Overcoming the barriers to implementing auto-contouring in external beam radiotherapy for cervical cancer
Loading...
Embargo End Date
2025-10-14
ICR Authors
Authors
Mackay, K
Document Type
Thesis or Dissertation
Date
2025-04-14
Date Accepted
Abstract
This thesis has developed methods for the clinical implementation of auto-contouring for external beam radiotherapy (EBRT) for cervical cancer. It has explored the barriers to auto-contouring, which include a lack of consensus on evaluation methods and challenges in generating high quality training data. These challenges arise due to issues such as multiple contouring protocols, reliance on clinical decision-making and inter-observer variation.
The important differences between published contouring protocols for cervical cancer EBRT have been identified. A contouring protocol based on anatomical substructures has been created, to remove the heterogeneity caused by clinical decisions. This protocol has been used to produce training data for a novel nnU-Net auto-contouring system that can contour the substructures for target volumes and organs-at-risk for cervical cancer.
This thesis has formally reviewed the auto-contouring assessment metrics in published literature. Significant variation was found, making it challenging to compare studies and recommend standardised practice. This thesis also identified the need for clinically relevant assessment methods, applicable in the context of usual inter-observer variation.
A manual inter-observer contouring study has been performed to establish a validation dataset and estimate delineation uncertainty for both manual and auto-contours. The delineation uncertainty was lower for auto-contours from a developed auto-contouring system, than for manual contours, suggesting superior performance. Delineation uncertainty may also be a novel clinically useful assessment method, as it can directly determine necessary treatment margins.
The novel auto-contouring system was found to produce contours that fall within the usual range of inter-observer variation. This range can also mostly be predicted by a pair of expert observers, which could simplify assessment processes in the future.
Finally, a novel standardised structured qualitative assessment has been developed and evaluated. This showed that inter- and intra-rater reliability is variable, an issue that must be considered when evaluating auto-contouring systems.
Citation
2025
DOI
Source Title
Publisher
Institute of Cancer Research (University Of London)