               full-volume quantification in cardio-oncology imaging.",
  pages =      "90",
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Research Unit of Computer Graphics, Institute of Visual
               Computing and Human-Centered Technology, Faculty of
               Informatics, TU Wien",
  keywords =   "PET-CT Analysis, segmentation, machine learning",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2026/eseroglou-2026-iic/",
}