University of Twente Student Theses
As of Friday, 8 August 2025, the current Student Theses repository is no longer available for thesis uploads. A new Student Theses repository will be available starting Friday, 15 August 2025.
Uncertainty Assessment for Improving Quality Assurance of Prostate Autosegmentation in MR-Linac Radiotherapy
Eek, D.M.C. (2025) Uncertainty Assessment for Improving Quality Assurance of Prostate Autosegmentation in MR-Linac Radiotherapy.
Full text not available from this repository.
| Full Text Status: | Access to this publication is restricted |
| Abstract: | The integration of MR-linac systems into radiation therapy has marked a significant advancement, particularly for prostate cancer treatment, due to their superior soft tissue contrast. This enables the use of online adaptive radiotherapy, enabling online adaptation of treatment plans. However, a bottleneck in the clinical workflow is the time-intensive contouring process required for each radiotherapy fraction. In recent years, the rapid development of artificial intelligence techniques holds great promise for automated segmentation in radiation oncology. However, it remains challenging to distinguish between accurate and poor-quality segmentations for individual patients. To address this, Monte Carlo dropout was introduced as a method for estimating model uncertainty, providing insight into a model’s confidence in its predictions. This thesis investigated to what extent uncertainty can support quality assurance for prostate autosegmentation in the MR-Linac workflow. We trained a DeepLabV3+ model for MR-based prostate segmentation and applied Monte Carlo dropout to quantify prediction uncertainty. Our results demonstrate that uncertainty, when properly calibrated, serves as a valuable tool for assessing model confidence, offering insights into the reliability of predictions and their geometric accuracy. Consequently, uncertainty can be a helpful tool in the approval process of autosegmentations. Furthermore, we show that targeting and correcting voxels with the highest uncertainty leads to the most significant improvements in both geometric and dosimetric outcomes. Therefore, uncertainty-guided correction offers a promising tool for editing autosegmentations. These findings emphasize the potential of uncertainty-based quality assurance for prostate autosegmentation in the MR-Linac workflow. |
| Item Type: | Essay (Master) |
| Faculty: | TNW: Science and Technology |
| Subject: | 44 medicine |
| Programme: | Technical Medicine MSc (60033) |
| Link to this item: | https://purl.utwente.nl/essays/106424 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page