Visual Analytics Support for Chemotherapy Treatment Response Prediction in Breast Cancer Patients | TU Wien

Information

Publication Type: Master Thesis
Workgroup(s)/Project(s): not specified
Date: 2026
TU Wien Library: AC17900963
Open Access: yes
First Supervisor: Renata Georgia Raidou
Pages: 196
Keywords: Brustkrebs, Neoadjuvante Chemotherapie, Vorhersage des Therapieansprechens, Radiomics, Diffusion Weighted Imaging, Tumorwachstumsmodellierung, Simeoni-Modell, Maschinelles Lernen, Visual Analytics, Feature-Redundanz

Abstract

Predicting treatment response in breast cancer patients undergoing neoadjuvant chemotherapy remains challenging, due to the substantial variation in how tumors respond to treatment. This thesis investigates whether mathematical tumor growth modeling, specifically the Simeoni model, can enhance a typical radiomics-based treatment response classification analysis. We analyzed Diffusion Weighted Imaging (DWI) data from 162 patients in the ACRIN6698 trial across four imaging timepoints during their treatment, extracted 428 radiomic features and 15 Simeoni-derived mechanistic features, and trained 116 Random Forest model configurations across different feature sets and timepoint combinations using Leave-One-Out Cross-Validation for five-class treatment response classification. An interactive visual analytics dashboard was finally developed for the systematic exploration of the experimental space. The best model achieved 90.1% accuracy using radiomics-only features from all four imaging timepoints. Combined models with Simeoni features consistently underperformed, with accuracy reductions ranging from 2.4% to 6.2%. Multi-timepoint imaging was the strongest predictor of accuracy, increasing performance from 71.6% with baseline data to 90.1% with all timepoints. Statistical analysis confirmed that Simeoni features are redundant with multi-timepoint radiomic features rather than being complementary. For clinical application, radiomics-only models are recommended for treatment response classification. The Simeoni model provides biological insight into tumor dynamics but does not enhance classification performance.

Additional Files and Images

thesis

Weblinks

BibTeX

@mastersthesis{stadter-2026-vas,
  title =      "Visual Analytics Support for Chemotherapy Treatment Response
               Prediction in Breast Cancer Patients",
  author =     "Matthias Stadter",
  year =       "2026",
  abstract =   "Predicting treatment response in breast cancer patients
               undergoing neoadjuvant chemotherapy remains challenging, due
               to the substantial variation in how tumors respond to
               treatment. This thesis investigates whether mathematical
               tumor growth modeling, specifically the Simeoni model, can
               enhance a typical radiomics-based treatment response
               classification analysis. We analyzed Diffusion Weighted
               Imaging (DWI) data from 162 patients in the ACRIN6698 trial
               across four imaging timepoints during their treatment,
               extracted 428 radiomic features and 15 Simeoni-derived
               mechanistic features, and trained 116 Random Forest model
               configurations across different feature sets and timepoint
               combinations using Leave-One-Out Cross-Validation for
               five-class treatment response classification. An interactive
               visual analytics dashboard was finally developed for the
               systematic exploration of the experimental space. The best
               model achieved 90.1% accuracy using radiomics-only features
               from all four imaging timepoints. Combined models with
               Simeoni features consistently underperformed, with accuracy
               reductions ranging from 2.4% to 6.2%. Multi-timepoint
               imaging was the strongest predictor of accuracy, increasing
               performance from 71.6% with baseline data to 90.1% with all
               timepoints. Statistical analysis confirmed that Simeoni
               features are redundant with multi-timepoint radiomic
               features rather than being complementary. For clinical
               application, radiomics-only models are recommended for
               treatment response classification. The Simeoni model
               provides biological insight into tumor dynamics but does not
               enhance classification performance.",
  pages =      "196",
  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 =   "Brustkrebs, Neoadjuvante Chemotherapie, Vorhersage des
               Therapieansprechens, Radiomics, Diffusion Weighted Imaging,
               Tumorwachstumsmodellierung, Simeoni-Modell, Maschinelles
               Lernen, Visual Analytics, Feature-Redundanz",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2026/stadter-2026-vas/",
}