Dose-Aware and Robust Contour QA for Radiotherapy

This site presents the PhD thesis of Amith Kamath, carried out at the ARTORG Center for Biomedical Engineering Research and the Medical Image Analysis Lab at the University of Bern.

Overview

Radiotherapy planning relies heavily on accurate anatomical contouring, and the quality assurance (QA) of those contours is central to the effectiveness of treatment. Yet the geometric metrics traditionally used to check contours — overlap scores such as the Dice coefficient — correlate poorly with the delivered dose distribution, and therefore with patient outcomes. This thesis works towards dose-aware contour QA: evaluating a contour by its clinical, dosimetric consequences rather than its pixel-wise overlap.

It tests the hypothesis that fast, dose-aware AI can make contour QA more consistent and clinically meaningful, provided that the dose-prediction and segmentation components are sensitive to consequential contour variations and robust to distribution shifts. The chapters build that case in three connected steps: first establishing the clinical need and testing automated assessment, then evaluating dose-prediction sensitivity and robustness, and finally examining the robustness of the segmentation architectures these tools depend on. Two proof-of-concept tools are embedded where they arise in that progression: AutoDoseRank in Part One ranks candidate segmentations by dosimetric quality, while ASTRA in Part Two produces local dose-sensitivity maps for contour review.

What’s inside

The work is organized by theme, in three parts:

• Part One — Dosimetric Awareness of Radiation Oncology Professionals (Chapters 3–5) shows, through a survey and a head-to-head comparison, that experts struggle to judge the dosimetric impact of contour changes and that a deep-learning dose predictor can outperform them, culminating in AutoDoseRank.

• Part Two — Sensitivity of Dose Prediction Models (Chapters 6–8) examines whether the underlying dose predictors are accurate, sensitive to realistic contour variation, robust to unusual cases, and fast enough for review, then turns that sensitivity into ASTRA.

• Part Three — Robustness of Segmentation Models (Chapters 9–11) studies how U-Net design and data choices — including skip connections, task complexity, texture, context, and foreground ratio — affect robustness under distribution shift.

From the thesis to the ongoing research program

The thesis is the completed foundation; my current work continues its questions along three connected axes:

1. Robust segmentation models extend Part Three from controlled architecture studies toward models that remain dependable across scanners, protocols, institutions, anatomies, and image quality.

2. Personalized contour review and correction extend Parts One and Two from expert disagreement, automated assessment, and sensitivity mapping toward models that represent multiple plausible contours and workflows that help clinicians resolve consequential differences efficiently.

3. Fast and sensitive dose prediction extends Part Two toward rapid treatment planning and evaluation across tumour sites, institutions, planning systems, and contour alternatives.

The axes are coupled rather than sequential:

• robustness seeks invariance to irrelevant acquisition and deployment variation;

• personalized correction preserves meaningful patient, observer, and treatment-context variation while helping clinicians act on the alternatives and uncertainty that remain; and

• dose prediction provides a fast treatment-aware signal for comparing those alternatives and guiding review.

The evolving theme pages above describe ongoing work, while this site remains the detailed record of the thesis studies. The publications page is the compact map from each axis to its supporting papers and project pages.

New readers may want to start with the Introduction for the clinical motivation, or the Background for a primer on the radiotherapy workflow and the AI methods used throughout. The Conclusions revisit the central hypothesis across all three parts, and every claim is backed by a linked reference.

Contents

• 1. Introduction
  • Clinical Background
  • Workflow Challenges
  • Unmet Clinical Needs
  • Hypothesis, Contributions, and Structure
• 2. Background
  • The Radiotherapy Workflow
  • An Artificial Intelligence (AI) Primer
  • Related Work
  • Evaluation Metrics
• Dosimetric Awareness of Radiation Oncology Professionals
  • 3: Predicting the Impact of Target Volume Contouring Variations on the Organ at Risk Dose: Results of a Qualitative Survey
  • 4: Comparing the Performance of Radiation Oncologists versus a Deep Learning Dose Predictor to Estimate Dosimetric Impact of Segmentation Variations for Radiotherapy
  • 5: AutoDoseRank: Automated Dosimetry-Informed Segmentation Ranking for Radiotherapy
• Sensitivity of Dose Prediction Models
  • 6: Deep-learning-based dose predictor for glioblastoma–assessing the sensitivity and robustness for dose awareness in contouring
  • 7: How sensitive are Deep Learning based Radiotherapy Dose Prediction Models to Variability in Organs at Risk Segmentation?
  • 8: ASTRA: Atomic Surface Transformations for Radiotherapy Quality Assurance
• Robustness of Segmentation Models
  • 9: The impact of U-Net architecture choices and skip connections on the robustness of segmentation across texture variations
  • 10: Do We Really Need that Skip-Connection? Understanding Its Interplay with Task Complexity
  • 11: How do 3D image segmentation networks behave across the context versus foreground ratio trade-off?
• 12. Conclusions
  • Part One: Dosimetric Awareness of Radiation Oncology Professionals
  • Part Two: Sensitivity of Dose Prediction Models
  • Part Three: Robustness of Segmentation Models
  • Synthesis
• References